The Nature of Consciousness

Here you will find some notes I am assembling while trying to develop a theory of how consciousness emerges from the 1.4 kg semi-solid mass of wet tissue that is our brain.¹ My bias, as you will see, is that consciousness should be able to be implemented in artificial constructs such as electronic control systems, monitoring devices, in research facilities and later on in more sci-fi like applications. (Please keep in mind that reading this will be like watching sausage being made - not a pretty site!)

In order to be seen as complete a new theory of something should:

• Provide a context
• Define the something
• Set criteria by which any theory of the something would be seen as adequate
• Review the current thinking
• Describe the new theory
• Evaluate current and new against the criteria
• Describe consequences and/or make predictions
• Demonstrate a practical implementation or application

This document is being restructured to conform to the above structure. It's likely to be even more of a mess the usual for a short while.

Who Am I, to take on such a task?

I am a practising Solution Architect at a large bank, which means I have formal computer training and experience. (It also means I will occasionally use computer technology terminology.) Almost 25 years ago I earned a M.Sc. in Physiology, which means I have a basic understanding of the scientific fundamentals. (…and will use technical scientific terminology.) And I've have had a long standing interest in this subject area, since AI by its nature is a child of both Computer Science and Physiology.

Like the Victorian armchair gentleman scientists and modern day amateur astronomers, I believe that there are still opportunities for motivated amateurs to contribute significant understanding in this field. Lab work is done by grad students and research technicians. The results are published by professional scientists and their findings may then be deeply considered by philosophers. The synthesis of facts and knowledge into fundamental insights and understanding however should be capable of being done by anyone with sufficient interest and background!

Why a Wiki?

I wanted a site that was accessible by me from multiple locations. The backup provision of this Wiki host was a natural fit. I will open it up for viewing, and perhaps later for comments and suggestions, once I get enough content. These are my notes however and not a collaborative effort, so I will probably never enable the real 'wiki' functionality of this wiki.

Why Study the Nature of the Mind and Consciousness?

Simple, for a variety of reasons. There is an intrinsic academic attraction to this field. Secondly, a good design for intelligent and aware machines would be incredibly useful to society. Some futurists predict that we might see such devices prior to 2045² and others predict even earlier than 2029³. Third, if and when humanity reaches other stellar systems (or vice versa!) it will be handy to have a working definition of who to greet and what to eat. Fourth, physicians must make decisions regarding the withdrawal of life support and eligibility for donating organs, which are partially based on an assessment of the patient's potential for returning to consciousness. It is legally and morally important to get this right. Fifth , our legal system grants rights and responsibilities to individuals that can reasonably be held accountable for their own actions. Should these rights be extended to other higher species such as chimps, apes, whales or dolphins? Should apparently conscious devices one day be granted these rights? Finally, humans are not likely to routinely go far out of the solar system. That job will be reserved for autonomous mechanisms. Independent, adaptable machines that are able to make appropriate decisions in novel circumstances would probably be implemented with a conscious, curious, caring and capable entity in charge of each. A generally accepted definition of what conscious is would greatly help in all of these areas.

Note that this will never be a formal thesis or article in a peer reviewed scientific journal. Nor will it be a public demonstration of the application of the scientific method to this problem - neither the idealized version, nor the practical way that is usually applied. It is purely a personal exploration of the subject. I have reviewed major, highly regarded secondary sources as noted in the bibliography at the bottom, as well as some published primary articles I found appropriate. I have freely incorporated some important ideas and interpretations and freely left out others that did not mesh with my thinking. What I have tried very hard to not do is ignore reasonable and successful arguments against the components that I have chosen to make up my point of view, that is I did not knowingly incorporate discredited concepts.

Note that this is not a rigorous theory of human consciousness, especially as it is contributed to by each of the specific brain structures. Instead, while considering the actual implementation in humans, I tried to remove the arbitrary specifics. I regard the fact that there is apparently a working memory as important, but less so that facts that the amygdala is mostly involved in managing the storage of declarative (facts) and emotional memories, and the hippocampus is more involved in facilitating the storage of episodic and autobiographical memory, and part of it is distributed in parts of the frontal cortex. I see that as more a matter of a specific implementation and less a matter of conceptual understanding how any kind of consciousness can work at all, and I reserve the right to either modify or simplify it for this use, or cut away at the irrelevant, specifically ‘human’ considerations which occupy much of the primary and secondary literature – considerations such as the explicit systems involved in human vision (including aspects like blind sight), human hunger, human growth and development etc.

What follows is my best guesses and speculations on what constitutes the essential set of components and the necessary architecture that can explain how we experience conscious, and that might define an artificial system that itself is a conscious entity.

Consciousness Defined (or at least attempted)

At the beginning of any field of human endeavour we start with nothing. Thinking about for example why there is day and night, why we die, why the sky is blue, where did we come from and so on starts as idle speculation which may escalate to myth and religion. In the vacuum of having no tradition defining how to approach answering these questions, philosophy rises in importance. The discipline brought to the subject by this structured way of thinking helps to focus attention on the right questions and identify ways to get at better answers. Once the real gaps in understanding are recognised we can start to accumulate preliminary information and build comprehensive theories to explain the results. It is at this point that a new formal branch of science is born and philosophy’s influence fades. We are not quite at that point in the study of consciousness. We don't have a widely accepted framework that defines how to even approach the subject. My background is in hard science and my peers and I used to look down our noses at the so called discipline of philosophy. Recently I revised my attitude towards the subject and now know better. Philosophy is the step after dancing witch doctors, but is a necessary step never the less. At this stage in studying the nature of consciousness it can contribute the most value. Snippits of scientific observations in one or another aspect of human thought can then be more appropriately placed in an initial framework that may someday become a testable theory.

In order to be able to say that something is or is not conscious, we need to rigorously define what that means. Whatever else it is, consciousness is the product of ongoing processes that can occur in brains of certain complexity and architectures.

Consciousness is a product of electro-chemical physical process in the cerebrum. Damage the brain and you damage consciousness. Change consciousness and you have changed brain processes. Stop the brain processes and you have stopped consciousness. I am aware of no credible conscious event that did not also involve a brain that produced it. The two are inexorably linked and a brain is required to be able to have a conscious experience. This is not to say however that biological brains and biological processes are absolutely required, it's just that until now biology has always been the substrate. This is also not to say that all brains and their processes can produce consciousness. I believe it is only brains that cross a certain threshold of complexity, that have certain architecture which will be described later (as soon as I finish it!), and that possess particular capabilities are capable of producing consciousness. The neural node (~brain) of a species of microscopic nematode worm⁴ consists of exactly 302 nerve cells (neurons). I assume it is generally agreed that this can only be an unconscious collection of reflexes. The node can perceive (= notice) changes in its environment and usually react appropriately to them but it almost certainly does not experience these changes. Detecting change is not equivalent to experiencing. Reflex responses are not the same as, or a kind of thinking, planning, deciding or otherwise conscious acts. Although the worm can be seen to perceive and react to noxious stimuli, reacting to pain by writhing around in an effort to avoid it for example, the worm brain is clearly not capable of conscious thought about or conscious consideration about, or fear of that pain, not even dimly. Every similar input always produces a similar pre-programmed stereotyped output, automatically selected from among the small collection of responses always available to it. This worm brain is not complex enough, is not arranged in the right way, and it can't do the prerequisite things that are needed for it to be conscious.

The psychologist William James described it succinctly as attention plus short-term memory. It's what you possess right now as you read this, and what you lack when you are asleep and between dreams, or under anaesthesia.⁵

Someone, I am not sure who, describes the primary function of consciousness as being the mechanism that musters appropriate resources required to solve ongoing complex problems.

Someone else claims concepts need to be in short term (actually, this should be 'working') memory in order to be available to the processes of consciousness, and that conscious attention to something is a prerequisite to it residing in long term memory - conscious experience is the gateway to long term memory. I am strongly inclined to agree with that claim.

Consciousness is the subjective feeling of personal awareness experienced by each of us. It is the thing that happens between all the sensory inputs coming in and any long term memories that might result afterwards. Consciousness is the impression that something is experiencing things and thinking thoughts, that the thoughts are those of this something. Things continuously stream into awareness, mostly fully formed, one after another, generating a point against which past and present are measured by. We are never conscious of arrays of individual light sensing nerves being activated (for example). We instead see a blue sky. This means we think in terms of symbols and concepts. There is also a perception of at least partial control⁶ over the content and direction of this stream of personal awareness. There is also some ability to return to or reflect on previously experienced points along the stream of consciousness. Something that is this subjective and first person isn't lilkely to be able to be rigorously tested. One can ask - in what way can we absolutely or even reliably determine if anything is conscious at all?

Consciousness has several characteristics:

• ownership - a first person viewpoint is always involved
• aboutness - by which I mean one is always conscious of something, and one is never conscious but thinking of nothing. I suppose one could be conscious of the absence of something, but that absence of course is something. This is the same as formally asserting that consciousness has the property of intentionality
• presentness - which means that it is always happening in the present. One is always presently conscious, even when thinking about some remote event
• locationless - it feels to me that there is no site of consciousness, no place where it happens. Try closing your eyes in a quiet room and think of something far away… the object of your thought might be removed, but I can't assign the thought itself to a specific location, inside my head or out. There are some cultures that ascribe consciousness to be behind the eyes, above the nose, in the heart, behind the head etc. but these seem to be culturally influenced locations
• agency - I have a very strong impression that it is the conscious "I" who, for the most part, is directing the flow of things that I am actively thinking about. After all, my consciousness is helping me to achieve my immedate and long term goals, not those of my neighbours. I identify this first person awareness with the core of who "I" am. I can even choose to make myself more aware of something, like paying attention to my aches and pains to see if I am still comfortable in this position, periodically listening for an expected visitor, periodically looking to see if the sun is setting.

Consciousness can be thought about through its contents, or as a state. I am less interested in things concerning the content of what one is conscious of, and more so in how one can be even conscious of anything at all.

A close approximation of the first person type of definition of consciousness is something that has the ability to participate in an active, first person manner in some mental process or physical behaviour. Obtaining independent and unambiguous confirmation of this perception is the key challenge.

The only definition that removes this subjectivity is an outside-in looking one. For purposes of this document, and as a proposal for a formal and testable criteria of higher awareness consciousness is characterized as a process that:

• is distinct from unconscious processes. There is no way an entity can be conscious without also having an unconscious state (either running in parallel or at least as an alternative state). This is almost a tautology in that all it says is that being conscious is different from being unconscious and that active processes are needed in order to be conscious, as opposed to the default state of non-consciousness. I will not consider pan-consciousness further
• operates at an aggregated summary input and command output level rather then on individual input/output (afferent and effector) nerves
• directs its focus on a limited subset of the incoming perceptive information available to it. i am not sure if this is just a current limitation of or a requirement for consciousness
• appears to experience a usually continuous flow of near-sequential awareness of a subset of the external events and internal changes available to it
• appears to be able to replay previously experienced sequences of this continuous flow
• demonstrates continuity with immediately prior episodes in what it perceives as a continuous flow of awareness, as well as demonstrates gaps or discontinuities between other episodes (i.e. between a previous stream of awareness and a subsequent stream there is a gap explainable perhaps by sleep or anaesthesia)
• bridges the functional space between newly arriving sensory inputs and long term memories, associating the incoming sensations with more permanently retained and categorized similar sensations
• Is a process contributed to by lots of beyond consciousness processes working in the background
• Allows for selection and focus on ideas/concepts/thoughts/internal representations of real world happenings for further (conscious) processing which otherwise wouldn’t get gone unconsciously
• Ideas/concepts/thoughts/internal representations of real world happenings can intrude on existing focus
• Sequences of previous external or internal ideas/concepts/thoughts/internal representations of real world happenings can be sequentially rerun
• Such reruns can occur independently of any external inputs
• New sequences of actions that have not yet occurred can be rehearsed and run through in different variations for evaluation or decisioning, independent of sensory input and motor outputs
• Ideas/concepts/thoughts/internal representations of real world happenings are bound with specific and consistent emotional ‘feels’
• Each idea/concept/thought/internal representation of real world happening is an integration point, or at least has access to all directly related sub-components
• The step-wise review/evaluation of each idea/concept/thought/internal representation of real world happening can be redirected at any point to traverse to a similar sub-component directly related to that component, and then continue stepping through that other idea/concept/thought/internal representation of real world happening – thoughts can trigger other thoughts
• Conscious processing can be an un-directed flow - daydreaming, musing
• Capability for intense focus (through an ‘act of will’) that excludes competing ideas from entering conscious awareness that normally would (i.e. pants on fire)
• bridges the functional space between newly arriving sensory inputs and intentional actions, associating inputs with appropriate resulting actions. It reacts to and initiates changes on its environment which are not simply reflexive. It can be seen as an exotic kind of control system that forms its own short and long term goals, chooses a subset of these goals to satisfy, and initiates physical or mental actions which advances towards fulfillment of the selected goals. Understand however that all control systems are not necessarily conscious - this alone cannot be its definition

Consciousness is the linking together of a shifting subset of the representations that make up thought which are distributed throughout the cortex. The vividness of a conscious thought should then depend on the depth and detail in each of the parts spread throughout the brain that are linked together at this time.⁷

Real experiences come in through multiple channels and in almost channel saturating force. Internally generated mental experiences on the other hand represent only the most intensive, the most needy, the most novel etc. of the multitude of possible 'next thoughts' i.e. the few at any one time winners of an internal fight to get linked into consciousness. This need not activate neural components as intensely as immediate experiences sourced directly from the outside. On the contrary, the winner is more likely the most intense of a very weak lot, the result of a brain in idle.

This means that the more rich, detailed, thorough and multifaceted a concept can be represented in a brain, given the same consciousness generating mechanism, the "higher the intensity" of the conscious experience. The more specialized and multifaceted the brain is for x, the more deeply consciousness revolves around (depends on, is connected to) x.

Imagine that someone is able to build an artificially intelligent device that exhibits all of the above. It might exhibit the same behaviours as would any human being in similar circumstances, but would it, could it actually be conscious? It would effectively sit in the very same black-box classification/categorization that we put every conscious person other then ourselves into. How can I be sure I am not a self deluded "simulation" of consciousness? It could even believe and say that it is experiencing conscious thoughts! The real question is - what is the gold standard of consciousness? Exactly what determines that this is conscious but that is not? This is why we need a solid definition! Lots of words here, but nothing definitive yet.

By the way, I firmly think that it is madness to build up theories of something we can't crisply and clearly describe. Most of what has been written is therefore junk (albeit with nuggets of value scattered here and there). It is akin to the middle ages argument about how many angels can dance on the head of a pin (42 I say!). I am leaning toward a clumsy and awkward, multi-part combined behavioural, functional and self-reporting definition right now - but I reserve the right at any time to substitute it with a better one. Achieving consensus in this definition is what I consider to be one of the most important parts of consciousness research that needs to be addressed.

What I am going to use as a temporary, stopgap definition is the term phenomenal consciousness (a.k.a. P-consciousness) which is the experience of qualia. In 1996 David Chalmers called this the hard problem of consciousness. I think that a different aspect of consciousness, namely self-awareness, is just a special case of this. Access consciousness (A-consciousness) is the phenomenon whereby information in our minds is accessible for verbal report, reasoning, and the control of behaviour. This too is fair grist to my mill. Mental events that are not within A or P-consciousness are unconscious, which I regard as being similar to multi-layered reflexes - local and autonomous. In short, for my purpose consciousness is defined as perception from a first person viewpoint. If we can demonstrate awareness of things from that viewpoint, then we will have demonstrated consciousness.

