Robot Logic

This is a simple example of how strange "robot logic" may be.

One of the problems with general-purose AI (should it ever be identified and/or acknowleged :) is that it's quite different from human intelligence. Maybe so much different that it's hard to understand the way in which robots/AI's arrive at their answers.

Some simple AI's have "explain" s/w that allows them to expand on how they arrived at a particular answer. But, usually, the explanation generated in this way is just as obscure as the original answer that you're asking the AI to explain.

Here's a non-linear planner solving the simple "number puzzle".

The object is to sort the numbers from 1-15 (plus "empty") into order.

A human usually does it by concentrating on simple goals like "get number 1 into the first position, then get 2 into its position, then ....". But this particular s/w tries to solve all goals in parallel -- the efficient way.

Looking at the way it's solving the puzzle is surprising. It's hard to see why the s/w is making SOME moves while other moves seem quite obvious. But in the end, you can see it has been making moves to get all the numbers into their right positions as efficiently as it can (the solution it finds is still no guaranteed to be optimal).

Yes, there's quite a communication gap between AI and human I.

I'll later post an example of a "humanized" algorithm that tries to solve the same puzzle in a linear (mostly one goal at a time) way -- it's far easier to understand what it's doing as it goes along than this s/w.

Category:

Science & Technology

Tags:

• robots
• AI
• puzzle
• animation

[jag9998]

Wow, I was only thinking about getting the top numbers to work out first. I guess I should upgrade my RAM. XD

[kymhorsell]

@jag9998 I don't know if you read the text for "Robot Logic" and "Human Logic" (companion vid), but my point is efficient solving of problems is maybe NOT the way to go. People can't follow them. So a domestic robot (for example) probably should solve things the way people appear to (see Human Logic) so a robot owner can get a good mental model so there are no "surprises". Many domestic appliances are designed to work in an "understandable" way rather than a super-efficient way.

[jag9998]

@kymhorsell Sorry, I usually read the text. Never thought of that before, but I guess if we want to able to fit into our lives then we shouldn't stop at giving them anthropomorphic bodies. It would be a lot easier for us to teach something that thinks like we do, and it would greatly decrease the possibilty of it hating/killing us in the future. Still, it would be sad not to take the technology to its limit.

[Swiftrr]

Very nice example of how robots are smarter than us =) 5 stars for good effort of explaining!