Post-Singularity Science Has Arrived – sort of

George Rebane

Machines are being developed that can analyze realworld observables (think of it as data from experiments with real things) and come up with not only the mathematics that describes the data, but also the mathematics that describes the underlying principles at work.  By ‘underlying principles’ I mean things like conservation laws (energy, momentum, information, …) that underpin the complex processes which generate the observed data.  You might say ‘what the hell, Man has been doing that for a few centuries now’, what’s the big deal?

The big deal is that humans know how to do that very slowly with relationships that have few variables and few constants prescribing their action.  (Think of the law of gravity F = G*m1*m2/r^2 which has three variables – m1, m2, r – and one constant G to describe the gravitational force between two masses separated by a distance.)  The real world is full of very complex processes that have oodles of variables and their related constants.  And underneath these processes may be entirely new principles and laws that are part of the fabric of our universe.  The hot flash here is that machines have now demonstrated that they can abstract existing laws from data much faster than humans can, and already distill out some new ones that we didn’t even know were there.

Here I want to acquaint you with an evolutionary algorithm called Eureqa.  This is a software system that is now mature enough for broad use by anyone who can master the care and feeding of the beast.  And hundreds of scientists and professionals ranging from micro-biologists to financial engineers are using Eureqa to discover things that they will share, and some more hush-hush stuff that they will tell no one.  For the near term, this points to a direction that man and machine will work together in the next decade or two to do science and engineering.  If you want to join them, go to this website operated by Cornell University computer science.

I described evolutionary algorithms here , and there is another good explanation from New Scientist here.  An excellent survey article on Eureqa, from which I filched the nearby graphic, is here.

My own familiarity with evolutionary algorithms (EAs) dates back to the days of my doctoral dissertation which involved asking a very simple question that required a very complex computational approach involving an EA to get its very simple answer.  The question is ‘what should I do until I know what I must do?’.  The answer required the invention and a solution of the Stochastic Anticipative Knapsack problem.  I say ‘a solution’ because the SAK problem can be so complex that many different methods to solve it may exist, and the one I came up with may not be the best.  But I will say this, I’ll bet you a real fancy dinner that all SAK solutions will center on EAs.

Today I am involved in an extremely interesting financial engineering project about which I hope to be able to report in the near future.  This project involves the modeling and managing of financial risk tolerance for actual human investors.  And it looks like EAs will again be part of my life.  The thought of it alone evokes a thrill chill up my back, because when you work with a GA machine, the feeling you often get is that the damn thing is alive, and you know that it can discover things you can’t.

Today we are still at a stage where a partnership is required between man and machine to probe together these unknown worlds.  I’m not sure how long this little interregnum will last before the machines will venture alone into such new conceptual and intellectual worlds, and then return to us with simplified answers that we will treat like the baubles for which the Indians traded Manhattan.

Comments

[Aaron Klein]

"like the baubles for which the Indians traded Manhattan"

Now that got a literal "laugh out loud" from me!

This is an exciting project, and I can't wait to see it launched...