There are two views of consciousness: (1) that it is part of a computational process and (2) that it is something we experience. These two views have to be reconciled, and this shows that the feelings we consciously experience must be part of the computational process. For example, having the feeling of pain is part of the process of deciding what to avoid, and failure to feel pain leads to failure to avoid what should be avoided. A major observed function of consciousness is to find solutions to difficult problems. The production/judgement model separates the production of possible solutions from the judgement process that decides whether or not to apply a possible solution. An analogy is drawn with decision-making processes in society, such as parliamentary democracy where political parties propose solutions to national problems and voters choose between parties. Feelings are identified as being part of the judgement process. A careful distinction is made between judgement and reward. Production/judgement systems contain one or more central control channels which are very low-bandwidth. Such a control channel in the brain would consist of a group of neurons all contributing to a single output. The noradrenergic, serotonergic, dopaminergic and cholinergic nuclei are all candidates for being one of these control channels.
The human central nervous system is a complex information processing system. Sensory neurons input information from the outside world. Neurons that control muscles output information. Interneurons receive information from their input neurons, process it, and output the result as action potentials which are then received by their output neurons. They may also store information about processing that has occurred, by altering parameters that control synaptic behaviour. The function of the CNS is to process information, and it has no other function.
Subjective experience and consciousness are strongly associated with activity in the CNS. There are two views of what consciousness is -
David Chalmers asks the Hard Question: even if a system performs the computation that a functionally conscious being performs, why does this give rise to subjective experience? (Chalmers, 1995) Where do the qualia come from?
In the search for a hypothetical beast that is functionally but not subjectively conscious, philosophers have constructed the zombie - the being that can and does perform all the computation and information processing that a conscious being performs, but which does not have conscious experience (Moody 1994).
Epiphenomenalism is the idea that subjective consciousness might just be an unnecessary but interesting accompaniment to functional consciousness.
Eliminativism is the even more extreme idea that subjective experience doesn't really exist, that it is just some sort of illusion. In which case we don't have to explain why it accompanies functional consciousness. But this is just too implausible.
Pain is a subjective experience that can be arbitrarily strong, so it focuses our mind on questions about experience. The mystery of the experience of pain is its horribleness. Why does pain feel horrible?
The function of pain is not too hard to ascertain. It tells us to avoid or prevent whatever it is that causes pain, and the stronger the pain, the more urgent the message. A message of pain can be ignored, but the stronger it is, the "harder" it is to ignore. I put the word "harder" in quotes, because it is a term that has subjective meaning, but it is not clear what its functional or computational meaning is. If something is subjectively "harder", how does that translate into function? Is it less likely to occur? Is there some limit as to how many times it is allowed to occur? These questions are equivalent to the question of free will. If there is such a thing as free will then it is free will that we use to do those things that seem "hard", for example, ignoring pain.
In the long term, the relationship between function and experience is observably strong. There are people who for some reason do not experience pain. They experience the sensations that would normally be associated with pain, but they do not experience the pain itself (Bear et al, 1996, p 340). In particular they do not experience the horribleness of it. And almost inevitably, they fail to look after themselves and die young (usually before 30) from the effects of repeated injuries. This happens even if they are made aware of the danger that their inability to feel pain puts them in and are taught to be careful not to hurt themselves. In normal people who feel pain, the pain they feel prevents them from allowing so many injuries to occur.
So in the case of pain, the association between function and subjective experience is quite obvious in the long term, where we can strongly predict features of observed behaviour from the nature of the subject's experience, but less obvious in the short term, where all we can say is that pain makes it "hard" to ignore the cause of the pain.
A similiar short term/long term distinction can be observed in the case of hunger and dieting. There are many people who feel that they are overweight (in many cases they are) and wish to lose weight. In order to do so, they must eat less, or eat less of certain types of food. They go on a diet and decide to eat less of what they should not eat. Subjectively they find it "hard" to not eat when they want to eat. "Hard" is not the same as impossible, so in the short term the diet may succeed. But it is almost inevitable that a diet based on using free will to ignore hunger will fail in the long term. That is, if a diet causes the dieter to subjectively experience the difficulty of forgoing desirable food, then we can predict that the diet will eventually fail and lost weight will be regained.
