[ExI] Hard Takeoff

[Samantha Atkins]

On Nov 21, 2010, at 1:20 PM, spike wrote:

> 
> The human brain has some inherent limitations, most specifically that we get
> tired, and there are not enough of us.  Consider top level chess.  Human
> elite players, top 100 in the world, can still play a competitive game
> against ordinary 100 dollar chess programs running on an ordinary 500 dollar
> laptop computer, but they must invest really intense concentration for about
> four hours, after which they are exhausted.  The computer on the other hand
> is immediately ready for another game, and can run two or more high quality
> games simultaneously, it can run day and night, it can replicate itself
> arbitrarily many times, all while the six billion strong human race is stuck
> right at around 100 or so players (and declining) capable of such
> concentration, at a rate of one game a day at most.  Silicon based recursive
> self-improvement is implemented by this ability to laser focus on the same
> problem over indefinite periods, in arbitrary numbers.

Great point!    I can beat the $100 dollar chess program quicker if I spend even more time probing and analyzing its weaknesses but the point is well taken.  

> 
> If we continue with the chess analogy, the human race has been playing the
> game in its current form for right at 500 years.  Many very focused players
> have dedicated their lives to this tragically wasteful preoccupation, and
> recorded their findings.  Chess playing software can re-derive that
> half-millennium of chess theory and surpass it in just a few days, or
> shorter if the task is spread over a sufficient number of processors.

If we knew how to add the right kind of learning to the chess playing programs (which are largely more brute force today) then yes, they could easily derive this knowledge.  Except chess playing programs do not generally look at the problem at all the same way humans do so the result would likely not be very useful for training would be human chess masters. 

- s