Communications of the ACM
Can Winograd Schemas Replace Turing Test For Defining Human-Level Ai?
Credit: Getty Images
Earlier this year a chatbot dubbed Eugene Goostman "beat" a Turing test for artificial intelligence (AI) organized as part of a contest put on by a U.K. university. Many observers and AI experts think the event does more to demonstrate how flawed a tool the Turing test is for measuring artificial intelligence than it does the sophistication of the chatbot, which was able to convince a significant portion of its questioners that it was human by claiming to be a teenager from the Ukraine that spoke English as a second language. This pretense enabled the questioners to excuse the chatbot's errors and tangents as the eccentricities of a young person speaking a foreign language.
A new proposed method of testing artificial intelligence are so-called Winograd schemas, which present an AI system with an ambiguously worded sentence, such as "The trophy doesn't fit in the brown suitcase because it is too big," and asks the AI to identify the ambiguous pronouns referent, something that is intuitive to humans because of their depth of knowledge and understanding about the relationships between the objects or people being described.
A new AI contest sponsored by Nuance Communications and CommonsenseReasoning.org is offering a $25,000 prize to an AI that can successfully answer Winograd schemas.
From IEEE Spectrum
View Full Article
Abstracts Copyright © 2014 Information Inc., Bethesda, Maryland, USA
No entries found