Based on these examples, we can formulate a hypothesis about consciousness – that its function, or part of its function, is to convert accumulated short term "hard"'s into long term "don't do it"'s. Discovering the precise nature of this accumulation process would then tell us more about the relationship between consciousness as a computational activity and consciousness as subjective experience, both in the long term and in the short term.
This would enable us to robustly refute the epiphenomenalists. If the computational processes occur if and only if the associated subjective experiences occur, then a subjective experience is not just something associated with a computation, it must actually be that computation.
If subjective experience and functional computation are to be so strongly identified, then for each and every qualia that we experience, we can ask the question: what computation is being performed by the experiencing of that qualia?
Another hypothesis about the function of consciousness is that its job is to provide solutions to problems, in particular problems that involve new or difficult situations, or problems that involve resolving conflicts between different considerations. Evidence for this hypothesis is that we can only solve these types of problems if we consciously pay attention to them.
Humans live in societies. In many cases there are identifiable groups of humans that must work together in some way to solve certain problems. Examples of such groups are countries, companies, clubs, gangs and families.
It is useful to ask what information processing systems allow the problem solving abilities of group members to contribute in an orderly fashion to the problem solving activity of the overall group.
In many cases a production/judgement model applies. In the production/judgement model, some members of the group produce possible solutions to a problem, and other members judge which of those possible solutions should be implemented..
In all these situations there are producers who produce possible solutions to problems. These solutions and the processes involved in arriving at them can be quite complex and difficult and may require on-going and continuous effort. There are also judges. The judges are presented with the results of the efforts of the producers, and they make a simple but final yes or no decision. The amount of time and effort that the judges spend judging possible solutions is often much less than the time and effort spent by the producers in producing those possible solutions.
These production/judgement systems are not perfect. The policies of the political parties may not correspond very well to the desires of any of the voters. Perhaps the voters could directly formulate legislation themselves. But production/judgement systems are common, and they are often the only practical way to decide upon and implement solutions to problems that affect all members of a large group.
In matching the production/judgement model to consciousness and brain activity, it is natural to attempt to identify production processes and judgement processes. We might be tempted to match judgement with what psychologists call reward. But reward is a measure of the success of a solution to a problem. This measurement can only be made after the solution has been implemented. Judgement about whether or not a possible problem solution should be implemented must be made before implementation. So judgement and reward are different processes that occur at different times, with judgement occurring first.
For example in the shareholder/directors example, reward would correspond to the profits that the shareholders receive. The process of shareholders receiving profits is quite distinct from the process of shareholders choosing company directors. In the entepreneur/consumer case, the "reward" is whether or not the purchased goods turn out to be useful to the consumer.
So judgement is something that precedes reward. Nevertheless, it may often be the case that there is a meta-judgement system where the rewards pass final judgement on the judges. After all, shareholders in firms that go bust will cease to be shareholders. But this meta-judgement is more mechanical and is of a different nature to the judgement that the shareholders make of their directors, which can involve complex calculation on the part of the shareholders. In the human brain, it would be the meta-judgement system that causes the above-mentioned conversion of short-term "hard" into long-term "don't". A "hard" decision is usually one that conflicts with immediately anticipated rewards – the implication is that the reward system in the brain slowly but surely acts on the judgement system and prevents it from indefinitely ignoring anticipated rewards.
I propose a multi-stage model of consciousness:
Some of the feelings that we experience can be identified with step 3, feelings such as:
These sorts of feelings correspond quite directly to the passing of judgement, (i.e. given a possible solution to a problem, should we go ahead with it?), and all have a clear positive or negative nature.
A plausible hypothesis about other experienced feelings is simply that they represent earlier stages in the calculation that results in this final judgement. This is certainly plausible for those feelings that have strong positive or negative nature – they would represent inputs favouring positive or negative judgements respectively.
Other qualia have a less obvious judgemental significance. They are not obviously positive or negative. For example, the redness of red. Red has its own distinctive qualia, but is it positive or negative?
Extending the hypothesis just given, we can suppose that these apparently non-judgemental qualia represent even earlier stages in the calculation. They are inputs that might have judgemental significance, depending on context.
