Friday, December 30, 2016

This is what it is like ...

My experience of research was in the 1980s, when I was delving into the mysteries of intelligence, via AI. The process model was this:

  • Take your most acute/insightful intuition
  • Find a simple-enough formal model/example
  • Analyse/program the model
  • Show some surprising/interesting results
  • Write/publish your paper.

AI at the time presented a superfluity of interesting problems - truly it was bliss to be a researcher. Only later did I realise that the most profound questions ('What evolutionary pressures selected for natural, human intelligence?') led into a conceptual swamp, where the right questions could not even be identified, let alone solved.

In physics, subject to centuries of theoretical and mathematical development, the situation facing the new postgraduate researcher is overwhelming and intimidating. As Bob Henderson observes,
"What we’d created is called a “toy model”: an exact solution to an approximate version of an actual problem. This, I learned, is what becomes of a colossal conundrum like quantum gravity after 70-plus years of failed attempts to solve it. All the frontal attacks and obvious ideas have been tried. Every imaginable path has hit a dead end. Therefore researchers retreat, set up camp, and start building tools to help them attempt more indirect routes. Toy models are such tools. Rajeev’s river almost certainly didn’t run to The Grail. The hope was that some side stream of one of its many tributaries (Virasoro, Yamabe …) would.

"Actually that was my hope, not Rajeev’s. Rajeev, I believe, just liked doing the math. The thing was a puzzle he could solve, so he solved it. For him that was enough.

"I, of course, had my sights set on bigger game."
Bob Henderson recounts an insightful, often tragic tale of hope and disillusion. Beautifully written to boot. Give it a try.

Via Peter Woit.

1 comment:

  1. "AI ... where the right questions could not be identified..." - it seems that this implies a similarity with this aspect of Theoretical Physics Research as indicated in the quoted story.

    You might want to study Fiber Bundles too, which are common in the more mathematical side of Theoretical Physics.

    Meanwhile matters related to the Penrose "non-algorithmic AI" issues provides a kind of focus of a mathematical approach to AI (or at least Cognitive Science). Thus the basic mathematical question in Cognitive Science might be "is (human) Cognition actually algorithmic?". If yes - prove it (despite Penrose); if no - find a modified structure that might work. In the latter case AI refers to the computable approximations to Cognition. Penrose sees a possible link between all this and Quantum Gravity remember - so you can have two difficult research projects for the price of one!


Comments are moderated. Keep it polite and no gratuitous links to your business website - we're not a billboard here.