Other candidate contributors to a definition that were discarded include:

• awake = conscious? not enough as most lower animals are considered to be wake-able but not all can be considered to be conscious
• simple perception = conscious? Not enough. Again, lower animals can be shown to exhibit perception, but so can thermostats, missile tracking systems, which are clearly not conscious.
• self awareness = consciousness? Not enough. This is just a special case of perception except it involves perception of oneself. It doesn't say anything about the 'thing' that is doing the perceiving.
• thinking = conscious? This is a difficult one. My ruling on this is that consciousness is closer to awareness of thinking, then the actual process of thinking. What is thinking, by the way, if it is considered to not be consciousness? I believe it is equivalent to or at least incorporates logical reasoning, as well as the recognition of being in a particular mental state ("I feel cold", the drawing of a conclusion ("Being cold is not fun"), and making of a decision ("I should put on some clothes"). It is closely entangled with consciousness but is not identical to consciousness. Thinking might be considered as the aboutness or intentionality (in the true philosophical sense) of consciousness, the thing that consciousness operates on
• claim to experience consciousness = conscious? not enough. A non-conscious device could easily be programmed to claim this
• demonstrated focus/attention = conscious? not enough since electronic devices, which are clearly unconscious, can be constructed that are able to track (focus on) objects
• Passes a Turing Test⁸ = conscious? This seem to me to be far too subjective.

The Classic Themes - My Take

The Chinese Room

It goes something like this⁹. If you had a room with a slot in a door and you passed in slips of paper with Chinese text written on it, and reasonable responses written on paper in Chinese came out afterwards you might think that someone in the room understood that language. However, if you knew there was a uni-lingual non-Chinese speaker inside who was simply following a script of very complex instructions for dealing with specific inputs, would there still be understanding happening? I say yes - the whole system understands Chinese, in the same way as a whole brain understands it. The man inside the room does not, in the same way that a neuron of that brain does not. Understanding depends on previously encoded information (the complex instructions script in the one hand, episodic and semantic memory in the other), a way to match current experiences (inputs) to appropriate parts of that previously encoded information, and a subsequent capability to act using the result of these two now integrated things.

But what is actually doing the perceiving?

If we don't use the word 'simulated' and instead use something like 'build from the ground up' in order to produce a conscious (a truly understanding to the best of its ability) thing, then we need to build pathways to and from it (so we can stimulate it and then afterwards see what it does with that understanding). The inputs need to be integrated in the right way to produce the resulting outputs. If we were to determine that self-awareness is necessary for consciousness, then some of the outputs would need to be fed back in as inputs. If pleasure and pain (or simple positive and negative reference or reinforcement) was thought a necessary prerequisite of consciousness, then some sort of arrangement would need to be constructed that would encourage or discourage future behaviour as desired. If such positive reinforcement needed to be grouped together with other positive reinforcements, then a new concept (pleasure) would need to be set up and integrated into the rest of the system. It is the whole system we are building that ultimately does the understanding, not individual parts of it. This parallels the human brain in that is whole systems within the brain that contribute to understanding, not the individual neurons.

There are already lots of existing theories of what consciousness really is - the mechanism, the process, the end to end flow of things necessary for something to be really conscious. Here are a few of the less outrageous ones.

HOT

Higher-Order Thought is supposed to explain consciousness in this way. An unconscious thought enters or is formed in the brain. A different unconscious thought is subsequently formed that is about the first one. The fact that a thought about a thought exists is somehow supposed to make the subject thought conscious. I am not convinced.

Cartesian dualism

Descartes thought that consciousness resides in the realm of thought and that the brain resides in the realm of physical things and that the two meet in some unknown part of the brain. I don't buy it. For one thing a non-physical thing by definition doesn't interact with a physical one. A variant of this is substance dualism in which the mind and thought is one type of substance that is less subject to the laws of physics and the brain is of a different type that is more conforming, but there is no evidence for this either.

?

something I have forgotten just this minute is supposed to work thus. An input into the mind is supposed to carry with it characteristics that make it either conscious, or not. No one seems to be able to define that something, so this argument carries little weight for me too.

Obviously not quite there yet.

Autonomous neuronal events happen in the brain all the time. Some enter into conscious awareness, most do not. The ones that do, don't always so there is probably nothing special about them other than the fact that they were somehow admitted in. Once in conscious awareness an event becomes available for direct conscious processing, whereas unconscious events are not. Conscious events are maintained in rough sequence within a rough time-line, become eligible subjects of an inference of cause and effect or agency, can be manipulated in the absense of any matching environmental actions, can be re-played, and so on. Unconscious events/concepts can participate in none of this. Perception of one's own perception of one of these events does not seem to me to be an absolute requirement of consciousness. Perception itself can be distinguished from the lack of perception. Perception requires a subject, an object and the act. The perceiver in this case can only be the concept of 'self'. Elsewhere I describe the general handling of 'concept' in my theory and 'self' looks to me like it is just a special case of 'concept'. What is now needed here is a better description of 'the act' of perception.

Qualia

The subjective quality of a thing (e.g. the 'redness' of red; the pain of a pinch) is a hard thing to understand if one thinks about it as a quality of an experience which is experienced. It implies both that there is something doing the experiencing and also that there is some aspect of an experience that is more then just the experience. If we instead understand that the totality of our experience comes from the biological processes going on in our brain, then the subjective quality of a particular thing is also just an aspect of experiencing that thing. I think the 'redness' of red has a lot to do with our previous total experience of red - all of the red things we have had contact with in the past and all of the associated inputs (originating either externally or internally) such as how good that apple tasted, how hot that red poker felt, memories from that red trimmed school room, how painful that cut with all the red blood felt etc. A current experience of red would activate to some degree each of these older memories, biasing our processing of the current experience with the previous lessons learned (and other associations integrated) in the past. Current sensory inputs that involve 'red' automatically brings up prior experiences that involved red in some way. This priming or biasing of the current experience with prior experiences seems like a useful evolutionary adaptation to me. I suspect something like this explanation will be determined to be the real explanation of 'qualia'.

Emotion

The loss of a loved one is a lot like physical pain. Falling in love is very much like being actually intoxicated. Anxiety really does cause physical abdominal disturbances that can be interpreted as feeling the same way that butterflies in the stomach might actually feel like. Severe disappointment can cause very much the same sensation as what a punch to the stomach might feel like. Excitement really does make one subjectively tingle (due to hyperventilation) and objectively hyperactive. Milder versions of these and related emotions all probably have much the same origin as my explanation for qualia just previous to this section. Past physical sensations, together with everything that happened each time the emotion was experienced, all serve to put the body and brain into a particular baseline or state unique to each emotion. That state biases both behaviour and the interpretation of inputs in different ways that depend on each emotional state, and each collection of biases is different from a neutral state set of behaviours and interpretation of inputs.

Jaak Panksepp[23] believes he has come up with at least seven networks of emotion in the brain: SEEKING, RAGE, FEAR, LUST, CARE, PANIC/GRIEF and PLAY (note that he capitalizes them because he believes they are so fundamental). If he is right then emotion is a very ancient thing and, as anyone who knows mammals can attest, should be viewed as a fundamental component of even basic consciousness.

The 'Thickness' of Time

If we didn't have the concept of time (or at least the concepts of 'before', 'now' and 'after') then everything would appear to happen at once. Clearly it doesn't. Is this just sequentially and subtly different associations forming over time, and later experiences forming new links to earlier ones, but not vice versa? Probably not. There are interesting experimental results in which perceptions resulting from fast inputs to the brain are delayed and remains outside of consciousness until corresponding slower inputs to the brain from the same external event arrive. Subjectively, both seem to occur together¹⁰. Clearly the brain struggles and goes to great lengths to keep a unified and internally consistent time line going. The question is why, to what end? Why is it important to the conscious part of an organism that all aspects of an external event are made to appear to happen simultaneously? Does this simplify models which only the conscious part works with?

This 'thickness' of this perception of instantaneous time (i.e. that fast input/slow input two step) is not as thin as it otherwise might be, and could be seen as the result of (for example) some stimulus first being processed in a sub-conscious neural network by means of spreading activation, where the resulting subset of activated concepts then serve as inputs to a second (or overlaid) neural network representing the neural correlate of consciousness. The longer time taken to finally activate the overlaid network is a result of network overhead. This delay could be undermined by direct inputs into this conscious network (i.e. the unconscious first network hasn't had a chance to fully process and propagate its stuff yet, or it is bypassed by late inputs which are fed directly into the overlaid network.)¹¹

If there is some few seconds that pass between the instant a fleeting thought or sensation comes into our consciousness (all parts of it: see the above paragraph) and the last vestiges of it, then it must be stored in some equivalent of short term memory.

Free Will

Experiments seem to strongly indicate that if allowed the time, parts of our brain reliably predict what we are about to choose as far as 10 seconds in advance of our conscious perception of having made that decision¹². This is in keeping with the 'lots of unconscious processes that are filtered and distilled into a more summarized and focused conscious thought' model I am building here. These processes all work together and form a consensus. It is that consensus that enters 'consciousness', that is it is that consensus that determines what is to be attended to next, out of the many possibilities. We have free will alright, but only at an unconscious level, which hardly makes it free. This makes religious, lobby, government and corporate deliberate manipulations of our unconscious far more scary and ominous then it otherwise should be.

The whole thinking system, not just the conscious part of it, should be thought of as a prerequisite for having has free will. This is analogous to a whole thinking system contributing to the generation of conscious experience and not so the individual neurons themselves. A selection of alternatives is presented to the whole system and the whole system chooses the one that best satisfies its immediate and long term wants and goals. The concept of determinism (dominoes of events triggered by generations of previous events all the way back from the dawn of time) limiting free will can be safely separated and addressed differently from the idea that unconscious brain processes influence or entirely determine the outcome of seemingly conscious decision making. In this latter case free will is reserved for the whole brain and is a fiction of the conscious brains awareness. In the former case, free will is reserved only for the prime motivator, the initial random arrangement and impetus that set the original first events in motion, and is a fiction that the whole thinking system is usually not unaware of but nevertheless is deterministically subject to. Free will exists, but not in the way that we intuitively think it does.

The big question here is why does the conscious part of our brain think it is in the driver's seat? Why does it think it has ultimate control when in fact it is probably only informed after the fact? Why is it apparently kept totally out of the decision making loop? For one thing, a world view that doesn't include the many and mostly unimportant unconscious component processes certainly simplifies things for a conscious brain. Also (and this time on a different track) even though the actual decision might be made unconsciously, that doesn't mean that preliminary conscious processing wasn't used as input into the unconscious processes leading up to that decision. Conscious rumination probably contributes strongly to unconscious decision making - otherwise why do we do so much of it? (weasel-words, I know, but they're all I've got at the moment.)

I will go further. 'Conscious' attention is a mechanism that is used to bias the processing of inputs in a direction that best satisfies current and long term goals. Once the goal is satisfied in some way (or a different goal's priority becomes dominant over the current goal) then the processes that satisfy the first goal are not given further attention and new processes that satisfy the next goal are.

Conscious attention is the brain's way of influencing which of many possibilities to consider next. It allows the mind to hold a train of thought and it allows the brain to bias the next iteration of processing that will be done in a particular trajectory. Concurrent conscious attention will make a mental concept encoded in neurons easier to trigger, when presented with a set of external stimuli, then it otherwise would be with just that set of stimuli alone. Conscious attention to a matter fertilizes the brain bed that will host the next thought.

I will go even further. The brain is extremely adept at detecting both differences as well as correlations. In the free will context, an individual brain throughout its entire life has historically experienced extremely high correlations between attention to a train of thought, and the selection of the same or similar aspects of that train of thought for further processing. Based on this high correlation it is easy for it to infer a cause and effect relationship - which is to say it is easy to come to believe that consciousness determines what the brain will next attend to, or causes its body to do next. I think this is an important deduction which itself is crying out to be tested.

What is a Thought?

In 2007 Steven Pinker wrote a book called The Stuff Of Thought - language as a window into human nature¹³. Ultimately I was disappointed by the book, but remain intrigued by his question.

For now, I consider a thought to be one or more concepts made conscious (i.e. enters into first person awareness, whatever that is). How is that done? What does it mean for a thought to be conscious? What does it take for a thought to be conscious? These are the big questions that I am trying to answer further into this document.

Why Are We Conscious At All?

I apparently can drive my car some distance without being able to later remember particulars of actually doing it. Similarly sleepwalkers cook, clean, engage in normal sounding light conversation, and supposedly even commit murder and other complicated things while not really being aware of it. Have they been conscious at all through these episodes, or is the problem (if there is one) with the laying down of memories of that experience and/or with the later recollection of them?

Such a mindless person performing activities (such as driving, cooking or eating brains) can be termed a zombie - not the walking dead kind, but a philosopher's zombie none the less - a sort of personified thought experiment. These people are envisioned as acting identically to everyone else, but there is nobody home, no consciousness at all. Yet everyone actually experiences things, from a first person viewpoint. Why aren't we all unconscious zombies? It must be that consciousness adds capabilities that are not possible in unconscious-only processes.

Our brains are always being stimulated by things that enter via our senses - 24 hours each day. It can be demonstrated that the vision pathways of the brain pick up changes in the visual fields of even anaesthetized animals. Sometimes the transition from sensory input, to working memory, to conscious awareness, and on to short term memory and finally into long term memory is interrupted - causing us to never become aware of the sensory input as an experience (during deep sleep for example) or to immediately forget the experience (e.g. the effects of administration of diazepam, propofol or scopolamine). It might be that we were never conscious for that period at all (a truly unconscious automata) but just appeared to be, so there is nothing even to make a memory about.

I think that the very fact that the brain can perform complex acts (like driving - steering, braking, lane changing and vigilance) by it self when the conscious mind is otherwise engaged means that it is acting unconsciously but still under a level of control and is not entirely abandoned to an unattended and simpleminded autopilot. The fact that an incident on the road can immediately cause the situation to take possession of the consciousness means that:

• the unconscious capability has its limits
• the unconscious capability knows what to do when these limits are exceeded (i.e. strongly influence the direction of full consciousness towards the situation)
• consciousness is necessary to deal most effectively with unexpected, complex or life threatening situations (a road accident ahead, a dropped cigarette, a loud noise etc.)
• consciousness is not absolutely required for most defined, rehearsed and constrained demands on our mental functions

The way that we do things consciously is probably the same as the way that we do things unconsciously. The only difference is that these things are attended to for some reason, and consciousness introduces something extra into the mix.

Nature never evolves something out of the blue. Consciousness most likely arose as an extension of preexisting unconscious processing. This means extending the lowest level of autonomous process with small autonomous processes that have oversight over the lower level, and those by more complex and richer overseeing processes (themselves unconscious). These overseeing processes are moderated and coordinated by even higher level processes, and so on. Note that the highest level processes don't deal directly with what the lowest level processes deal with. They only work at a more abstract 'summary' or 'aggregated' level. Consciousness is associated only with one or more of the highest level(s) of this hierarchy. For example consciousness does not deal with a single receptor nerve, but rather the highest useful categorization of what each field of receptors has is set up to receive (a pain in the stomach, pressure from the surface of a chair, sunlight on one's back, or what ever).

