Re: software progress (RE: Hardware Progress: $319/GF)

From: Samantha Atkins (
Date: Thu Sep 05 2002 - 02:18:36 MDT

Ben Goertzel wrote:
>>Anyway the conversation reminded me of something that you guys on
>>this list
>>should know about... an advanced approach to CPU design which is not being
>>followed up by the hardware industry at all, but really should be. VERY
>>cool stuff... based on applying the TimeWarp algorithm from discrete event
>>simulation theory, on the chip level...
> As an exercise in mind expansion, try this. Read about the TimeWarp
> algorithm, and then try to imagine the psychology of an mind that used it!
> Basically, in the TimeWarp approach, computing is kept reversible, by
> (within each computing component) keeping information about previous states
> even when a new state is moved to. Computing is then done aggressively,
> with new computations being done *before* the information needed to support
> them is there, based on "best guesses". Then, if the guesses proved wrong,
> time is "rolled back" and the computation is done again with the correct
> information. There are some strong theorems showing why this is an optimal
> approach under pretty loose assumptions.
> It's sort of loosely like the backtracking algorithm in AI, but *massively
> parallel*...
> Imagine if each portion of your mind/brain could think speculatively, based
> on the speculative thoughts of other portions, etc. -- and then roll back to
> previous states precisely, as determined appropriate.
> The rollback would feel cool, huh?? ;-)

Probably. The mind/brain does think speculatively and "jump to
conclusions" before the data is in. It seems to have trouble
with the rollback though. It wastes too many cycles attempting
to find justification for its "best guess" leaps. At its
extreme we call this "rationalization". :-) It is worse than
that because those guesses are already part of the next cascade
and the processing is asynchronous. I don't know how big the
state information needed to really rollback in such an
environment would be.

- samantha

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:40 MDT