a blog about things that I've been thinking hard about

Consciousness is a Centre of Accountability

29 November, 2005
consciousness, like a president

Consciousness doesn't make all the decisions in your brain.

Nor does consciousness propose possible decisions.

But consciousness does make a final "yes"/"no" choice on all important decisions proposed.

And consciousness is held accountable for those choices that it makes.

The Unity of Consciousness

One of the primary features of consciousness is its unity. Consciousness is the "me" in your mind, and there is only one "me".

Another feature of consciousness is that it appears to be responsible for all the high-level decisions that an individual makes about what to do. Most of the time consciousness doesn't decide which foot to put in front of the other, but it almost always has the final say on decisions like "which girl shall I marry?", and, "should I take the job?".

Taken together, these features suggest that consciousness is like the "boss" or the "president" of the brain.

The Job of President

The president of a country doesn't directly control everything that happens in his country. He doesn't even know everything that happens in his country. What he does do is make decisions about the most important things, and he has aides and government officials who provide him with the information which is most relevant to making those decisions.

One mysterious fact about human consciousness is that sometimes a conscious decision to do something is made after the person actually does it. This seems to imply that maybe consciousness isn't the real decision maker, that really someone or something else is doing the decision making.

But we can understand why this happens by considering analogous situations in the job of president. There are some decisions in the running of the country that may have to be made by individuals other than the President, because the situation requires a response faster than the President can deal with it. At the same time, the considered action may be so risky (to national security etc) that the President should be consulted first.

What to do in such a situation?

One approach is for the officials dealing with the situation to act immediately, and for the President to be consulted. If the President agrees with the action, then this confirms that the officials acted correctly. If the President disagrees, then it's too late to retract the officials' actions, but they will know that they shouldn't do the same thing next time (or at least not without consulting the President and actually waiting for his OK).

Something similar may be going on with consciousness – the conscious decision always matters, even if sometimes it comes after the completion of the required action. If the conscious decision disagrees with the non-conscious action, then the non-conscious action will be prevented next time.

The Rubber Stamp President

The President's aides sometimes decide what to do before asking for the President's confirmation. We can consider the possibility that they always decide what to do, and the only thing that the President ever does is say "yes" or "no".

One advantage of this arrangement is that it makes it extra easy for the President to delegate certain decisions. All he has to do is OK the same type of decision repeatedly, and then he can tell his aides that they don't even need to ask him in future. And if he disagrees repeatedly with a proposed action, then the aides will learn not to even consider that type of action in the future.

Responsibility

The President not only makes the important decisions, he is also responsible for them, and can be held accountable for them. In a democracy, that mainly means that he can be voted out of office.

In the brain, things are a little more complicated. The analog of a President would be an individual neuron. But to have a single brain cell in charge of everything seems a little fragile. Also the brain is known to use multiple neurons to represent even single items of information (by population encoding), so this suggests that if there is a neural equivalent to a President of the brain, it is a population of neurons.

These neurons would be like a Presidential Committee, and each Presidential neuron would vote on each proposed course of action, with overall votes deciding whether or not the action went ahead. Each member of the committee would be held responsible for the consequences of their vote, by considering the success of any action in relation to whether or not that committee member voted for it, and by increasing the weighting of the members that voted for more successful actions and decreasing the weighting of those who voted for less successful actions. The weightings would translate into how much each member's vote counts for in future votes.

The one loose end in this arrangement is how to calculate the "success" of a given action. The behaviourist theory of operant condition is based on the idea that success is measured entirely as a function of the satisfaction of an animal's or person's biological drives in the very near future, "very near" being perhaps no more than a few seconds after an action is taken.

This theory seems radically inconsistent with our human experience of acting over the long term, and in its pure behaviourist form it appears to ignore all possibility of internal thought and other complications of mental state.

However, and this is a big plus, it does admit of a straightforward implementation. As long as the decision-making neurons retain a memory of how they have voted for or against a particular action in the last few seconds, this memory can be combined with information about the animal's or person's current satisfaction of drives, to determine an alteration to the neurons' future weightings (i.e. their voting power).