The difference between a conscious state and an unconscious state looks to me at this time to be a difference of degree, not a difference of kind. There is probably very little that a conscious state could do that a high-level unconscious state could not if placed in the right circumstances. A difference of degree, but what a difference that little difference makes. The specifics of this difference is crying out for explanation.

The above point seems quite important. Any implementation of an artificial consciousness must first have solved and implemented the unconscious problems of monitoring, motivation, goal setting and follow through, attention, daydreaming and a host of other problems, before any of this list is made to enter into a consciousness.

Taking this levels of oversight concept further… Since consciousness is not dealing with the minutia of everything, it can deal most effectively with the bigger problems in life, longer term plotting, planning the immediate ones that can't wait, the instant by instant monitoring and decision making that has to happen, but also if appropriate the larger problems that can wait and long term and analyzing. This is where the extra ingredient, the value that consciousness provisions is brought to the table.

Consciousness in this narrow discussion refers to an ability to perceive (that is to mechanically make the fact that has something changed from the previous state available for use) first person involvement in some mental process or physical behaviour. The important point here is the 'first person' viewpoint reference. I think that this is a key definition that should be considered in the context of the 'big problem' of consciousness. If you don't have this, you aren't conscious. Things may be happening, but not to you.

Attention and Prioritizing

I can see both aspects of a necker cube, but only one aspect at the same time. My wife can attest that while searching for something I often look directly at it without seeing it. Once pointed out, it is easy to see. And like everyone else I am subject to change blindness, as well as the various visual masking techniques that some vision researchers use. There is nothing wrong with my visual system. I simply am not consciously attending to what my eyes are picking up. External inputs make it into the visual centres of my brain, but might or might not not enter into my conscious awareness.

This fork in the path (to consciousness or not) must have little or nothing to do with the incoming information. My consciousness is just not attending to it. Therefore there is a gating mechanism that makes these decisions, itself unconscious. This mechanism must take into account characteristics of the incoming information, moderated by some measure of its importance, as well as my internal drives and goals. I will even bet that the net result of things that have made it into my consciousness can influence these decisions too. After all, I can hold a thought in spite of sometimes intense pressure to turn my attention to something else.

There are apparently separate systems to prioritize or attend. Wikipedia has an interesting article on the neural mechanisms behind shifts of attention. Utah researcher J.S. Anderson¹⁴ found that the intraparietal sulcus has at least seven connectivity maps in each hemisphere, topographically organized maps of 'all the things one can pay attention to' - apparently the auditory, visual, somatosensory, and default mode networks. This architecture makes the intraparietal sulcus ideally positioned to mediate all attention, if not only conscious attention.

We need to differentiate automatic reflex type central processing from much more complex and longer loop processing and also from processing that is complex to such an extent that is it at least eligible for (or actually gets) conscious notice. I suspect the fundamental difference is a difference of degree only, except for the latter in which a fundamentally different high level process becomes involved.

From first principles (i.e. me thinking about this) consciousness needs:

• non-trivial inputs, which are not necessarily from the outside
• options in how these inputs can best be dealt with (ignore, leave to automatic processing, select the best, ruminate for new options…)
• a mechanism to evaluate and value the intrinsic result of each of the options (along various lines of criteria, but it must be simple to be quick and effective, so a global valuation for each would be best if that can be done.)

Wikipedia also has an article about attention itself that lists these core processes. I have reworked a few to better fit my ideas.

• Information is made available for detailed analysis.
• Competitive selection is the process that determines which information gains the right to further processing.
• Through top-down sensitivity control, higher cognitive processes can regulate signal intensity in information channels that compete for access to high priority processing, and thus give them an advantage in the process of competitive selection. Through top-down sensitivity control, the momentarily activated content can influence the selection of new information, and thus mediate voluntary control of attention in a recurrent loop (endogenous attention).
• Bottom-up salience filters automatically enhance the response to infrequent stimuli, or stimuli of instinctive or learned biological relevance.

Implementing these core processes as described would result in a mechanism that will always and only focus on one thing at a time - the specific thing could change over time, but attention would happen sequentially. What is missing is the almost universal ability of people to pay attention to multiple candidates for long term attention, and only after due consideration/comparison start to concentrate on one. Said differently, mental 'trial runs' must be evaluated to see how they might benefit from attention, and then kept somewhere 'in mind' while other trial runs are happening. A valuable end point can be achieved by a variety of means, and selecting the best one somehow involves briefly devoting attention to each alternative, and then further concentrating on the one chosen.

In this way the single in situ representation of a unit of thought need not be moved into and out of some hazily defined concept of 'global work space' (movement of which I think is not necessary). Instead, different alternatives would activate their various representations where each was initially stored, more or less strongly depending on their neural net type weighting (a virtual global work space, defined only by what is activated at any one time together with the record of the prior activations that led up to this one). This is a dynamic activation variation of Bernard Baars workspace theory. The activation of a set of such units would be momentary, but enough to prime some kind of longer term valuation¹⁵ that happens to be transiently associated with each set. The best valuation corresponding to the best set of activated neural units would represent the best candidate to devote further processing towards.

This implies that a thought held for a period of time in consciousness does not need to be continuously activated including all of the concepts that are strongly related to that thought. This also implies that 'pointers' to that selected complex of interrelated concepts do need to be maintained over time. Periodically, probably quite regularly and possibly several times per second, all the competing sets of complexes will be re-examined to see if it is time to devote 'attention' to a different complex. The large related neural complex that represents each candidate 'thought' does not need to be activated to hold that thought. In fact a background valuation process can cycle through these candidates, briefly activating one after the other, in order to judge the winning value available at that moment. (This decision making site in the brain is probably in or functionally near the striatum.)

Possible Uses/Functions of Artificial Consciousness

Look at it from a different angle. In exactly what ways would society make use of artificially conscious entities? What features are so special about them that would cause them to be more useful then traditionally programmed non-conscious devices? It seems to me that adding conscious control mechanisms to self-opening doors or toasters (a la the plot device used in the Hitchhiker's Guide to the Galaxy series of novels!) would be overkill. However in complex situations simple if-then-else or case logic fails - especially in cases where a human would traditionally be called in to make a definitive determination or judgment call. A device that is aware of itself, of its previous and potential actions and their consequences, one that can be reasonably expected and trusted to pro-actively identify every contributor to a situation and to weigh the pros and cons of each in isolation and together, might be useful. This is the very kind of machine that science fiction writers have the human race unadvisedly abdicating key responsibilities to. This is also the kind of machine that could autonomously pilot a space ship, diagnose and treat your illness, contribute to corporate, government, legal or military decision making, perform original research and if given a personality it could even act as a companion. It would have to:

* function autonomously
* function in a more complex way then is possible with simple reflexes
* be motivated to strive to achieve a useful level of timeliness, skillfulness, reasonableness, knowledge, background, fairness, and possibly produce some sort of self-satisfaction afterwards as a means to these ends
* adapt to and successfully cope with novel situations
* prioritize its tasks
* logically come to reasonable conclusions even when based on incomplete evidence
* converge on a conclusion when presented with ambiguous inputs
* select the next thought (thing to focus attention on)
* carry a thought over time when other thoughts are competing for this attention
* recognize significant or otherwise meaningful differences
* bring all available considerations to a problem together and make defensible judgment call or conclusion
* poke around at an idea (concept/thought?)
* in short - think for itself.

Notice that 'learn for itself' is not on this list. I think you can have a conscious and engaged entity that keeps repeating the same thing when presented with the same inputs. I don't think that an inability to learn and 'self program' would disqualify an entity from being capable of achieving consciousness - it would just be less useful but perhaps more predictable. The intrinsic knowledge required could be pre-loaded as a factory default, rather then learned from scratch.

…However, I am deeply uncomfortable with this notion at present. The diurnal cycle is a very ancient one which is observed in all mammals, birds, and in many reptiles, amphibians, and fish, and that when deprived of sleep rats will die after about three weeks (of unknown causes). It is believed that during sleep unstable (or transient) short term memory is consolidated and re-consolidated into long lasting representations. Apparently either SWS (slow wave sleep) or REM (rapid eye movement) sleep is better then no sleep for such benefits, but both types of sleep in one night is best. This means that the only normal loss of consciousness we hope to ever experience serves an important purpose. The act of learning consolidation (presumably) is tied to a need to interrupt consciousness repeatedly during our lives. This implies that consciousness is somewhat incompatible with learning consolidation, or at least that unconsciousness is somehow required. What can we do while unconscious that we can‘t while conscious? Dream perhaps, but that mostly but not always occurs in REM sleep. A more satisfying answer is that we can disconnect (most of) our brains from the consequences of our motor actions during this time. That alone tells me that most of the same mechanisms that result in motor actions are also used or re-used for the consolidation of learning. Our sleeping brains are most probably replaying or rehearsing fragments (or even entire episodes) of our conscious experiences from the previous wake cycle. The only thing differing from awake experiences and sleeping replays is the origin of the activity. Remember that experiences are not recorded like a film or tape, but are broken into their constituent aspects and retained widely, probably across the whole brain, but mostly in the hippocampus and amygdala and only in a temporaryway. This needs to be incorporated into the form of long term memory that is only slightly influenced by most day-to-day activities…in quite a different form then the working or short term memories are. Repetition and rehearsal, repetition and rehearsal, repetition and rehearsal…why does that sound familiar and so important? I can’t think at the moment as I am currently sleep deprived. I think I will sleep on this question. In any case learning and consciousness are quite tightly tied together somehow. The question however is - is a period of unconsciousness (sleep) really required for learning, or is this just the result of nature reusing some of the materials at hand?

Tononi [31] suggests that consciousness in an entity is (or at least is extremely closely related to) the degree of integrated information available within that entity. Reducing the number of integrated nodes of information (even if they are not firing during the experience of any particular qualia) reduces the quality of the consciousness that is doing the experiencing of that qualia. Conversely, increasing the number of integrated nodes would enhance that quality. A machine or AI version of a conscious entity could probably be designed and constructed to surpass the human limitations concerning the integration of the information available to it. There is no telling what capabilities this might endow the machine or AI with. Hitherto unsolvable problems in all fields of endeavour might be able to be finally addressed by such a 'being', and new ones identified.

How Consciousness Might Be Reliably Determined

A completely paralyzed quadriplegic with a fully functioning brain (unfortunate soul!) should still be considered to be conscious even it this person cannot initiate an action or respond to an input. A dead person, or a thermostat, probably should not be considered to be conscious. How should we reliably infer the existence of consciousness in borderline cases? Where is the borderline? Here is a list of some candidate criteria (which is probably going to be redundant with the specifics of the definition as listed above):

• be recognizably distinct from an unconscious state, itself recognizably distinct from an inactive or dead state
• demonstrates selective attention, filtering and focusing raw input into higher level concepts (someone is paying attention)
• exhibits serial attention – by which I mean is able to focus on a set of inputs over time, and then intermittently switch focus and hold its attention on another set (someone can hold a thought over time)
• displays intentionality in that it can initiate novel trains of "thought" or motor actions (someone is starting something)
• exhibits motivation, determination and goal seeking behaviour (someone has a purpose)
• reacts and adapts in non-stereotyped but appropriate ways to changing conditions (someone is awake, and actually driving)
• exhibits non-stereotyped emotional state fluctuations independent of changes in its internal or external environment (someone is living an inner emotional life)
• claims or appears to be aware of being aware (someone is conscious of being conscious)
• exhibits or claims that "thoughts" pop into its awareness fully (or mostly) formed. (someone has a division between unconscious and conscious mental activities)

What Consciousness Is Not

Dennett [9] describes several kinds of creatures which routinely exhibit intelligent and seemingly conscious behaviour. I like his list, but feel that no single species or individual must fit entirely into only one category. These increasingly complex arrangements of brain functionality, by themselves do not count as evidence of consciousness, but consciousness is built upon the designs found in these foundational systems:

• Darwinian - creatures possessing this design each commit to a particular "hardwired" approach to a problem, but together as a population exhibit a variety of similar alternative approaches. The individuals that happen to express the most successful approach are the ones that survive and prosper and most influence the make up of the next generation
• Skinnerian - creatures set up in this way are wired to blindly and sequentially try multiple, mostly random approaches to a problem. Creatures that survive the most incorrect choices and those that luck out early on the most successful approaches are the ones that prosper. The best choices will be among the first to be tried next time
• Popperian (named after the philosopher Sir Karl Popper) - creatures that use this architectural approach first preview and evaluate the already available candidate "canned" approaches and then select the most promising to act upon, the first time, thereby reducing the carnage that can result from bad first choices of the previous designs
• Gregorian (named after British psychologist Richard Gregory) - creatures with brains set up in this way can learn from the experience of others, from analogous previous experience, and from direct observation and logical reasoning, and make far-better-then-chance correct behaviours the first time, by virtue of being able to encode all these experiences into symbols and to manipulate the symbols.

Each of these differently architected control systems is capable of producing complex, subtle, seemingly conscious, and intelligent looking behaviour. But none of this is necessarily proof that there is someone "at home" experiencing, deliberately evaluating and directing any of this, consciousness is not required in any of these types of creatures (which themselves can be seen as being composed of increasingly complex and layered collections of reflexes). Similar behaviours can be artificially programmed into computer games, and no one at this time is arguing for self-aware and conscious computer games - yet. These kind of brain arrangements I still consider as collections of increasingly sophisticated and dynamically adaptable reflexes, but reflexes none the less. There does not seem to be a need to include the assumption of any sort of consciousness to these entities.

While top down and bottom up inhibition or potentiation is necessary for proper cortical functioning, I am not yet convinced that these mechanisms are essential for basic consciousness.

While prediction and error detection of actions that deviate from that predicted state are necessary for brain controlled guidance of movement, I am not yet convinced that these mechanisms are essential for basic consciousness.

While input from external stimuli and output to end effectors (i.e. muscles) is necessary for externally detectable evidence of active guidance and internal control, I am not yet convinced that these mechanisms are essential for basic consciousness.

While most people understand that other people have their own point of view, their own ideas, there own motivations etc. (this is called a Theory of Mind and it is deficient in disorders such as autism), I am not yet convinced that the ability to understand this is essential for basic consciousness. I have read enough self-reports from high-functioning autism spectrum afflicted individuals that I am convinced these individuals are quite conscious.

While knowing that a bug is an insect and that self propelled insects are living things and that the group of living things also includes people with beating hearts but not people with non-beating hearts is useful, I am not convinced that the ability to categorize inputs into concepts and concepts into one or more hierarchies is essential to basic consciousness.

