5
$\begingroup$

Stuart Russell and Peter Norvig pointed out 4 four possible goals to pursue in artificial intelligence: systems that think/act humanly/rationally.

What are the differences between an agent that thinks rationally and an agent that acts rationally?

$\endgroup$
1
  • $\begingroup$ just an interesting side note, but there is no well-agreed upon definition of rationality in the psychology literature $\endgroup$ Sep 2 '18 at 14:15
5
$\begingroup$

Maybe a good example to think about would be something like the Sphyx story. The wasp in the story appears to behave like a rational being: it seems to have a plan of action, it seems to be able to do advanced operations like counting, and it seems to execute the plan well. However, if you disrupt the wasp's plan, it becomes apparent that it is not thinking rationally. Instead, it has evolved a very complex behaviour that appears to include rational components, but is still just an instinct.

In the context of AI, consider a GOFAI system for planning like GraphPlan as opposed to a machine learning system for generating plans. While the former is a general-purpose algorithm for reasoning about planning problems, the latter is an input/output mapping that may mimic reasoning, or may be more "instinctual". Some AI'ers would say that the latter system is not really engaged in rational thought, while the former is. Both systems exhibit rational action, however.

Instinctive systems often work very well, and I'd say that the increasing effectiveness of machine learning approaches since that edition of R&N was published (almost 10 years ago now?) makes this a fuzzier distinction in practice than the book might suggest.

$\endgroup$
0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.