Does it make sense to consider operant conditioning plausible because it is easy to implement, without regard to its failure to explain the more complex aspects of human thought and cognition?

We have to remember that biological organisms are not the result of far-seeing architectural design; rather they are a result of piecemeal evolution. A learning system that only takes into account short-term consequences is better than nothing, and it is relatively robust. It is also consistent with the behaviour of "lower" non-human animals, whose behaviour tends to be much more "short term" than that of people. And we know that even people suffer from problematic "short-term-ness" in their behavioural strategies. These observations are consistent with a system which initially evolved to learn according to short term rewards and punishments, but which later evolved various adhoc additional mechanisms to take some account of the long-term. (A full discussion of all the ways this could happen deserves a whole separate article, and I have partly treated the issue in my article Bootstrapping the Mind.)

Conscious Perception and Qualia

If consciousness is this high-level decision-making process, then where does conscious perception come into it? And what about the qualia, the "what is it like" aspect of conscious, e.g., "what is it like to see red?".

Considering subjective consciousness as an decision-making function, conscious perception must consist of no more and no less than the sum total of information relevant to those decisions. Returning to the presidential analogy, conscious perception is everything that the President is told about by his aides so that he can make the best possible decisions.

Much of the discussion about qualia has to do with intensely subjective experiences like colour, smell and taste. For example, "red" is a common topic of discussion, presumably because red is a subjectively stronger colour than most other colours. The redness of red remains strong, even when it is obviously not very important which colour a particular object is. For example, the stapler on my desk is red, but the stapler's colour has little effect on my decisions about whether or not I use it to staple things, and a blue one would do just as well.

It is, however, well to remember that many of the colours of things that we see are artificial colours. In a past prehistoric environment, bright colours were more likely to have particular significance. Red is the colour of blood, and it can be the colour of a ripe fruit. Green is the colour of plants, blue is the colour of the sky, black might mean that something is dark (and things could be hiding there). Red, orange, white and yellow are the colours of fire, something quite important in prehistoric times.

Our ancestors' brains evolved strong conscious responses to colour because colours were quite often immediately relevant to conscious decision-making processes. The same applies to smell. Even though humans are much less smell-oriented than other animals, if you can smell something, it is often quite significant, whether it be the presence of a fire, or that the food you are about to eat is rotten.

One major fact about qualia is their quality of "differentness". Colours sufficiently distinct from each other seem to have completely distinct qualities. Similiarly for distinct smells.

The information processing interpretation of this observation is that the effect of distinct qualia on conscious decision making are independent of each other. In other words, if the colours black, blue, green and yellow have particular effects on a particular type of conscious decision, this has no a priori implication about what effect other colours such as red, white or brown should have on the same decision.

Attention

One aspect of consciousness is attention. Attention relates to consciousness in two ways:

In the presidential analogy, I said that conscious perception corresponded to the information that his aides made available to him. But this arrangement seems to leave him somewhat vulnerable to manipulation by his aides. One way to deal with this problem is for the President to have the ability to take – or at least confirm – actions which consist of efforts to acquire more information. (This analogy seems particularly ironic if we consider recent events where a certain President took actions based on information provided which turned out to be wrong, and where the President seems to have avoided reponsibility for the quality of information that he acted on.)

Does This Theory Solve the "Hard Problem"?

The "Hard Problem" is based on David Chalmers' claim that even after we solve the "easy" problem of explaining the information processing functionality of consciousness and its physical implementation, we still won't we able to explain the "why is conscious feel like it does?" kind of question.

My own approach to explaining consciousness is based on the assumption that Chalmers is exactly wrong, and that the mystery of consciousness will be solved precisely by considering its subjective aspects and assuming that they are a direct indication of the information processing function which consciousness represents, with the added assumption that the information processing function has a straighforward physiological implementation, without any additional non-physical "magic".

This probably won't convince the "hard-problemists", and I probably haven't yet explained all the subjective aspects of consciousness that need to be explained, but I think my theory is a plausible move in the right direction.

Vote for or comment on this article on Reddit or Hacker News ...