Intelligence vs. Consciousness

Note that a mind that is conscious is not necessarily an intelligent one. I can point to a couple of acquaintances to justify this statement. I think of intelligence as a measure of the degree of successfulness of an organism's ability to deal consciously (in non-reflex ways) with novel stimuli - the more successful the strategy and the more novel the stimulus that can be accommodated the more intelligent the thing is. Most normal organisms more complex then insects are probably conscious when they are awake (in the something-is-experiencing-awareness sense). But a conscious mind is not equivalent to an intelligent one. Most of us are probably intelligent (especially if you are reading this!) and remain so indefinitely but we are not always conscious…i.e. not during sleep, etc. The two terms have little directly to do with each other and one is not necessarily dependent on the other. 'lower case' artificial intelligence is being built into all sorts of devices - in digital cameras, cars, medical diagnostic software and bank fraud detection systems for a few examples. None of these are ever thought to be even close to reaching conscious but it is can be argued that the decisions they produce are appropriate to their inputs and they can often match or beat the responses of the best human experts in the respective fields to those same inputs. On the other hand, an infant is mostly a sometimes-conscious collection of instincts (with a lot of potential, of course) but not particularly intelligent.

In popular literature the two terms are often incorrectly used to mean the same thing. I notice that in some more rigorous papers this confusion happens too. Be careful when you next consider the terms 'intelligence', 'A.I.' and consciousness, so as not to confuse them. My interest is the study of consciousness (~awareness and experiencing), rather then understanding the nature of intelligence (~better or more adaptable coping strategies; a.k.a. 'smartness'). A successful A.I. implementation does not necessarily imply an implementation that is conscious[footnote]upper case" A.I. does however.[/footnote], and vice versa.

old stuff I was unhappy with the way my first attempt was developing (style more then content) and moving it here keeps it available for my reference while I work on its replacement. I told you this was not going to be pretty!

Mind/Personality and Consciousness

To me a personality is a reference to the set of major individualized characteristics (or trends, preffered directions, preferences) of some person's conscious thought patterns. The mind is another name for the set of a person's conscious thoughts. No mystery here.

Some Neurophysiology Background to the Technical Terminology I Use

In these notes when I say a concept in the brain is ‘activated’ I mean that stimulating a region in the brain will cause a decrease in the magnitude of an electrical potential difference across the membrane of the nerve cells (negative on the inside) in each cluster of neurons in this area corresponding to the physical implementation of the concept, as compared to its normal resting state. Enough of these input events (either simultaneous or in quick succession) will cause the state of the cell membrane to cross a threshold and trigger a travelling ‘action potential’ in which a wave of membrane depolarization spreads over the whole cell and then quickly re-establishes its former state. This travelling wave can jump from cell to cell and thus spread throughout much of the cluster. Even if the input isn’t sufficient to start this happening, a decrease in the potential difference across a cell membrane can make subsequent stimulation of these neurons more likely to result in this spreading electrochemical wave (I sometimes refer to it firing or discharging). I call this higher likelihood of a cell firing at the next stimulus potentiation, stimulation, biasing or making it more twitchy. The opposite effect, increasing the potential difference across a cell membrane inhibits the subsequent responsiveness of the neuron. These are also the mechanisms of positive and negative feedback loops, the stimulation or inhibition of the involved cells. The propagation of action potentials from one cell to another is actually the result of neurotransmitters crossing synaptic gaps and interacting with receptor cell membranes. I am proposing nothing new here in the nature of the fundamental mechanisms of physical propagation of information throughout a neural network. These low level plumbing details don’t seem to me to be too relevant to the study of consciousness, which happens at higher levels of organization. I think the specifics of which neurotransmitter goes which which human brain function, the neuromodulators, the inhibitors, the re-uptake mechanisms etc. can all safely be ignored for the time being, at least by me for these notes. A great overview can be found in LeDoux's Synaptic Self [19]

Concepts as stored within the brain are hierarchically related to other concepts (i.e. face is related to its components like eyes, nose, mouth, chin…) A concept can also be related to peer concepts (i.e. right eye, left eye); and also distantly related to distinct concepts by way of overlapping categorizations (i.e. eye witness testimony, a gun, a wire tap transcription and a bloodstain are all related as evidence at a murder trial). In practise there seems to be no limit to which concepts can be related with others, including the concept of nature of each relationship. If everything can relate to everything in every way, how are specific concepts appropriately active at one time, while others are not? This has to do with the degree of salience and relevance with what is active at the time as well as quiescence of non related concepts, perhaps by negative feedback or just the absence of stimulation. An ‘eye’ concept is probably never active in isolation, but all of its directly related concepts are, if not actually actively firing off action potentials, then at least more likely to fire. Other more distantly related to ‘eye’ concepts are only mildly biased toward being more ‘twitchy’. But the web of concepts directly and indirectly related to another active concept is likely to overlap components of our ‘eye’ concept. Elements that overlap are especially stimulated and in turn reinforce the activation of concepts directly related to both of them. This makes information that is most relevant and related to the small groups of concept at hand (i.e. the current focus of consciousness) also available.

My Particular Interest and What This Theory Covers

Millions or billions of sensory receptors (sight, taste, smell, balance, fullness, CO2 levels and so on) all converge on the brain-stem, the mid-brain and on to the neo-cortex. Along the way the signals (all of which are neuron action potentials - regardless of the primary sense organ that started them) are compared, collected, filtered, summarized, ignored, grouped, subjected to feedback and feed-forward circuits, and otherwise heavily processed. All of this is as yet outside of conscious perception. You might refer to the end product of this as pre-conscious, proto-conscious, potential-conscious, pre-thoughts, proto-thoughts, potential-thoughts. Whatever you call it, it is the things out of which thoughts are constructed - the prerequisite building blocks of consciousness. This theory is not about all that. Much fine research is going on in that area, including the construction of computer models of the fine details of the visual system all the way from light sensors to something approaching unconscious perception¹⁶. Some lines of investigation include actual budgets too!

Most of the time these things happen automatically, outside our awareness. Often a physical action is initiated that results from something unconsciously detected in this way. An itch is scratched, an eye blinked, a leg position shifted, or something like that. More rarely, quite complex automatic and apparently unguided actions can be produced that still never rise to consciousness. Automatic driving while engaged in an intensely interesting conversation with a passenger is an example - the driver might deny any memory of the act of driving perhaps for miles at a stretch when later questioned. The sensory inputs are appropriately dealt with in a myriad of subtle ways. Since there is so much raw material coming in and so many options to respond with, it follows that there has to be some mostly automatic processes going on that connects the most important inputs to the most appropriate responses. At this point however, the organism that this is happening in would be considered (or would consider itself, if asked) to be unconscious of the automated task. This is a close analogy to the philosophers' zombie, and perhaps also to some particularly advanced robot of current or planned designs. The lights might be on, but nobody is home. It is not consciously aware of (cognition), or concerned for (emotion) nor inclined to do anything (motivation) about the situation. This New Scientist article describes the similar activations generated by an unconscious impression being similar to those activations generated by a similar impression that actually enters conscious awareness, and goes on to describe a specific wider and synchronized activation.

More on the question of automatic driving…
Awareness was not on the act of driving, but was instead on some competing stimulus (e.g. the conversation) - but there was continuous consciousness nonetheless. It was just directed in an unexpected and perhaps not quite safe direction. An example of the other end of the scale is paying conscious attention to breathing. We can consciously regulate breathing, that is we can choose to pay attention to it/to be consciously aware of it, but most of the time we aren't. In this case however ignoring it is no more unsafe then paying attention to it. Breathing and automatic driving are examples of how attention to something is deeply involved in the mechanism of consciousness. It may even be the only factor that determines if a particular nexus of stimulated inputs gets into the conscious mind.

At some point a few of these heavily processed sensory inputs, or internally generated proto-thoughts become available to the mental process that is consciousness. This is what I hope to rigorously think through and come up with a reasonable, testable explanation of. Why is there such a thing as consciously being aware - as opposed to automatic perception and reflex reaction? What is the function of being conscious? Why is it an evolutionary advantage over the presumably minimally aware "lower animal" mind? Why does it work and can the same concepts be made to work in artificial mechanisms - like computer programs or androids?

The exact nature or the specific way these granular sensory inputs are stored, categorized, related, summarized or otherwise processed, and made available to consciousness is probably not critical to the existence of consciousness itself (at least if this theory pans out!). It should not be necessary to simulate or reproduce individual neurons themselves to be able to generate an “entity” that believes itself to be conscious.

What is the area of coverage of the theory that I am trying to build? I can't condense it into a single 'catch phrase' other then to describe it as portions of the hard problem, so here is a list of what's covered and what is not going to be covered. Note that much theoretical and experimental research is being conducted on the remaining areas, just not by me.

Not at all Covered

• categorization of input sensations
• knowledge classification or the internal representation of sometimes contradictory hierarchies of information
• long term storage
• matching of input to a particular internal frame of reference
• normalization of input to a standardized reference frame, or modifying the frame as a result of new input (The letter ‘A’ is a letter ‘A’, even if it is lower case, written on paper, in smoke from an aircraft, engraved on the head of a pin, carved in the side of a mountain, or coded in Morse or semaphore flag).
• functional or mental attribution to specific biological or neurological structures or processes
• prediction, when given some starting input conditions
• error (deviation from that prediction) detection
• mechanisms of short term or long lasting learning
• development of a theory of other minds, that is - how I represent what is going on inside your head
• Neurons, astrocytes or synapses
• Neurotransmitters or inhibitors
• polarization, depolarization or nervous propagation
• Specific brain structures
• Neural column physiology
• Connectivity of the cortical layers within or between neural columns
• Representation of cortical inputs as topographic, retinotopic, tonotopic, somatotopic maps on its single plane surface

Very Little Coverage

• Representation of the hierarchies of composition and aggregation
• Bottom up experience vs. Top down prediction
• Error detection

Peripherally or Indirectly Covered

• direction of attention to the 'winning' goal satisfying output
• free will, or the conscious illusion of free will
• influence of conscious attention on future mental processing
• neural initiation of subsequent neural or physical actions
• goal setting
• valuation of an incoming process with respect to previously set goals
• selection of one or more of subsequent potential processes that best satisfy specific or general short of long term goals
• development of a world view based on the sum of prior inputs
• translation of inputs into potential goal satisfying outputs

Covered in this theory

• experiencing
• introspection
• sense of the past, the present now and a future
• the sense of an I
• the sense of an I in control
• attention
• selection and satisfaction of long and short term goals
• valuation of potential actions (including attention) w.r.t. those goals
• evaluation of inputs that may or may not contribute to achieving these goals
• selection of actions that further a goal
• biasing towards the continuation of a specific action (e.g. continuing a thought in progress, interrupting an action in progress for a higher priority action)
• Nature of integration with previous and future selected actions

Great! Where's the Beef?

Synopsis

• Consciousness is fundamentally made up of processes that occur in a brain. 'Brain process' here refers to the ongoing activation/deactivation of one or more concepts. 'Activation' as used here refers to those concepts that have passed a threshold and are stimulating their peer concepts, perhaps also into activation.
• To be conscious a 'self' needs to be made aware of (notice) things. Each distinct 'thing' is herein termed a concept.
• This 'self' is itself an ever changing envelope consisting of the selected activated concepts that have been made known to it, as well as additional processes that enhance and support this. An example of enhancement and support processes might include a mechanism that pursue the consequences of different decision point results to the most likely outcomes, providing each with an associated value of desirability. This would provide for directed thinking and mental dry run exercises. An additional example is a mechanism that provides for the explicit self-identification of sensations and output from previous cogitation with itself, one that links all of the so-called conscious activities of the 'self' to the self. That would provide for the belief of agency that the conscious mind has. A third example is a mechanism that binds simultaneous inputs into what this process that is consciousness identifies as constituent parts of a whole.
• Individual concepts are almost never activated alone, but more often in groups of loosely linked concepts. Consciousness is realized in this model as an ongoing selective awareness (noticing) of groups of these 'activated concepts' by the almost isolated process representing consciousness. A particular activated complex of these collections of concepts could also be termed a 'thought'. There has to be a mechanism that is the gatekeeper of which of these 'thoughts' will actually enter into consciousness, that is which of these collections of linked concepts is allowed to be made available to, to be acted on and to be processed with with the subset of neural processes that is consciousness.
• Concepts (each can be defined as a specific collection of semi-permanent weighted neural connections to other concepts) are the content of conscious thought.
• Particular conscious thoughts correspond to different loose combinations of selected activated concepts. Concepts are activated and linked to previously linked groups, whose aggregate composition drifts over time as well as through sensory input and feed forward re-entry of previous activations (resonance). Single concepts normally would not be enough to define a particular thought.
• Simple activation of a group of concepts does not imply that this group automatically enters consciousness awareness. It only enters into awareness by means of its inclusion into a contiguous and changing envelope of previously selected activated concepts.
• The selection of which activated groups are to enter consciousness is itself an unconscious process. How to differentiate complexes of activated concepts worth being selected from the rest of the fire hose of new input sensations and re-entrant previous products of cogitation (both conscious and unconscious) is my biggest implementation problem right now.
• Consciousness depends on a separation of activated concepts from the act of selecting them. This act of perception (i.e. simply detecting something) is itself a concept, fully capable of also being included in these ever changing groups of activated concepts. The activation of this 'perception' concept tied with simultaneous activation of some particular group of concepts is a prerequisite to that thought's entry into consciousness, and should be a neural (or quasi-neural) correlate of consciousness.
• The selection or filter criteria used is important! I am in the process of trying to figure this out.

Consciousness Fundamentally is Process Happening in the Brain

The experience of consciousness resides ultimately in rapidly changing patterns of neuronal activation. Automatic mechanisms heavily pre-process every little sensory input and automatically generated subconscious 'thoughts' about them, and then select which ones actually "enter consciousness". These processes have access to everything our conscious minds do since they are one of the mechanisms our conscious minds. The brain also has numerous mechanisms that automatically look for and make matches to (and find contrasts from) what has been experienced and/or understood previously - we can't help this, we are just programmed for it to happen continuously and unconsciously. For example, visual areas in anaesthetist cats are still stimulated by neural signals coming down the optic nerves. The cat is not conscious of this because other downstream pathways are interrupted, but not the raw inputs. The highest, most abstract of these patterns of activation have access to sensory experience, to needs, to drives and emotions, to memory, to past lessons learned, to what something 'feels' like, and to everything else that makes up conscious experience. The "gut feel" of some idea of a sensation is the same as the real gut feel of that sensation since is the same neuronal cells being activated for both. This (however poorly described) is what I feel consciousness ultimately is…the ultimate convergence and the ultimate abstraction each of us is capable of.

The Sense of 'Self'

We have to be careful here, as I am using the word perception in a way that should not be overloaded with qualia, but instead is meant here as a more mechanical notice or detection of something.

It might not be necessary to crisply define consciousness at all. If we say consciousness is perception from a first person viewpoint, then as long as that viewpoint is exhibited and if the thing being tested for consciousness perceives itself as being the agent of that viewpoint, we will have demonstrated consciousness.

A continuous and self-navigable continuity of perceived sensory input contributes the the creation of the idea of a discrete consciousness existing across a span of time. One perception smoothly follows another in mostly predictable ways. The conscious thing may come to indirectly understand that stuff happens to other conscious agents too and that it can't directly perceive those inputs in the same way that it can for some of its own inputs. That lifelong separation confirms the rift between it and everything else and therefore continually reinforces its concept of self as distinct from non-self.

