I would like to describe a thought experiment that I would like to discuss. Assume that I was able to set up a computer simulation of my brain and its initial conditions as of some instant in time. Now I set the simulation going and feed in the sensory inputs over the next, say, 5 seconds.
(i) Would the behaviour of my simulation diverge from my own behaviour ?? - if so, why ?? e.g. Is my brain's sensitive to initial conditions ??
(ii) Is my assumption that I can set up such a simulation *fundamentally unreasonable* ??
Note that my interest is in emulating me at the behavioural level. I am not concerned with the accurate emulation of the internal operations of my brain at any level.
I realise that what I have said above is not particularly rigorous. I would like to be able to discuss the above thought experiment with you in order to *tighten it up* and explore it further.
I realise, of course, that the above thought experiment is just a way of looking at the basic aims of the study of Artificial Intelligence, i.e. the emulation of human cognitive capabilities. However, I feel that the way I have put the questions is interesting and I will certainly be putting some effort into refining and extending the experiment.
My own intuition is that it is essentially possible to set up such a simulation and that the behaviour of the simulation will slowly diverge from my own behaviour due to the differences in both the starting conditions and the subsequent sensory inputs. But I have no idea what "slowly" means in this situation. It may well be that the simulation exhibits aberrant behaviour, e.g. "goes mad", in 5 seconds or maybe 5 days. I have no idea.
Although I believe that such an emulation is "essentially" possible there is one area where I know that it not possible with today's technology and that is in playing the game of go. I can play go better than any computer program.
Since I believe it is possible the next question is where to start. Does the architecture need to be a precise copy of the brain at the neuron level ? Can it all be done at the symbolic level with perhaps a mixture of expert systems and semantic nets.
Another interesting question to ask is how much of this "emulation of a human brain" has already been done. Where are the challenges today. We already have systems which understand what they see, understand what they hear, hold conversations and reason. If I just bolted these together and set up the intial conditions appropriately, how closely would the resulting system emulate me, or you, or could we manage to emulate a 5 year old child ?
A good place to start with an emulation of me would be to use the Turing Test scenario. The success of the experiment would be determined by how long it took a close friend of mine to decide which of the two entities they were interacting with via text chat software was me and which was an emulation.
A further simplification to the above Turing Test scenario would be to choose a subdomain of my knowledge such as my knowledge of the game of go. Would it be possible to emulate that well enough to convince a go novice the the emulation was actually me ?? I think it would be possible. I wonder how long it would take - and what it would prove ??
Assistance on the above project would be greatly appreciated. Just email me.
Last updated 30 July, 2002