For example, consider the qualia of redness. If it is the color of an apple, then it tells us that the apple is ripe and good to eat, and represents an input to a positive decision to start eating the apple. If it is the color of blood, then it tells us that someone has suffered a severe injury, which depending on further details of context could contribute to many possible positive or negative judgements.
Are all perceptions qualia? Or are some more qualia-ish than others? The strongest qualia would be those that make the strongest contributions to judgements or those considered most likely to make a strong contribution.
Some perceptions could be classified as qualia in advance, this being specified genetically. This would probably be the case for those sorts of qualia that we all agree on as qualia, such as colours and smells. Other perceptions could acquire qualia status as a result of experience showing that they can usefully contribute to the judgement process. For example, a chess master could have feelings about chess positions, and claim that to him or her they are qualia. A non-chess player is not going to be able to understand "what it is like" to have these feelings, and if they are indeed qualia then they are qualia acquired by experience.
A critical feature of production/judgement systems is that the centre of control is a very low-bandwidth channel.
For example, the running of a democratic country involves enormous information flows. But the centre of control is the election. Typically, an election may occur once every four years. Often the electorate makes a choice that is effectively between two major contenders, in which case the bit rate for this channel is one quarter of a bit per year.
Such a low bandwidth channel may seem insignificant in the scheme of things. Hundreds of bits get transferred every time one civil servant sends email to another civil servant. But the significance of the electoral control channel becomes obvious if you take it away. Then you are no longer living in a democracy. Conditions of life in democracies are usually significantly different to those in non-democracies (even if some would claim they are not always better).
There is necessarily an enormous convergence towards and divergence from these control channels. Decisions made by millions of voters go into the final election result. Electors receive information from many sources that contribute to their decision-making: from the media, friends, and organisations they are associated with. The result of the election chooses the government, and there is then an enormous outflow of information from this choice, i.e. the passing of new legislation, the implementation of that legislation, and the application of policies for the administration of the whole apparatus of government.
In a more precise analysis of a democracy we may discover several qualitatively different low-bandwidth channels. In some democracies the party affiliations of MPs are unreliable, so we cannot consider the election result to be simply a choice of one government or the other. We can identify two consecutive judgement processes. Firstly the voters of each electorate choose an MP, from two or maybe slightly more major choices, once every four years. Secondly the MPs then vote on legislation presented to parliament, with their votes adding up to a final yes or no for each item of legislation. The voting-for-MPs channel has a bandwidth equal to N/4 bits per year where N is the number of MPs. The voting-for-legislation channel has a bandwidth of one bit per item of legislation presented and voted upon.
The other production/judgement examples given above also have identifiable low-bandwidth channels. In each case this channel is where the result of the judgement is summed up so that it can be applied to the decision-making process. Consumer preferences for a product sum up to a simple sales volume, which together with other factors determines the company's net profits resulting from sales of that product.. Shareholders vote for a limited choice of directors. Shareholders that consider a takeover bid make a simple yes or no judgement.
With these examples in mind, it seems logical to search for one or more low-bandwidth control channels in the brain. Conceivably such a control channel could be implemented by a single neuron, whose output represents a judgement about whatever is currently being judged. But implementation of consciousness in a single neuron would be a rather fragile solution. What seems more likely is a large group of neurons (at least in the hundreds or thousands) all performing the same calculation, or performing calculations in such a manner that their outputs are summed together by some means other than input into a single neuron. This is in contrast to what is normally found in the brain, which is a group of neurons in some particular area, all performing the same sort of calculation, but nevertheless all performing different calculations of the given sort and all passing on their individual output results separately to distinct inputs of other neurons.
Investigators searching for the physical substrate of consciousness have perhaps failed to separate production and judgement. They have looked for a high-bandwidth central control channel, a "little man" inside the machine, and have failed to find him. They have been looking for the wrong thing and have even come to the conclusion that maybe there is no central control, that it is all done in parallel.
The most obvious candidates for central control channels are the various nuclei that broadcast neurotransmitters non-specifically, i.e. the serotonergic, dopaminergic, cholinergic and noradrenergic nuclei. A critical feature of their outputs is that they involve special neurotransmitters and these neurotransmitters are allowed to diffuse into the brain medium before reaching their targets.