When you look at a gelatinous mound of wrinkled tissue freshly harvested from inside of someone's head, you can't help but think that this real-world piece of flesh and blood couldn't possibly generate the sublime thoughts and experiences that make up our lives. However, when you look at it from inside, imagining its neurons passing millions of chemical messages between themselves, circuits from the lowest to the most abstract levels automatically doing just what they were set up to do (be it edge detection, pattern recognition, best outcome determination or whatever), external inputs coming in and internally generated and heavily modified inputs going out, it is plain to see that something approaching the functioning level of a super-sophisticated robot or automaton must be happening. My contention is that with just the addition of a set of self referencing and self modifying circuits over top of this, and from the point of view of the whole system which makes up that brain and only that viewpoint (it can never intimately have knowledge about an others ' viewpoint) that brain can be considered to be functionally self aware. We are ultimately all isolated from one another, just like that dissected brain is, and our opinion of an others' state of consciousness is only based on the behaviour and reactions to things exhibited by these organisms. Our brain assembles its own experience, and sophisticated brains can apparently assemble awareness and ultimately consciousness itself.

If consciousness is simply connected threads of neural events corresponding mostly to current sensory inputs and overlaid with the rich collection of re-entrant outputs of previously processed neural events, then some AI consciousness simulations already have that (minimally). There must be more. Taking the example of 'twilight' sleep (procedural sedation or conscious anaesthesia) in which a person is apparently awake and cooperative during a medical procedure but has no conscious recollection of it afterwards, perhaps we are always instantaneously conscious (by means of a lot of concurrent 'threads' of neural activity) of multiple things at any one time, but it is only the few threads which are laid down in short term memory that we are set up to think we are actually conscious of. The term 'laid down in short term memory' here must include not only the simple retention of this new information, but also its incorporation or connection into the part of long-term storage that we regard as our continuing consciousness. Part of this incorporation must happen at the time we experience things (or more properly as we process these experiences), and a more substantial incorporation could occur during sleep (at last, a real function for sleep instead of the ill defined 'restoration' it provides!). I suppose taken to the extreme, we are only conscious in conscious hindsight.

But I don't think that self awareness is identical with being conscious. The thing that is conscious can be aware of being conscious, but there must be a conscious thing in order to do the 'aware-ing'. Self awareness is just a special case of generalized consciousness - just looking inward.

Concepts

External events enter through our sense organs and are generally relayed to topographic, retinotopic, tonotopic, or somatotopic maps on the cerebral cortex. Coincident activations are recognized as such by coincidence detectors and dynamically and transiently linked together (in different brain regions - the so-called association areas) and can be dynamically clustered into discrete 'things' that can be further manipulated as a unit by the higher convergence areas. With repeated referencing of the same transiently linked neurons, the dynamic relationships to other 'things' can crystallize into more permanent, 'hardwired' representations that I call concepts here. These can be thought of as useful handles to often invoked clusters of activated neural patterns and which are used by the brain to represent things, anything and everything actually. All of the components of our thoughts - all of the nouns, verbs, adjectives, adverbs, phrases and syntax that we use…even the ill defined feelings, apprehensions, hunches etc. I have a hunch that these 'things', these concepts may be embodied in the brain in the form of the cortical columns, although the theory presented here does not hinge on this. What is important is that inborn mechanisms within the brain automatically organize similar and repeating patterns into these discrete semi-permanent logical (at least, if not physical) functional clusters. This is termed self-organization.

When you see a finger, hear the word finger, think of a finger, feel through your finger, bang your finger, move your finger etc. the concept 'finger' is invoked. It has a physical place (albeit probably redundantly stored and distributed all around) in the brain. When referenced, this virtual location is 'activated' in that it's component neurons actively fire off action potentials and spread this activation to whatever they are wired up to, at whatever strengths they are wired up with. When not referenced, there is no spiking of action potential generation and thoughts cannot be connected in any way to this concept. The referencing stimulus can be a result of external sensory input, or the re-entrant result of prior neural processing which equates to thinking.

Fingers are part of a hand and fist. They are similar to villa of cells, and to fiords on a map, they are used to play piano, to make rude gestures, to work with tools, to caress things etc. When each of these concepts are activated, the concept of finger is also at least stimulated, if not also activated. The relationship (connectivity) between one concept and others is really what defines the concept and distinguishes it from the rest.

Concepts would play heavily into the idea of archetypes. For example, everyone has an idea of what a chair is. It might take any number of forms, but - mostly based on if it is possible for something to sit in it, most of us would agree that some object is or is not a chair. The idea of a generalized 'chair' thing that can be manipulated and related to other things (including numerous instances of particular chairs) could be termed an archetype and would have a very similar concept associated with it in different minds. Each instance of this concept would be organized in that its detailed relationships with other concepts within that mind are implemented differently - as a result of the different environmental and timing variations that the minds of individuals undergo as they learn about and repeatedly deal with real chairs.

Since concepts are fundamentally only nodes of differing collections of association to other nodes (differing in specific targets and connection strengths), there need not be specialized structures in the brain to hold each kind of them. Once you can organize input stimulations into one concept, you can keep anything in mind.

There is one last thing about my concept of 'concepts', perhaps the most important. There is nothing special about a concept in which the underlying neurons are stimulating their targets (activated, as I say in other parts of this discussion). This happens unconsciously all the time. But only activated concepts can be admitted into conscious thought. And there is absolutely no difference between an activated concept which has entered into consciousness and one that has not. The difference lies in a gating mechanism which either lets it join into conscious awareness, or not.

In the sense that I use it here, concepts truly are the stuff of thought.

Thoughts

stuff

Conscious Awareness

stuff

Gating Consciousness or Determining Which Content will Enter into Consciousness

See Elaboration. Consciousness is a process that happens within the connectionist system that is the brain, differing from unconscious processes. How? One way is that it has a point of view. That view is always with respect to an 'I' - a 'self'. Automatic (and for that matter unconscious) mechanisms organize linked inputs into temporary circuits and if often reused physically links neurons into groups or units (cortical columns?) representing concepts, as needed. In this theory a concept is something that relates to small collections of other concepts with unique patterns of weighted connectivity, and is an internal representation of something (nouns, verbs, adjectives) in that it carries meaning when put in context with other concepts. When related concepts are stimulated either by the outside world or internal cogitaton, they activate whatever they are connected to in what I will call a base network.¹⁷ These connections can be dynamically created and maintained in addition to later physically instantiated through biological changes. Spreading activation of these clusters of concepts in a richly connected network, triggered from both external and internal stimuli, will converge on whatever is most in common with the bulk of the input activation stimuli trajectories. In my theory the concepts at these converging nodes then activate additional network(s) that are superimposed on the original base network.¹⁸ These superimposed networks can themselves potentiate or activate nodes in the original base network (i.e. can be re-entrant), which can again spread back to one of the superimposed networks. Re-entrant stimulation can explain how a thought might be held in mind, how goals are maintained when sidetracked by subsidiary goals, how time is kept and estimate internally, how to anticipate and plan, etc. Most importantly, some of these overlapping networks are cycled through a couple of (for want of a better word) 'trial' iterations, with each network weighted towards one or another of the main already-activated concepts of the original network. The each one of these 'trials' would result in goal-concepts within that same network being activated to varying degrees. The superimposed network that activates a goal in its network the most compared to the rest of the goals activated in the other networks is then selected to reentrantly stimulate the base network - for real this time, not as part of a trial mental exercise. This is how the next thought to concentrate on is selected and also how long term planning is done. The keys are 1) automatic discovery of i.e. self organizing concept nodes¹⁹ and²⁰, 2) automatic maintenance of connection strengths to other nodes based on usage pattern reinforcement and 3) multiple overlapping levels of networks that are hooked up both horizontally {between concepts) and more sparsely vertically (overlapping networks containing the specific concept as a node).

Conscious perception appears to depend on the output from heavy processing of input receptors, where the associated multiple features and aspects of a particular experience are identified and represented in a distributed manner across many parts of the brain. Perception is not the direct result of this processing but it depends on this background processing.

Consciousness is primarily distinguished from other mental activity by the absence of unconsciousness (circular I know, but it is what it is). This theory proposes that there is a physical distinction between the two - in the form of a distinct network or networks for consciousness which might even overlay the same cortical columns (used here as the fundamental unit of mental concepts) which is implemented in the continuously active but fragmented unconscious network. Activation of any specific concept by unconscious processes can cause other connected nodes (columns) on the unconscious network to be stimulated and possibly to become activated too, without necessarily causing the corresponding concept of the conscious network to activate. What activates the conscious version of any particular concept is specific mechanisms designed to focus conscious attention onto subsets of these unconsciously activated concepts. By 'attention' I simply mean the concept is available for processing within this conscious network. The simplest mechanisms would reside within the same cortical column and might just recognize that there has been a change. More complex mechanisms would involve value functions, and a neural means of assessing the best overall best choice.

My answer to the perennial question of "But what is the thing that is doing the perceiving?", the first person viewpoint if you will, is that it is physically implemented in this semi-autonomous and distinct but overlapping network or networks. This is not to say there is a conscious network watching the unconscious brain circuits. That just pushes the conscious entity one step to the left without really addressing the question. What is being presented is saying that conscious perception may be best implemented by having two (physically or at least logically or just hierarchically) different networks. The whole system 'perceives' things (in the sense that it notices change away from a background). Part of that incoming set of stimuli is composed of outputs from previously processed concept activations in the network, - re-entrant inputs. In this way the same concept doesn't need to be represented in a number of places at the same time (i.e. in the neural seat of consciousness which might be termed working memory - if such a place exists, as well as in one or more unconscious sub-systems). Instead the concepts that are capable of entering into consciousness are simply double wired up with their peers. One category of peer connections facilitates the automatic unconscious momentary activations that happen all the time (activation of the beer concept subsequently activates liquid, bubble, glass, hops smell, distinctive taste, party and so on even if they are not also directly activated by input sensations; but does not usually activate the concepts of grandmother, diapers, house work, surgery, etc.) This set of transiently activated connections between these same concepts would, provided the proper gating mechanism permits it, instantiate a hierarchically higher level of activations of this same subset of concepts, one that represents the instantaneous content of consciousness. Thus, a concept could be activated at a low level but not at the highest, conscious level, and vice versa. Note that an individual concept can be pre- or hard-wired to specific other concepts, and can also be coincidentally and temporarily activated in concert with a set of other concepts. Activation caused by either mechanism (hard wired weighted connections, or temporal coincidence) is probably the cause of the so called gamma synchronization that is supposed to be associated with consciousness. I believe that this gamma synchronization is a side effect of, rather then essential for consciousness.

A recent paper by Langsjo et al. [18] studied the activation of numerous brain regions as volunteers were rendered unconscious by the use of various anaesthetic agents, allowed to return to consciousness and then rendered unconscious again. They note that the essential neural core of consciousness probably resides not in the neocortex but in deeper, phylogenetically ancient structures of the midbrain and brainstem. This makes sense as in this theory the cortex is more responsible for the content of consciousness, and a groggy post-anaesthetic individual does not think deep thoughts. The midbrain and brainstem are well positioned to function as control switches, gating which of the multitude of potential sensory and cortical content of conscious thought actually gets admitted into consciousness. Persons who are minimally consciousness would be expected to exhibit reliable activation of only this central gate keeping region, and much more variable activation of whatever the sensory or cortical structures were that embodied the particulars of their final or initial conscious impressions. Experiments that study subjects as they dance around the line that separates awareness from oblivion are a promising avenue to determining the neural correlates of consciousness. But when a site or site is unequivocally determined to be a NCC, so what? We still need to understand the underlying gating mechanisms.

The Langsjo et al. conclusions are troubling in that this means that many more creatures than is commonly agreed upon today should probably be considered to posses some form of consciousness. The here-to-fore record of humane treatment of such creatures has not been good.

Unfortunately this doesn't yet explain the apparent continuity of consciousness, or the apparent executive ability of consciousness to direct thinking.

Input into the decision to admit something into consciousness is best performed with a fully attributed input…that is all of the emotional, drive, goal, novel or not information should be available at the time of the decision for this process to work efficiently. A proper architecture would have as many of these extra bits of information necessary to make the decision already assembled at the time a decision is needed.

Closing the loop - activated nodes of this proposed consciousness network would have the ability to stimulate peer nodes but also nodes of the underlying unconscious network. The subsequent input stimulation would come either from consciousness network peer node activation (i.e. normal spreading activation), or as the result of some fragmented collection of nodes in the unconscious network suddenly being made salient, that is being able to trigger input stimulations in the conscious network.

Characteristics:

• Inputs:
  • sensory input concepts (with connected emotional connotations?)(with novelty indicator)
  • re-entrant input concepts (an existing train of thought)
• references:
  • Goals (general goals…is this actually re-entrant input?)
  • Drives
• Evaluation
  • does input concept help to satisfy particular goals? If so, how important is this?
  • does input concept help to satisfy particular drives? If so, how much is this worth?
  • is this new? If so, is it important?
  • is this laden with emotional significance? If so, can it be ignored?

Elaboration

Background processes happen automatically and continuously (even when asleep or lightly anaesthetized). In the pathways to visual awareness for example cone and rod cells in the retina fire in response to light stimulus, edges are recognized as such and reinforced, similar masses of colour and texture are recognized and linked together into neural representations of larger forms, foreground and background features are isolated, partly by the synchronized movement of groups of these features (in the human brain this is might be a function performed by the thalamus), component features are recognized as probably belonging to higher order things (e.g. eye is part of a face), missing information is tentatively filled in, emotional significance is assigned (in the human brain this probably involves the amygdala), immediate and remote past experiences of similar sights are matched and are made available (in the human brain this probably involves the hippocampus), projections and expectations are made (in the human brain somewhere in the neocortex)…

All of the above is accomplished by multi-use networks of cells that (in effect) calculate the probability that a current input is similar to what a previous input was determined to be (a pair of eyes emerging out a shadow are more likely to be part of a single face then to be nailed to a wall; a group of edges moving from the left to the right in sync are more likely to be physically connected then they are to be multiple objects coincidentally moving in the same direction and speed). The probability impacts which pathways are most activated in higher levels of the cortical hierarchy. In this way very granular inputs can be transformed into activation of specific high-order summaries or conclusions (i.e. retinal cells activated by 475 nm wavelength light is automatically identified as a blue sky, or as a bedroom wall, or as a sports car…depending on the context and the specific other sensations contributing.).

These multi-use hierarchical networks should generate reproducible results on different occasions in so far as the inputs can be reproduced, because they are very like 'hard wired' reflexes - that is they are the product of the physical state of synaptic connections between axons and dendrites. This physical state can change slowly over time as a result of ongoing learning by means of growth of the axon or dendrites, and also as a result of physical synapse changes that potentiate or inhibit the conduction of nervous impulses across the gaps between neurons, but in short time frames they can be considered to be as reliable and repeatable as reflexes. Note. The slow calcium waves that are known to be propagated by the glia cells may impact this conclusion and this repeat-ability. Any understanding, conclusion, identification or awareness resulting from the automatic exercise of such pathways through this hierarchy on a particular occasion should be the same as what would emerge on a different occasion - say after a sleep interruption. This kind of perception does not depend on transient, situational or coincident brain processes. The output of these processes (the automatic recognition, classification, etc.) is not consciousness, but the stuff of consciousness, the thoughts that consciousness works with.

