When a child knows a thing, he automatically assumes everyone else does. This false belief is demonstrated by the famous Sally-Anne test, in which a child watches an object get moved from its original place, then is asked where someone who hasn't seen this move will guess where the object currently is. Children under 4 generally predict wrong, because they're not yet aware of separate consciousnesses other than their own. According to this an announcement from the Rensselaer Polytechnic Institute, researchers with the school have been able to simulate this phenomenon with an AI in Second Life, using "an automated theorem prover coupled with procedures for converting conversational English in Second Life into formal logic". There's more: when the child AI guesses incorrectly on RPI's virtual version of the Sally-Anne test, it learns from its mistake. This is potentially groundbreaking artificial intelligence. Hat tip: Joey Seiler. Image credit: RPI.
Update, 3/11: A YouTube version of the demo video is here.
pfffft... he's not so smart.
*pouts*
Posted by: Adz Childs | Tuesday, March 11, 2008 at 05:22 AM
In the future, we will compete with robots for jobs. Wait ... that is today.
Posted by: Dedric Mauriac | Tuesday, March 11, 2008 at 09:40 AM
The test wasn't performed properly in the video. It's lacking the memory and reality questions. Also, I'd like to know more about how the AI works.
Posted by: Myra Loveless | Tuesday, March 11, 2008 at 06:00 PM