If there was only one such nucleus broadcasting one neurotransmitter, it would be easy to identify it as being the control channel of consciousness. Unfortunately we are spoiled for choice, and have to guess which of these nuclei if any could form part of a production/judgement process.
This is activated most strongly in response to unexpected non-painful stimuli (Bear et al, 1996, p 420). In relationship to consciousness, this seems to correspond to the identification of those situations for which the production of solutions is required, and corresponds to step 1 in the multi-stage model of consciousness given above. Judgement is something that must come later.
This is strongly implicated as a "reward centre" (Pinel, 1997, p341-7). As explained above, reward is something that occurs after and separately from judgement (step 5 in the multi-stage model).
This seems to play a crucial role in the initiation of voluntary behaviours (Bear et al, 1996, p 422). As such, it could be regarded as a final "OK" to a proposed consciously initiated behaviour, and may be one of the judgement systems that we are looking for.
The functional role that these play in the awake brain is not very well understood. (It is known that they also play a role in the control of sleep/wake cycles.) Drugs that affect serotonin transmission and reception (in particular the famous fluoxetine otherwise known by its brand name "Prozac") are used to treat various mood disorders including clinical depression and anxiety disorders (Pinel, 1997, p 461 to 464).
Both depression and anxiety have a general effect on decision-making processes, and this implicates serotonin as a possible carrier of the results of a judgement process. Both anxiety and depression can be interpreted as preventing the performance of voluntary behaviours. Anxiety prevents those behaviours that conflict with the subject of the anxiety. Depression causes pessimism about the outcome of any behaviour and thus leads to inaction.
These two examples suggest that serotonergic nuclei play a role in suppressing negative judgements about behaviour. In other words, at stage 3 of the multi-stage process of conscious judgement described above, when a possible solution is considered, the expected negative consequences of the possible solution are considered, and the serotonergic nuclei have the job of deciding whether or not to allow those negative expectations to prevent the possible solution being accepted. By suppressing the effect of negative judgement, the overall effect of the serotonergic nuclei on conscious decision making is positive.
Some of these are more active during REM sleep than during wakefulness (Pinel, 1997, p 313).
It is conceivable that the production of dreams during REM sleep is itself the result of a production/judgement process, i.e. that the content of dreams serves some purpose, and that judgements and rewards are applied as to how well particular dreams serve that purpose. One possible purpose of dreams is to test the ability of the brain to comprehend and react to the situations that occur in those dreams. Dreams are often more bizarre or extreme than normal real life. Their purpose could be to ensure that the brain does not behave in unexpected ways when a bizarre or extreme situation does occur in real life.
One piece of evidence for this theory is the normal non-lucidity of dreams, i.e. dreamers do not usually realise that they are dreaming. Dreams would be less successful tests of reaction to dream situations if the part of the brain doing the reaction realised that they were not really real. The occasionally observed phenomenon of dreaming of waking up from a lucid dream can be seen as a deliberate stratagem to restore non-lucidity, applied by those parts of the brain responsible for the production and success of dreams (if you become aware that you are dreaming, but then you dream that you have woken up, you have effectively lost the awareness you had that you were dreaming).
What I'm suggesting here is the existence of a completely separate consciousness – one responsible for inventing dreams, and which has its own separate control channel or channels. The existence of separate consciousnesses in the brain is not implausible, because split brain patients (those with their corpus callosum severed) are sometimes observed to have two separate consciousnesses operating within their brains.
Libet (1996) reports various stages in the production of a self-initiated voluntary act, as observed in EEG's:
There is therefore a 100 to 150 millisecond window of opportunity between stages 2) and 3) for conscious control to veto the willed action. Libet concludes that conscious awareness does not control the initiation of the voluntary act, and only exerts control via its veto power. This model of voluntary behaviour is very similiar to the production/judgement model that I have described in this paper.
A production/judgement system separates the production of possible solutions to problems from the judgement as to their suitability. If this is a correct model for human and animal consciousness, then the conscious feelings that we have are naturally identified as the final stages in the calculation of the judgements. Perhaps surprisingly, given the traditional association of consciousness with the complexity of the brain, the neural correlate of consciousness that we should be looking for is one or more control channels with very low bandwidth, possibly consisting of groups of neurons all performing the same calculation and outputting a summed result.