Consciousness probably depends on this predictable, automatic pre-processing of incoming sensations, but also includes the transient characteristics that are specific to each point in time (the factors that make each occurrence of some set of inputs unique; the things that are not yet remembered or even recognized!). It is fundamentally based on varying potential differences across neural membranes which either potentiate or inhibit neural action potentials, like every other brain process. These excitations or inhibitions, the result of just happened experiences, are held in this temporary way until physical changes have had a chance to stabilize the experience into a longer lasting memory - physical changes such as dendrite or axon growth, pre or post-synaptic changes, protein synthesis, ion channel protein changes etc. These transient experiences are at risk of being victims (i.e. lost) of a sudden 'reset or reboot' caused by a blow to the head, faint, anaesthesia, or even by being distracted by more urgent matters until they become more permanent memories. Contrast this with the long term stable automatic processes mentioned above. A period of unconsciousness or distraction should not affect these more permanent memories (i.e. memories not liable to being lost by these events). These too are fundamentally based on varying potential differences across neural membranes which either potentiate or inhibit action potentials. One additional difference is that these action potential predispositions, in response to a new stimulus, should be repeatable and more or less stereotyped, whereas the transient 'stuff of thought' responses is variable due to being exquisitely sensitive to events of the moment. Again, it is the disposition of the membrane potentials of the network of nerves that brings these two different mechanisms together in a single interface. That interface is very close to the physical manifestation of consciousness, but it is not quite all of consciousness. That interface is actually the same set of nerves. Real experience on the one hand produces specific patterns of activation. Historical experience (long term memory) on the other hand is what is superimposed on these patterns, skewing the raw input, trying to more faithfully correlate with the most probable cause. The same neurons are stimulated. It is the strength of the resulting activation pattern that is influenced by prior experience and automatic matching with prior experiences is attempted. Only in the case where a good match is not made (i.e. a large error of prediction) that alternate, more unlikely pattern matching, or even new learning happens.

Imagine brain areas dedicated to describing the degree of various emotions (for example Panksepp's[23] seven mammalian emotions: SEEKING, RAGE, FEAR, LUST, CARE, PANIC/GRIEF and PLAY). Neurons (let’s say but not really believe) on one side of one of these areas represent low levels of an emotion, neurons in the centre represent a moderate degree of emotional involvement or attachment, and those on the other side represent intensely high levels of this emotion. Imagine now that each experience which is stored in a distributed way around the brain as described above is linked with a particular cluster of cells in these emotional-degree brain areas, in such a way as to activate the area every time the corresponding concept-containing neurons are activated (in the human brain the amygdala is well placed - between the incoming signals climbing up through the thalamus to their cortical destinations, and signals coming down from the cortex). Providing that this link is made around the time of memory formation (either the transient or the more permanent type of memory) then we have a mechanism to evaluate the relative importance of each concept (idea/thought/memory/experience etc.) A process that evaluates all of the current experiences in terms of their degree of emotional importance (or ranking) for each emotional axis can then happen, and even select (by excitation of the winning concepts and inhibition of the others) an emotionally important subset for further consideration (processing). Apparently the medial prefrontal cortex²¹ is involved in this. It could serve as a handy focusing or attention selecting mechanism. The medial prefrontal cortex is part of something called the "default network"²² which is otherwise active except when the conscious brain is focused on a task, and is now believed to be involved in undirected daydreaming, inner dialogues, and possibly categorization and consolidation of short term memory into more permanent forms. Note that the concept of ‘emotion’ as used here is an attention or importance ranking mechanism, and is not the associated subjective feeling. These feelings may turn out to be something as simple as different connections to previous somatic experiences being linked for each. When not daydreaming however, the default network is not active, and I speculate that the thalamus (now appearing in its other non-signal modifying, bottleneck role) acts as an attention selecting device, using the degree of unexpectedness (novelty) and also the degree or urgency of emotional significance which the amygdala previously added. I have subsequently stumbled on a series of papers by Basilis Zikopoulos and Helen Barbasthat indicates that this might actually be what happens in real life²³).
O.K. We have lots of specific detail and/or generalized concepts or ideas with which to work, and we also have both transient and long term storage memory mechanisms to keep them in. We also have context sensitive retrieval mechanisms to activate only the most relevant stored information. We have evaluation mechanisms that rank the relative importance and make only the most salient available for the limited resources of working memory (where conscious awareness and focus has to happen). In the human brain working or short-term memory is probably a joint function of the medial (anterior cingulate), lateral and ventral (orbital) pre-frontal cortex and amygdala²⁴). Throw in inputs and an input focusing mechanism (the thalamus), a decision/relative ranking area - the striatum, as well as one of the major connection 'hubs' of the brain - the precuneus²⁵ necessary for self referential processing and you pretty well have all the pieces necessary for consciousness. How does all this make some one actually conscious? Here comes the problem of qualia, and the physical feeling of being conscious, and also the purpose or function of the very ability to be conscious, and the problems of determining exactly what consciousness is. I think the end result will be that consciousness turns out to be nothing more then the highest level of the selected for attention neural process just described. A high order mechanism that receives pre-processed unconscious but selected inputs which can also influence the outcome of some of what is going is a very good start. As long as it 'thinks' that it itself is conscious, and it exhibits conscious seeming behaviour to outside observers, it is practically conscious and probably is exactly conscious too. Explaining how each of the characteristics of consciousness listed above are generated is next.

Working Notes to Myself

Note - some more random ones

Assume consciousness has something to do with the concepts automatically selected for it to dwell upon. The system needs:

• Concepts and pre-existing weightings of connections between these concepts
• A starting condition in which some concept already has attention in a network that represents what the system is supposed to be conscious of
• A stream of simultaneously arriving concepts at an 'unconscious' network, sort of a staging or proving area
• A mechanism to evaluate each incoming and existing candidate for highest importance, significance, or potential value. This is done with:
  • Defined primary goals or objectives or preferences
  • A biasing towards concepts that have just previously received attention, in as to be able to hold a thought
  • A trial evaluation of each incoming concept against the goals
  • Selection of the winning concept, which in this case means it now starts to influence other concepts in the 'conscious' network
• If it is to be replaced, the concept previously receiving attention is ‘de-selected’ (no longer affect other concepts in the 'conscious' network) and perhaps suppressed
• The winning replacement concept is ‘selected’ (in that it is now allowed to influence other concepts in the 'conscious' network) and enhanced, that is strengthened (it's important, right?)
• Other, unconscious network, candidate concepts are slightly suppressed
• The 'conscious network' system cogitates a bit on the winning concept
• Rinse and repeat

This represents a brain in its default mode, not concentrating on much of anything. The mind just wanders. The input stimuli represent potential sensations that could terminate this default mode and replace it with attention to a stimulus.

When something has the attention of consciousness, these input stimuli still fight to be selected. The already selected concepts however are separately also generating inputs into conscious network's own the evaluation mechanisms.

Think about feedback (both top down and bottom up). Think about connecting concepts with goal fulfillment.

Note - 'Temporal thickening'

Consciousness is about the now. It quickly fades into memories of things just past. Ideas and sensations enter it and are there for a brief but non-zero length of time. I just read an idea from Nicholas Humphrey²⁶ to the effect that an internal feedback loop (or "re-entrant circuits, resulting from some cerebral output) if timed just right to feed back in to the next moments neural input streams could create a sort of self resonating/reinforcing system. This feedback from previous outputs combined with new inputs results in what Humphrey terms 'temporal thickening' which I interpret as the mechanism that stretches out the moment of whatever 'consciousness' is, from infinitesimal to somewhat longer. This is an interesting mechanism, but I think not essential to consciousness. However, perhaps this self-resonance is the source of the reasonably well documented ~40 Hz so called gamma brain waves - the still unexplained near synchronous firing of neurons at this frequency which is often seen in the EEG and quite restricted to states of conscious awareness. I can see it contributing to maintaining a higher potentiation of certain specific regions (concepts) that are already activated, and also to building the long term potentiation of synaptic connections (i.e. laying down a permanent declarative memory) through the mechanism of "neurons that fire together wire together" as voiced by Donald Hebb²⁷.

Rather then defining the content of consciousness as being those activated concepts that are simultaneously (i.e. gamma synchronoulsy) activated, we might say it is that and also those that are also activated in a distinct network. We can even alter this and say that consciousness represents those activated concepts in the distinct network who's outputs re-enter that network in a positive feedback loop, as well as into the original base network, in a longer feedback loop.

Nota bena

Inter-nodal connection strengths represent a kind of running average of of inputs that have come before. If novel inputs come in and stimulate two overlapping networks, one that maps to the pure inputs, and another that maps to those same input activation patterns but also is modified by allowing propagation of the resulting spreading activation (using the connection strengths) to stimulate more nodes, then we have an ideal architecture in which each node (playing perhaps different roles in both networks) can most easily identify differences between the current incoming stimulations and previous inputs. The more different the same node is acting in the input network and in the spreading activation network, the more salient and relevant it should be for consciousness. These differences can be propagated|promoted|communicated to an independent network layer, which might represent or instantiate attention and play a fundamental part in the mechanism of consciousness.²⁸

The representation of things, such as individual faces, in our brains is probably done by storing all characteristics (dimensions such as roundness, feature shape, position, spacing and proportion, texture, colour, etc.) of the average face in great detail into a perceptual whole, or Gestalt, and then storing each individual's face as a set of distances from that Gestalt, distances across the various dimensions of face-space. A concept that relates the relatively unchanging average of a class, as modified by sets of these dimensional distances is much more efficient at storage and retrieval of class instances then is the storing in full of every characteristic of each instance of a class. This frees up a whole lot of neural processing power, and is almost certainly a mechanism used in life, although I am not aware of an experimentally observed basis for this conclusion yet. This kind of knowledge representation is interesting but not directly related to actual consciousness. Class instances can still be at least conceptually referred to by the term 'concepts' as is done here.

I envision spreading activation would work on a small world type network in which most concept nodes are not neighbours of one another, but in which most nodes can be reached from every other by a small number of hops or steps. Given my vision of multiple networks superimposed on these nodes, each network can have different characteristics. One might be a network that does not restrict activation of concepts unless the total activated number goes over some threshold (a baseline reference network). Another might be a network that favours spreading activation from newly activated concepts over that of continuously activated ones (a novelty detection and enhancing network). Another might favour the continuing activation of those concepts that are already activated (this network preferentially propagates already long lived concepts). Still another might favour the activation of concepts only if there is a strong association with some category of concepts (a network which favours activation of concepts with strong emotional associations/ a network which favours activation of concepts strongly associated with concept-goals). These different characteristics give better control on the required gating decision to admit each activated concept into consciousness or not, while providing mechanisms for the simpler autonomous processes to fulfill the duties that the brain was first called upon by evolution to perform. Consciousness is most likely composed exclusively of these already ancient functional components and mechanisms which are already unconsciously doing useful things. The only difference is that there is more extensive integration in the conscious network then there is in the myriad of autonomous unconscious processes? This difference needs to be extremely simple to implement and only subtly different from unconscious processing…a difference of degree but not of kind, perhaps? Or is it one of a completely novel architecture or connectivity, or functional mechanism? I don't think so.

Notes on a new modified "conscious pilot" model

Hameroff [12] gives a nice summary of the current state of knowledge about the Neural Correlates of Consciousness field, covering most of what is also found here. Where he and I diverge is in his description of the mechanisms of cortical synchrony, especially his assigning of an extensive role for gap junctions, and also in his speculations about solutions for the hard problem.

His extensive role for gap junctions would I think overly couple huge areas of the brain, tying constituent neurons together in such a way as to make these syncytia useless for normal neural net function, if I understand his proposal. The need to rapidly coordinate all of the gap junctions necessary to shift to a different syncytial configuration also seems be unlikely to be easily satisfied. I suppose it is possible that only particular second messengers and the like are allowed to mingle, perhaps forming multi-cellular gradients, but certainly not a free flowing river of ions between cells. This is a likely mechanism for generating the calcium wave phenomena which traverses the surface of the brain, especially since astrocytes seem to be strongly electrical coupled by gap junctions. Such waves can locally (or more extensively) influence synaptic neuronal firing and may play a role in the laying down of memories from the transient, almost standing wave-like patterns of neuronal activation and conscious attention.

Hameroff's speculations about the agency capability of gamma-synchronized dendritic webs strikes me as being bang on. Where we differ is when he references "others" speculations that the involvement of "sub-neruonal functions" somehow results in consciousness. These functions include "quantum computations in microtubules", a hazily defined role of electomagnetic fields, "molecular reaction diffusion patterns", quantum entanglement and quantum coherence. These poorly understood phenomena regularly are dredged up as candidates to explain the poorly understood phenomena of consciousness. Somehow the two 'poorly's are supposed to cancel each other out, but I can't accept this as an explanation. As previously noted, here I instead rely on the brain's seemingly automatic tendency to attribute causes to effects, which is heavily based on previously observed correlation. The agency of Hameroff's transiently activated but nearly continuously operating network of neurons, his 'conscious pilot' could be attributed by this automatic tendency to a concept of 'self', constructed/invented/self-organized for just that purpose. The 'self' may be an invented or created-for-the-purpose placeholder, of sorts.

Hameroff's [12] conscious pilot theory involves a mobile spatiotemporal envelope, composed of a gamma-synchronized and dynamically defined dendritic web, that performs collective integration and volitional choices. In my view groups of neurons are activated by external or re-entrant internal inputs, bound together by temporalcoincident firing, and the various concepts' strength of activation (synaptic potentiation) modified by relative value and goal attributions. The infamous gamma frequency correlation with consciousness seems to me to be a side effect of massive numbers of simultaneously firing concepts, not the driver of this. The real question revolves around how this envelope integrates, and make deliberate choices.

Neural inputs can activate concepts. Activated concepts are 'hard wired' to relatively small subsets of peer concepts. The more related these peers are, the more they can drive each other to activate too (see my concept of qualia in this document). But inputs might regularly present the brain with novel patterns of concept activation. These unrelated concepts, when attention and consciousness is upon them, need to remain bound together by some mechanism that results in their coincidental activation. If we assume that each concept is not 'hard wired' to every other concept (the brain would be composed of 99.999% wasteful white matter!) and instead is reciprocally hardwired with weighted connectivity to one or a few central nuclei (hubs), then it would be easy for that (or those) nuclei to maintain and slowly modify the coincidental activations. This (or these few) central structure(s) is/are nicely positioned to be the site in which relative value and goal modifiers are applied to the activated concepts.

The above construct gives us a mechanism to select the next concepts (thoughts) that are based on the relative interest/worth/novelty/value of each current concept, as well as furthering the most compatible drives/wants/needs/goals.

Note - behavioural expectations/interface specs?

This section will be incorporated into the main body once it is half way completed

Assumptions:

• Inputs can come from without or within. Both should be treated the same way
• Inputs can be valued relative to other inputs
• Inputs can dynamically converge on internal representations
• Internal representations can be valued relative to other representations
• Highest value internal representations can turn into long-lasting steady state internal representations

When Given:

• No, or sparse inputs
  • pay attention to (generate diverging representation through free association) inputs originating from long-lasting internal representations

• A stream of similar inputs
  • Notice (pay attention to) input stream particulars
  • Arrive at some internal representation about them

• A long relatively unchanging stream of similar inputs (add in a memory component)
  • Suppress attention previously paid to the input stream
  • Turn attention to other things

• Novel inputs after a long relatively unchanging series of inputs (add in a difference detector)
  • Immediately detect the change
  • Start to pay attention to novel inputs
  • Suppress attention previously paid to the 'other things'

• Ambiguous or conflicting inputs (deal with ambiguity)
  • Arrive at a steady state of multiple possibilities
  • Continuously attempt to resolve using subsequent input
  • Converge onto a single winning representation when able, suppressing the other now discredited possibilities

• Multiple inputs (add value evaluation)
  • Arrive at a steady state of multiple convergences
  • Select the most salient and pay attention to it (?how to define this?)

