[FoRK] A Theory of Products: Magic, Alchemy, Science... and Beyond?

Jeff Bone jbone at place.org
Sat Feb 6 12:14:33 PST 2010


Ken says...

> I disagree.

But of course you do. ;-)

However, you're not actually disagreeing with what I was asserting;   
you're talking about human beings and human organizations --- Stephen  
and I were discussing whether *machine* creatives would inevitably  
become stuck in local optima.

All of your statements about optimization vs. human nature are  
absolutely true in my experience, and I agree with them.  However,  
what's true for humans and human optimization and creativity isn't  
necessarily --- and generally *is not* --- true of *machine*  
optimization and creativity.  The techniques mentioned work *better*  
for machines than humans;  they don't generally have the hang ups  
humans do --- no limbic systems and all that, we humans just can't  
seem to get out of our own way most of the time --- so that  
"jolting" / getting unstuck is even easier for machines than it is for  
human beings.  That, in particular, is why I'm bullish about the  
prospects for machine "creativity" and problem-solution search.  Just  
one small problem for most sufficiently complex and interesting  
problems...

That problem, the biggest problem with machine creativity today, is  
creating a sufficiently rich yet dimensionally / computationally  
tractable representation of the fitness landscape for the problem you  
want to solve and for the shape of the individual solutions.  But  
significant progress is being made on that front in various places,  
including (an area I'm particularly interested in) mechanical system  
design automation...  (A second problem, particularly when the  
landscape is dynamic, is signal-to-noise in the data as represented,  
in the capture / sampling process, and often in the landscape /  
problem space itself.  To be fair, for some domains that is *THE*  
problem...)

BTW, my understanding of this stuff isn't intuitive, perhaps  
unfortunately.  I've gained a lot of first-hand exposure to the  
machine side of this over the last several years.  I'm sure there are  
folks for whom understanding the limits (and the *opportunities*) in  
the machine learning and intelligence space is a lot more intuitive  
than it is for me.  For my part, I am continually surprised to find my  
assumptions about such things to be exactly wrong, though usually in  
hindsight these "discoveries" make perfect sense.

Understanding human nature, on the other hand, is a lifetime endeavor  
--- and probably inherently constrained to the point of almost- 
impossibility in the general case.

  ;-)


$0.02, YMMV.



jb



More information about the FoRK mailing list