Researchers at New York University have developed an artificial intelligence (AI) that can skillfully play a Battleship-like game by determining how to ask questions as efficiently as possible.
The AI perceives questions as miniature programs, which enables it to learn from only a few examples and to formulate its own questions on the basis of the knowledge it gains.
The researchers first had humans play the game and recorded the questions they asked, and then translated them into conceptual components. The team encoded the questions with a programming language and constructed a probabilistic model to ascertain which questions should return the most helpful information.
Their technique is more efficient at discovering smart questions to ask than most AI strategies, in which a computer is fed immense quantities of example data and then has to generate its own examples.
The researchers are investigating applying their methodology to the enhancement of dialogue systems such as chatbots.
From Technology Review
View Full Article - May Require Free Registration
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA
No entries found