« Comedy | Main | Living in France »

Sunday, April 20, 2008

Warm robot

There's been a lot of buzz recently about an experiment which purports to show, using FMRI scans, the evidence of a decision taking place in the mind some significant time before the occupant was aware of making that decision - or even of the question arising, it is said.

This comes many years after another experiment where, using some sort of electrical tracking rather than FMRI, a trace of an intention could be detected before the subject decided to flex a muscle. That is, in some electrical sense the subject's brain initiated a movement before the individual had conceived mentally of the action.

Another piece comes in the well-established idea of 'confabulation', the tendency of the brain to adopt a memory as its own, and therefore as fact, though it may have been planted recently by another, intent, perhaps, on enlisting the support of a witness. It would seem that the brain shows an inherent love for consistency in a constructed memory when contrasted with its own patchy recollections and will overwrite any gaps with any adequately well-structured fiction. This is seen in some criminal investigations with susceptible witnesses but is considered to be common in all of us with our fond and firm memories of childhood.

Then there's the thing that has always worried me. We're not made of wire. Our nervous systems and our brain work by chemical signalling not by electricity. Signals between our fingers and our brain (and back again) are conducted in a manner more like a rippling reaction in a fluid than an electromagnetic wave in a vacuum. The speeds involved, the speeds with which all information flows in our bodies and minds, are more like the speed of sound than that of light, measured in tens of metres per second rather than millions.

It is true that the brain works in a parallel fashion, millions of parallel computations happening at once but change based on the propagation of information is limited by the speed at which it arrives.

I've been a programmer now for forty years and feel in my gut the limitations of a single process working on a flow of information. A single serial process can act only as fast as change is presented to it and is propagated through it. Equally a billion processes ranged to act in parallel can still only deliver even their wonderfully complex conclusions after the arrival of a stimulus and can deliver it only as fast as that propagation can cross the medium of its computation.

Thus, to me, it seems incredible that I can be feeling, thinking and acting at the speeds I do.

Free will is the last issue in the mix. My will is imposed between my perception and my action. The complexity of my decision making is down to the vast parallelism of my brain, but the timing and the justification for it is limited by the need to integrate many, many channels of sensory input and to distribute the results of these decisions to a complex array of muscles.

It doesn't seem right. It pushes the limits of my belief.

Now what I propose is not unreasonable. It is certainly not unreason. I am not going to propose something unscientific, something born of a leap of faith. This is a proposal, a model built on my perception of the world, apparently consistent as far as I can test it. A scientific hypothesis ready to be destroyed by anyone with more knowledge. It just seems to me to be a reasonable idea.

Imagine an organic robot, programmed to respond more-or-less perfectly to its surroundings. Stimulus arrives in expected or unexpected forms. The expected, internally modelled, environment is handled by a set of long-established and nurtured programs. Rituals have been programmed into this robot through custom and learning.

Also, this robot is programmed with the capacity to respond to unexpected situations, to shocks and hurt by a set of low level reactions and to less immediate but equally challenging changes by a process of assimilation and adaptation.

This robot is a machine. It is not self-aware in any sense that we respect.

The robot, being organic, cannot respond very quickly and relies heavily on a predictable environment and cooperation from other robots. Sometimes, things happen which leave whole groups of robots short of a response and they often react relatively inappropriately while a suitable action is devised. It is not (evolution has asserted) good for this random searching to be incorporated in the robots' programming and so no explicit records of these moments are kept.

An interesting part of the programming of this robot is this adaptation mechanism. A program runs which watches and evaluates the robot's activity as to appropriateness and success or failure. This evaluation consists of constructing a set of theorisations of the big picture which the robot has neither the time, the processing speed or the omniscience to gather. This program constructs a continuous evaluatory narrative formed of the recent past.

This narrative, separated completely from reality (since it is constructed as a consistent theory of what is observed gathered and collated post hoc), is a confabulation. It is however, for the explicit purpose which it serves, entirely consistent and reliable.

So much so that this program forms a belief, a model of reality, solely from this confabulation.

You are this program. You do not control your actions. You are merely the observer riding in a warm robot, programmed to be, as far as the other robots are concerned, you.

So where is free will if you are not controlling your body?

Well, I see the reason for evolution constructing this program is entirely (and solely) to improve the robot's activating programs. You observe, you judge and you retrain the robot's programs. The next time the changed program is called for, it performs more to your latest standards. The review program sees the new behaviour and blesses it, reinforces but mainly confabulates it (for its own sanity and analytic purposes) as its own deliberate action.

I just stopped typing and scratched my nose. It's something I do. During the action, it felt like 'I' was scratching 'my' nose. That is my confabulation. I needed to model it like that in case it wasn't a useful action. I'm fairly happy to think that the robot scratched its nose because it's what it does and only now (while it's typing away on this paragraph) my control program, me, has retrospectively constructed this confabulation of the event simply to allow me to consider whether it was a good piece of programming and whether it should be suppressed or reinforced.

Long term free will is just that project of finding oneself in a body and planning to make it work as well as you can. Confabulating continuously that you are actually living and moving and interacting, just modelling the possibility that your body was even chemically capable of such immediate action, you work on the reprogramming of the robot for your long-term ends. When these come to pass, as your host body gets by, you feel that you are really doing it.

It takes some thinking about, I guess. When the separation and the isolation caused by the confabulation process hits you at its extreme, it delivers quite a philosophical punch, especially as the ultra-dual separation between the slow, slower than the world could tolerate and much slower than you think, body and the completely detached mind leaves one simply as a custodian of something which, if it is you, is not you at all.

Posted by Martyn Horner at 22:02
Categories: General