• Multiple inputs, one or more sets change after a while (the whole shebang)
  • Arrive at a steady state of multiple convergences
  • Select the most salient and pay attention to it
  • Notice change
  • Select the most salient and pay attention to it

Note - the Architecture of a Conscious Implementation

A brain comes with feedback loops, self organization of inputs, lots of convergent abstracting of this information and various levels of feedback control. We then have an unconscious zombie-like implementation, like so many other AI programs and/or robots. Next we need a gating mechanism that controls what subset of zombie pre-proceesed inputs are allowed entry into consciousness for further processing, and a relatively virgin, isolated substrate that does the same sort of thing as the zombie brain. This one has (must have?) re-entrant feedback such that its own output becomes one of its inputs. This latter bit is a material difference from the zombie version (perhaps?) or at least much more implemented then in the zombie version. More then one proofing region would allow multiple versions of potential thoughts and actions to be evaluated at the same time, with the 'best' one selected as the source of further processing or action signals and the others simply discarded. Alternatively each source potential thought and action would come with its own evaluation of how successful it will be, allowing a much simpler and quicker evaluation. There should be a comparison between action results and the selected source anticipated results, and an error identification (difference) and correction mechanism (learning?) applied.

Local valuation state transfer decision points:
- inactive to newly activated; and activated to newly inactive
- activation coincident with non-peer group concepts activations (novel combinations of activated concepts); and corresponding inactivation
- inactive to activated with strong direct or secondary connections to concepts that are emotionally laden; and corresponding inactivation

This really just describes an autonomous controller, with a separated executive. The magic happens in the interaction of the executive with the autonomous controller and within the executive itself. The executive will have selective inputs from the controller and also its own output as inputs. The executive's own internal workings, components and subsystems will be opaque to the executive however.

Note - consciousness is built upon the same foundation as unconscious neural processing is

The exact things that happen in a neural network that embodies conscious awareness, must have been happening in the rest of the unconscious networks that have existed since the first nervous system enlargement millennia ago. Different implementations of the layers of abstraction will exist, as will gating, but the same fundamental mechanisms must be used in purely unconscious reflex-like processing as do in sophisticated conscious thought.

NB Nota Bene

The function of consciousness is likely to maintain an input-to-timeline record of a wake cycle's worth of awareness of things important enough to keep in a record of this type. During a sleep cycle the discrepancy between sets of connection strengths linking concepts in this record at various points in time, and between connection strengths linking the concepts maintained in long-term (general) memory is identified, and the long term representation is adjusted slightly with this new learning.

Where are the 'Three Laws'?

Asimov's Three Laws of Robotics are nothing more then a fictional plot device. There is nothing that says that they must be part of every mobile computer (robot), even if they might actually be a good idea to implement!

Draft Specifications for a Minimally Conscious Software System

Below is an unfinished Use Case-like word description of the flow of external events: through input transducers, on into first representation in the software, how these patterns of on/off activation (or of course sub-threshold biasing or anti-biasing) are properly matched up with prior stored representations, are matched up with different biasing or anti-biasing drivers to be used in later rank evaluations, are evaluated and selected for focused attention, are made conscious…..TBC

Why do I think a machine can be made to be conscious when the human brain is arguably the most complicated thing in the universe (well, argued by some people anyway)? The natural neural pre-requisite for consciousness is a much smaller subset of the 100MM neurons and 100BB synapses found in the adult human brain. For one thing fully ½ of the neurons are involved cerebellar function (co-ordinating movement) which has little or nothing to do with consciousness. In addition, consciousness is almost entirely unaffected in those patients who have had their corpus callosii severed, so ½ of the remaining number is redundant for “minimal consciousness”. Much of the remaining quarter is taken up by homeostatic monitoring and controls, proprioception, primary sensory input areas, movement initiation etc. which are not intimately necessary in a demonstration of artificial consciousness. Finally, the whole breadth and depth of human consciousness is not necessary to demonstrate the mere fact of a working artificial consciousness – only a logically consistent and “reasonably complex” subset. A demonstration of the principle only shouldn't require anything close to equivalence to a human brain.

1. Each element of the 'areas' I will discuss next correlates with either individual neurons or with small groups of neurons (cortical columns).
2. External stimuli continuously affect a matrix, array, raster or other temporary data repository region with their granular inputs corresponding closely to environmental fluctuations (from outside the system and from points inside also). Each mode or channel of input updates its own area (a primary visual area, a primary auditory area, etc.) but the specific orientation is not important. It doesn't matter that inputs from adjacent receptors in a visual field could easily be mapped to representations in the primary area that area far apart. The transducer-to-primary-area-element mapping is fixed however; stimulation of a particular receptor transducer will reliably be represented in the primary area in the same way, time and time again. The information in these areas is almost directly correlated with external stimuli and is just as transient and just as 'noisy'. It is a working 'memory', a working area, a workspace for the temporary representation of transient patterns…places in which to temporarily represent the on-off and grouping patterns of external inputs.
3. Each of these primary information input areas are also hardwired to multiple secondary elements, in overlapping paths. The secondary nodes that are activated at any one time are those that link together only those primary area elements that have just been synchronously activated (e.g. all contrasts from the background; all similar colour; all similar sound frequency; etc.). These correlations of input patterns are themselves represented in the form of transient nodes in temporary working areas. These new nodes each correspond to the average of the small groups of raw inputs that they connect to, and the recent history of these raw inputs since activation of each of these new nodes is longer lasting then are the patterns in the primary areas. In this way many simultaneously activated elements caused by an outside stimulus are linked together in multiple more stable representations, integrated over short spaces and over brief times. Different aspects of the input are separately represented in different secondary nodes.
4. Note that some time afterwards, when a long term memory is physically laid down (in effect enhancing the connection strengths between nodes to the currently active higher order nodes) similar subsequent input will activate these same nodes easier then they will other non-involved nodes. In this way responses to similar inputs are reinforced through repeatedly being exercised and potentiated over time. (e.g. an edge; a left edge of a mass; a left edge of a blue mass; a left edge of a blue mass attached to a larger mass. Similar inputs should produce activation of this approximate cluster again.
5. These new temporary nodes are themselves linked together in overlapping and convergent ways, by applying the "what has been recently activated" algorithm, and are themselves written as nodes in yet another different summary/hierarchy/composition representation areas. Along the way some of the rapid fluctuations of the raw inputs are lost/smoothed over since these nodes are even longer lasting and (effectively) summarize or otherwise integrate larger numbers of the primary pattern elements. This process iterates to ? (less then 20, perhaps < 7) more hard coded levels which results in stable, long lasting representations of the initial causes which lie behind the rapidly fluctuating raw inputs (a mass moving across the field of view would be represented as a single activated group of nodes in a top level area instead of as the original series of discrete impulses in the visual input channel.)
6. At some point synchronously activated (and linked together) nodes from across input modes are themselves linked together. In this way cross mode (i.e. vision and hearing) nodes are brought together and represented as input-mode independent patterns of relatively long lasting activation. The end result of this summation has in neurology been simplistically called 'granny cells', since a particular small cluster of such nodes is presumed to be always activated when the concept of grandmother is active - for example.²⁹
7. These stand-alone and self-discovered concepts are still held in transient storage. They need to be matched where it is possible with corresponding concepts held in more permanent storage (the kind of storage which is committed to memory, dropped from focused attention, and then can later be recalled.) This matching must be done on the basis of similarity of hierarchies, context, composition etc. In those cases where a candidate concept is not readily found, a new one needs to be permanently stored and discretely represented.
8. This auto-discovery is not performed from scratch every time. In the first place existing hierarchical pathways to the upper nodes are preferentially biased towards firing with similar subsequent inputs – a mechanism for maintaining focus over short times. This is due to the reinforcement triggered by simultaneous occurrences potentiating background process of Hebbian learning. Similar input patterns will tend to activate similar nodes at each level at different and perhaps widely separated times. Similar secondary activations (i.e. subsequently generated through the activation of a primary pattern) are also potentiated and linked together through this same simultaneous potentiating process.
9. note that even though the above steps in the process were itemized in some detail, none of this alone results in consciousness. All we have is some self-organizing, semi-independent subroutines that work with extremely granular data and which message their newness, novelty, salience and/or their degree of need into a process which selects a few that are to be allowed to do further processing. It is an interesting control system, but do you see any kind of conscious awareness here?
10. which brings us back to what I started with, the need for a solid definition of exactly what constitutes consciousness.
11. .
12. .
13. . need to better place these ideas
14. An attention focusing method that does not require a central hub is in situ activation of persistent complexes of nodes in the multiple levels of the multiple hierarchies, in which each activated complex also weakly activates its logically connected neighbours as well as a hub (perhaps an emotional centre such as the amygdala?) which normally keeps everyone suppressed. The whiniest complex is the one that gets the hub’s attention (positive feedback/potentiation by removal of its inhibition). The centres with the most activated neighbours are the ones that others are most dependent on and thus the ones that are allowed to continue. (i.e. pants on fire causes activation of the following complexes pain, heat, smell, sight, invoked fear, cognitive concern, adrenalin release effects, etc. – vs. background hunger which might only involve activation from cognitive and gut sensory complexes.)
15. The activated complex in the pants on fire scenario would be encouraged to explore (more strongly activate) node-once-removed and more distant connections – presumably ones involving consequences, possible escape actions, “stop drop and roll” memorized sequences, smothering with hands, yelling for help, locating sources of water, etc.).
16. The neediest complex might also be the one with the highest prediction error signal. In the pants on fire scenario all deviations from expected states of pain, heat, sight etc. introduce a strong prediction error signal. In the case of a couch potato watching TV scenario, non-completion of chores generates a planed vs. actual gap (deviation) in a household job completion neural node complex. Related secondary node activations involve plans for accomplishing each job, expected consequences of completing them, expected consequences of non-completion. Each of these also generates this prediction error signal that results from differences between the expected prediction (usually a continuation of steady state) and the theoretical prediction of each alternative.
17. .
18. Note to self - get a head start on implementing some concepts with word association data - something like the [Edinburgh Associative Thesaurus (EAT)], the [University of South Florida Free Association Norms] or perhaps the [WordNet semantic network]
19. Spreading Activation is a great concept, but the devil is in the details of how it is implemented and what termination conditions to use. I think the whole cluster should first be tested for contribution to one or more goals or to satisfying one or more drives. Then the 'best' next thought, next idea attended to, the best next action etc. should be chosen on that basis. How many iterations must this trial or test go to determine a good result? How many can be evaluated given the real-time nature of the need? How many can be tested per iteration and how can non-overlapping subsets be identified for such parallel evaluations? What is the algorithm for determining best (short term payback, longer term gain, more gratification or longer gratification)? Are 'calculations' done on a single unit of normalized values from all goals and drives, and if so how best to normalize the values of all of the possible outcomes?

Design Assumptions and Decisions

These assumptions are being made for a Java based minimally conscious POC demo application that I am concurrently working on, while I develop this thesis. You might ask me - to what end? I would like to be able to start it up and legitimately ask the question - why shouldn't this entity be considered to be truly conscious? The more difficult that question is to answer, the more I will have succeeded.

A single layered network of interconnected concepts that stimulate each other by waves of spreading activation is probably not enough. At the very least you need two networks superimposed on the concept/nodes, one that is the unconscious processes and one that embodies the 'conscious' ones. Careful gate-keeping of the spreading activation across the two networks can cleanly isolate conscious processes from unconscious ones. The isolated conscious processes can be kept "artificially" or purposely clean, that is less affected by the noisy signals that an unconscious (or pre-conscious) network would of necessity have to deal with (it is the real world we are trying to make sense of after all). In my theory of consciousness this gate-keeping is critical.

If the perception is a large mass rapidly moving toward you, your train of conscious thoughts might be different depending on whether the large mass is a lady with a Dim Sum trolley, or if it is a car on the same road you are. In the case of the lady it would be different if she was looking at you (= aware of you) or not (i.e. about to run into you). The same input needs to be evaluated in different ways to find the most appropriate next thoughts that should be surfaced to consciousness.

The ability to plan different things based on the same input knowledge (= activated concept/nodes) means that different combinations of nodes need to be tested in turn for 'goodness' of the plan/next thought - with the one that produces the most valuable result/potential being the one that is selected for continued cogitation. Is this planning and evaluation an unconscious thing, or is it a conscious one? The number of alternatives considered could determine the number of overlapping networks needed if all are stored 'on-line' at on time - probably not. More likely each bunch of related concepts to be tested has a persistent pointer, and the relative value is associated with each pointer. In that case you would only need one overlapping 'test' network which is used in turn to evaluate each nodal combination.

• neural nets are not needed to implement 'Concepts', although a complete from-the-ground-up implementation might choose to use that approach
• the existence of, the frequency of, and the strength of connections between these Concepts are needed. Get from web based standard reference sources such as the Edinburgh Associative Thesaurus (EAT)
• two independent connection networks link and are superimposed on these Concepts - one corresponding to subconscious (i.e. unconscious or pre-conscious) processes, the other corresponding to conscious processes which taken all together are identical to conscious thoughts
• at any point in time multiple inputs can arrive (from the external environment or by internal recursion), which activates one or more of these Concepts which are nodes in the unconscious network. The demo does not attempt to implement this input-to-concept mapping. Right now, a script containing external inputs is executed and concepts are set to 'active' at pre-determined intervals
• spreading activation proceeds in the subconscious network, seeded by each of the newly activated Concepts
• spreading activation ends when one or more associated 'goal' concepts (just a specialized type of Concept) is activated, with concepts reaching activation from multiple pathways (e.g. 'light', 'smoke', 'heat' and 'burn' can all activate 'fire'). Selecting the appropriate boundary condition is crucial to the success of this endeavour
• the activated subconscious concepts closely associated with the highest goal(s) are now considered as being 'selected'
• the conscious network is then stimulated by these concepts selected by the unconscious network
• a wave of spreading activation then similarly courses through the 'conscious' network. This is not identical to the subconscious pattern that triggered it and starts from a much reduced subset of some of the activated subconscious concepts. This subset can/should move in an entirely different direction from the unconscious networks
• feedback of (conscious network) activated concepts can re-enter and bias the next iterations of the unconscious network
• with this arrangement goals and drives can sub-consciously compete for the ability to select closely related concepts for entry into the conscious network
• conscious and subconscious mechanisms to bias the baseline connection strengths, and in effect learn from experience, are possible but will not be implemented at this time in this Java based POC. These bias changes could be accumulated and perhaps used to permanently update the inter-concept connection weightings.
• further conscious and subconscious mechanisms to additionally bias the baseline connection strengths would include sleep/wake cycle simulations, arousal, cyclic drives such as hunger, companionship, reproduction etc.
• later I will add some rudimentary grammar that can differentiate nouns from verbs, and some sort of text based UI to display the free association-like output that this system produces. In the meantime I write instead to the Java Console.
• a sense of "me" as well as a sense of past, present and future time can be added later. The simulation as it stands however would forever live in the now.

