MIT Finds 'Grand Unified Theory of AI'

aftab14 writes "'What's brilliant about this (approach) is that it allows you to build a cognitive model in a much more straightforward and transparent way than you could do before,' says Nick Chater, a professor of cognitive and decision sciences at University College London. 'You can imagine all the things that a human knows, and trying to list those would just be an endless task, and it might even be an infinite task. But the magic trick is saying, "No, no, just tell me a few things," and then the brain — or in this case the Church system, hopefully somewhat analogous to the way the mind does it — can churn out, using its probabilistic calculation, all the consequences and inferences. And also, when you give the system new information, it can figure out the consequences of that.'"

Comments

Re:The real summary

[Trepidity]

Mostly, he or his university are just really good at overselling. There are dozens of attempts to combine something like probabilistic inference with something more like logical inference, many of which have associated languages, and it's not clear this one solves any of the problems they have any better.

This looks familiar

[Meditato]

I looked at the documentation of this "Church Programming language". Scheme and most other Lisp derivatives have been around longer and can do more. This is neither news nor a revolutionary discovery.

Re:The real summary

[Trepidity]

I should add that this is interesting research from a legitimate AI researcher, not some kooky fringe AI. I suspect it may have been his PR department more to blame than him, and his actual academic papers make no similarly overblown claims, and provide pretty fair positioning of how his work relates to existing work.

Re:Interesting Idea

[Chris Burke]

In an example, we're told the cassowary is a bird. Then we're told it can weigh almost 200 lbs. Okay. Now you're telling me that it might revise its guess as to whether or not it can fly? Come on! Am I the only person that can see that you've just given me an example where the program magically drums up the rule or probability based rule that "if something weighs almost 200 lbs it probably cannot fly"?

For fucks sake, it was just an example of the kind of inferences a logical rule system can make, not a dump of the AI's knowledge and successful inference databases. I mean you might as well complain that the example given was not written in Church and ergo not understandable by the AI whatsoever.

As the article explains, just not explicitly in the context of that example, it devises these rules from being fed information and using the probabilistic approach to figure out patterns and to infer rules, and that it does this better than other

So in the actual version of the Cassowary problem, you would have first fed it a bunch of data about other birds, their flying capabilities, and their weights. The AI would then look at the data, and infer based on the Emu and the Ostrich that heavy birds can't fly and light birds can, unless they're the mascots of open source operating systems (that was a joke). Then you tell it about the cassowary, but not whether or not it can fly, and it infers based on its rules that the cassowary probably can't fly.

In a sense it does "magically drum up the rule". Yes you still have to feed it data, but the point is that you do not have to manually specify every rule, because it can infer the rules from the data, and the create further inferences for those rules, combining the abilities of a rule-based system with the pattern-recognizing power of probabilistic systems.

So the point is it takes less training, and a relatively small amount of explicitly specified rules.