brain simulations

(this is a VERY random thought - more a wondering than a pondering!)

A major goal of technology has always been to emulate human behaviour, especially reasoning, and far from getting closer to a solution, our modern advances seem to simply reveal more and more how hard the problem really is.

One problem is we are looking for a 'design', and this requires first an understanding of what it should really do, and then how it should do that. But this is of course a 'top down' approach, whereas the thing we're trying to copy, the human brain, is the product of a 'bottom up' mechanism - the evolution of our species over millenia.

Ironically our 'higher level' conceptual approaches can't seem to match the mindless simplicity of the natural process. The reason being of course, that we are trying to produce the same 'work', and work is a product of power and time; we have the power, but nature had the time, and for the moment our power pales in comparison.

There are though two ways to achieve something - through planning, and through trial and error. In principle any method could be discovered by just trying out all the possibilities, and (within the limits imposed by 'adjacent possibilities') this is basically what happened during our evolution - over time multitudes of organisms of increasing complexity were thrown into the testing ground of the world, and slowly whittled down until just us were left.

Even for the simplest problems, we don't ever really have time (or inclination) to really apply trial and error, but we have a powerful substitute : computer simulation. It could be that since we don't fully understand what we are really trying to emulate from the off, then maybe a better approach would be to use simulations and then see if anything produced actually matches what we were looking for (our ignorance in understanding the brain then is out of the equation, since we might not know much about intelligence but we know what we like when we see it).

The question then is, is this feasible, and if so what would be involved?

What i'd be interested to know sometime, is if (and that's a big if) we had a model of an earth environment, and could model some basic creatures in it (we could cheat and start of with some primitive lifeforms we DO understand better) - then how many iterations would be needed to match the simulation runs performed by evolution?

There are admittedly many parameters involved - at the very least not just the number of generations (years/lifespan?) but also the number of relevant interactions (reproductions/terminations) per generation.
But surely there is a number which could be worked out,even from what we currently know about the history of the earth and biology?

And if so, then what is the number of computer operations, or the number of years for a computer running at 1GHz (1,000,000,000 operations per second) it would take to run?

And I don't even think it is a problem that nature basically ran one large simulation, but for a very long time : that large simulation (history) consisted of a multitude of smaller simulations in parallel. Of course there needs to be room for cumulative effect, but it is also probably the case that the starting points in nature (the number of relevant creatures) was initially small and it is in the 'cone' of development that the amounts mushroom.

My point is not really that this is a valid approach for modelling human behaviour, but it would be interesting to quantify just what sort of domain space is involved.

And of course, there is still that big IF of how to model the environment, but probably we could again take a leaf out of evolution's book and not model ourselves a complex environment, but allow complexity arise through the interaction of a sufficient number of simple models.

I know in areas like game theory such long running interactive simulations are used, but I must do some research to see what sort of sizes of runs are involved, and how the addition of parameters increases them.

I would just be interested to know what for example would happen if took some of the most powerful computers today and ran a simulation of some simple interacting evolving structure for say a year?

Nature didn't have the guns, but it sure had the numbers. My question is - how many?!