"Shit happens" to the system [via activated concepts triggered from a file of concepts to be activate at particular system timing cycle counts]. Some stuff is more valuable and important to the system then other stuff [thanks to the embedded goals]. The stuff that happens is closely associated with all of the characteristics of whatever it is, even if not directly perceived at the time [qualia? activated by the mechanism of spreading activation]. The system's response or reaction to something will very much depend on it's prior state. That prior state is unlikely to be reproducible because of the large number of paths to possible states that can be triggered from the external simulations [perceptions] being fed to the system. Thoughts about whatever happened ['consciously' activated concepts] can trigger other thoughts [also via the mechanism of spreading activation]. These trains of 'conscious thoughts' can influence or bias the non-conscious state of concepts, which in turn will influence which subsequent thoughts will next be brought to 'consciousness'. The system will only be able to report the things it is 'conscious' of.

In what way is it not conscious? (whenever I get it actually working that is!)

Glossary

1. Amygdala - also called the archistriatum is located deep within the cortical forebrain in the medial temporal lobes and is involved in the processing of emotion as well as learning. It receives both low level ascending sensory information from the thalamus and more fully processed information that descends from the cortex. It is ideally placed to add an emotional context to experiences, which perhaps is a factor contributing to which subset of all perceptions or thoughts are selected for attention's focus
2. Cartesian Theatre - A supposed place in a brain where everything comes together and consciousness happens. It is supposed to be somewhat like a movie screen, combined I guess with snif-o-rama, a feel suit, a taste simulator, etc. The problem is that there doesn't appear to be anything that can experience the show; there is no provision for an audience.
3. Concept - an area in the brain (perhaps only virtual, logical, and/or distributed) that is specific to any of the components of thought or awareness, with a role to play in maintaining relationships between itself and other concepts. One instance is a strong relationship between the concepts 'finger' and 'hand'. There is much less likelihood of any relationship between 'finger' and 'star' for example. These relationships can by dynamic, but as they are reused over a short time, physically implemented.
4. Consciousness - in my view is a similar or identical concept as each of: the awareness of being aware (as opposed to reflex-like "perception" which is just automatic responses to environmental perturbations), active perception, awareness of ongoing thought (especially logical thinking), sentience, a subjective feeling of being an "I", and the reason for a belief that "the lights are on and somebody is actually in control and pulling the levers". Consciousness is perception from a first person viewpoint.
5. Cortical Column - a highly interconnected collection of nerve cells, arranged radially and in six layers on the surface of the brain. Each has a stereotyped architecture with defined layers for inputs (in general - L4 for inputs from outside of the cortex, L3 for intra-cortical inputs), outputs (L5 and L6) and intra-columnar connections (L2 and L3, with lots of inhibitory connections across the layers thrown in also). Individual neurons and neural networks are less interesting to me than the column they make up, which in this theory is the mechanism that represents concepts, relationships and hierarchies³⁰. Neocortical columns can be thought of as being 'logically' stacked in arrangements representing more and more abstract and overlapping hierarchies, even though they are actually physically arranged side by side in a manner similar to a collection of pencils standing on end around the nooks and crannies of the cortex, connecting by long distance axons running along L1 to other columns. Groups of columns responsible for similar functions (vision, touch, understanding speech etc.) comprise and should not be conflated with the areas Brodmann and others³¹ enumerated, or with the six layered structure of the cortex itself. I believe consciousness arises from the function and arrangement of these units, and that detail at the level of the neuron is just the plumbing (it is much easier to program and implement artificial consciousness at this higher level also). What is important here is that external sensory input to a column is distinguishable from the input from peer columns, leading to an interpretation that this is the basis of the separation of reality from internal dialogue-type activity such as dreams, plans, expectations, wishes, speculations, memory reviews etc.
6. Dualism - the belief that the mind and brain (i.e. all of matter) are distinct and separate things, and that the mind is not made up of material stuff
7. Edge - a connection between two nodes (vertices) of a network (graph), in the arcane terminology of network theory
8. Global Workspace Theory - is where everything processed comes together and is made available to the rest of the system (this would be a flavour of the Cartesian Theatre)
9. Graph - a network consisting of verticies and edges
10. Habenula - The lateral habenula (tissue adjacent to the stalk of the pineal gland) is pivotal to reward processing, in particular with regard to encoding negative feedback or negative rewards
11. Hippocampus - located on the medial (inward facing) aspect of the temporal lobes is involved in the processing of episodic and autobiographical memory, identifying (making the connection, and holding?) relationships between different things stored or held elsewhere in the brain. Concepts are born here.³² Afterwards concepts appear to be made available to the ventromedial prefrontal cortex (vMPFC), an area important in decision-making which would need to use these concepts.
12. Intentionality - In its common usage it is having or forming an intent or motivation. In philosophy it is the property of being about something or being directed toward a subject, but that is not how it is used here
13. Meme - a unit of culture which is somewhat analogous to a gene in that it can also exist outside an individual and spreads by reproducing itself. The idea of a meme can be described in terms of survival of the fittest. Memes could be implemented in this theory as (compound) concepts
14. Memeplex - a collection of interdependent memes, such as a consistent belief system or some form of complex and deep understanding (formal science, math, etc.) which has the most of the properties as does a single meme
15. Phenomenology - is the study of subjective experience
16. Qualia (singular - Quale) - the subjective quality of a sensory experience (e.g. the redness of an apple)
17. Striatum (or striate nucleus) contains the dorsal striatum which contributes directly to decision-making, especially to action selection and initiation; and the ventral striatum which is functionally strongly associated with emotional and motivational aspects of behaviour
18. Small World type network - in graph theory is a network in which most nodes are not neighbours of one another, but in which most nodes can be reached from every other by a small number of hops or steps utilizing long distance connections provided by small numbers of 'gateway' nodes.
19. Thalamus - almost all sensory signals to and motor signals from the brain pass through and are modified by the relay nuclei of this structure. It can be seen as a waypoint in the flow information to and from the cortex - ideally positioned to modify this flow and select subsets of this info to focus attention upon. A recent study intimately associates the thalamus with the neural core of consciousness [18], along with the hypothalamus, inferior parietal cortex and the anterior cingulate
20. Vertex - a node of a network (graph) linked by edges (connections), in the arcane terminology of network theory

Bibliography

Here are some interesting books and articles related to this subject that I highly recommend.

1. Anderson, Jeffrey S., Ferguson, Michael A., Lopez-Larson, Melissa, and Yurgelun-Todd, Deborah. (2010). Topographic maps of multi-sensory attention. PNAS November 16, 2010 vol. 107 no. 46 20110-20114.

2. Bast et al. From Rapid Place Learning to Behavioral Performance: A Key Role for the Intermediate Hippocampus. PLoS Biology, 2009; 7 (4): e89 DOI: 10.1371/journal.pbio.1000089 (as reported in Science Daily)

3. Basten U, Biele G, Heekeren HR, and Fiebach CJ. (2010).How the brain integrates costs and benefits during decision making. Proceedings of the National Academy of Sciences 10.1073/pnas.0908104107 PNAS November 30, 2010

4. Blackmore, Susan (2007). Conversations On Consciousness. Oxford University Press.

5. Boucsein C, Nawrot MP, Schnepel P and Aertsen A. (2011). Beyond the cortical column: abundance and physiology of horizontal connections imply a strong role for inputs from the surround. Front. Neurosci. 5:32. doi: 10.3389/fnis.2011.00032.

6. Carter, Rita (1998). Mapping The Mind. Phoenix Paperback of Orion Books Ltd.

7. de Costa, NM and Martin KAC (2010). Whose cortical column would that be? Front. Neuroanat. 4:16. doi:10.3389/fnana.2010.00016.

8. Damasio, Antonio (2005). Descartes' Error: Emotion, Reason and the Human Brain. Penguin Books.

9. Dennett, Daniel C. (1998). Kinds of Minds: Toward an Understanding of Consciousness. BasicBooks.

10. Freeman, Walter J. (1999). How Brains Make Up Their Minds. Phoenix Paperback of Orion Books Ltd.

Freeman has one of the best descriptions of what I call 'patterns of activation' throughout this discussion. The writing style is quite technical and slow going, however it is worthwhile and based on almost a half century long career.

11. Goldberg, Elkhonon (2009). The New Executive Brain: Frontal lobes in a Complex World. Oxford University Press.

12. Hameroff, Stuart (2010). The "conscious pilot"-dentritic synchrony moves through the brain to mediate consciousness. J Biol Phys (2010) 36:71-93 DOI 10.1007/s10867-009-9148-x.

13. Hofstadter, Douglas, R. (2007). I Am A Strange Loop. Basic Books.

14. Hagmann P, Cammoun L, Gigandet X, Meuli R, Honey CJ, et al. Mapping the Structural Core of Human Cerebral Cortex PLoS Biology Vol. 6, No. 7, e159 doi:10.1371/journal.pbio.0060159

15. Hawkins, Jeff and Sandra Blakeslee (2004). On Intelligence. Times Books. It is almost scary how close some of Hawkins' views match my own view of how the mind works. It's too bad he didn't go deeper into the details. I suspect it involves patents or something like that.

16. Nicholas Humphrey (2000). “How to solve the mind-body problem,” Journal of Consciousness Studies, 7, 5-20.

17. Kaschube M, Schnabel M, Löwel S, Coppola DM, White LE, & Wolf F. (2010). Universality in the evolution of orientation columns in the visual cortex. Science (New York, N.Y.), , 2010; 330 (6007), 1113-6 PMID: 21051599 as reported in Self-organising principles in the nervous system.

18. Langsjo, JW; Alkire, MT; Kaskinoro, K; Hayama, H; Maksimow, A; Kaisti, K; Aalto, S; Aantaa, R; Kaaskelainen, SK; Revonsuo, A; and Scheinin, H (2012). The Journal of Neuroscience, April 4 2012 - 32(14):4935-4943.

19. LeDoux, Joseph (2002). Synaptic self: how our brains become who we are. Penguin Books.

20. Minsky, Marvin (2006). The Emotion Machine: Commonsense Thinking, Artificial Intelligence, And The Future Of The Human Mind. Simon & Schuster.

21. Montague, Read (2006). Your Brain Is Almost Perfect: How We Make Decisions. [previously published as Why Choose This Book?] Plume

22. Neokleous KC, Avraamides MN, Neocleous CK, Schizas CN (2010). Cognitive Computation Volume 3, Number 1, 321-331, DOI: 10.1007/s12559-010-9063-2.

23. Panksepp, Jaak (forthcoming July 2012). The Archaeology of the Mind. W. W. Norton & Company. Referenced in the May 2012 edition of Discover magazine.

24. Pinker, Steven (1999). How The Mind Works. W. W. Norton & Company.

25. Pinker, Steven (2007). The Stuff of Thought: Language as a Window into Human Nature. Penguin Books. Contrasted with his other book referenced above, this one was a bit of a disappointment. It had more to do with language than it does with insights into how the mind deals with language - which unfortunately was how it was billed on the cover.

26. Searle, John R. (1997). The mystery of consciousness. New York Review Of Books. I don't agree with some of his reasoning, but he seems to reach similar conclusions as I do anyway.

27. Singham, Mano (2010). Benjamin Libet's 1989 famous experiments, as discussed in Machines Like Us

28. Singham, Mano (2010). Chun Siong Soon, Marcel Brass, Hans-Jochen Heinze, and John-Dylan Haynes. Unconscious determinants of free decisions in the human brain, Nature Neuroscience, vol. 11, no. 5, May 2008, 543-545 as discussed in Machines Like Us.

29. Sporns, Olaf (2011). Networks of the brain. Massachusetts Institute of technology.

30. Stanovich, Keith E. (2004). The Robot's Rebellion: finding meaning in the age of Darwin. The University of Chicago Press.

31. Tononi, Giulio (2008). Consciousness as Integrated Information: a Provisional Manifesto. Biol. Bull. 215: 216-242. (December 2008)

License

Basically if you like what you read here and can find any original thoughts, feel free to use them but also please attribute your use to this site and to me - Reigh LeBlanc.

This work is licensed under the
Creative Commons Attribution 3.0 License

Social Bookmarking

Footnotes

1. Brain Facts and Figures

2. Wikipedia entry on the Technological Singularity

3. Implications about the accuracy of this date

4. see The Worm Brain

5. see IEEE Spectrum: The Consciousness Conundrum I really like this definition!

6. Whatever you do, DON'T THINK OF AN ELEPHANT!

7. some time after I wrote this I came across Giulio Tononi's 'manifesto' which I think is mostly on the right track. see Consciousness as Integrated Information: a Provisional Manifesto and [31]. Tononi makes a chilling prediction for those of us with relatives who have Alzheimer's disease or other brain insults - namely that the quality of an experience depends not only on the neurons that are firing but also on those that are not. Even though not firing, the removal of some of these neurons will fundamentally change the experience. I am not sure I like the implications.

8. Wikipedia entry on the Turing Test

9. see Searle [26] page 11-12

10. see for example Soon [27] and [28]

11. Page 127 of Narrative knowing and the human sciences By Donald Polkinghorne makes reference to William James' 1890 concept of a six to twelve second period that encompasses the seemingly instantaneous present.

12. see for example Soon [27] and [28]

13. see Pinker [25]

14. see Anderson [1]

15. see Basten et al. [3] which shows how the brain keeps track of benefits and costs and which one is the winner

16. see [22]

17. This really means 'directed edges' in graph theory terminology

18. making this kind of network arrangement a directed, weighted, mixed multigraph see the Wikipedia entry for multigraph

19. similar to Kaschube [17]. Note that Kaschube et al. are discussing the development of fine brain structure while I am alluding to the laying down of new nodes and fast neural pathways (short term learning) which over time become more structurally based (consolidated memory) - similar concepts but not identical. My point is both are self organizing and might even use some of the same structures and mechanisms.

20. see also Fruit fly nervous system provides new solution to fundamental computer network problem for how neurons can self-select some but not others

21. see The secret life of the brain

22. see the previous footnote

23. see this sciencenews article for a description.

24. see LeDoux [19] page 292

25. see Sporns [29] page 122

26. see Humphrey [16], of the original discovery of blindsight fame

27. see Hebbian theory in Wikipedia the Wikipedia entry on Hebb's therory

28. This idea was inspired from Goldberg's speculations starting on page 263,[11] but in this case does not depend on his hemispheric differences (which themselves probably contribute significantly to the generation of consciousness too).

29. see Bast et al. [2] for an example involving hippocampal neurons firing "when — and only when — the animal passes a specific place"

30. even if concepts are not physically represented in actual columns, the idea of separate logical functional areas that are interconnected still works. For a critical look at problems with the cortical column concept see da Costa [7]. Cortical columns seem to be a structure in search of a generalized function. In primary sensory areas it looks like the evidence is clear enough, but in associative and higher abstract reasoning areas there are lots of problems concerning the nature of the role(s) they play. There does seem to be emerging evidence that many if not most of the connectivity is from long distance connections vs. connections from adjacent neurons.see Boucsein [5].

31. see the Wikipedia entry for Brodmann area

32. See the NewScientist article