Artificial intelligence (AI) researchers want to develop more comprehensive exams to test their systems. For example, Facebook wants to create a digital assistant that can participate in a real dialogue with humans, and the company has created 20 tasks, which get progressively harder, to test the system.
Any potential AI system must pass all of them if it is to develop true intelligence, according to the Facebook researchers. Each task involves short descriptions followed by some questions, with the harder examples including determining whether one object could fit inside another, or why a person might act a certain way.
"We wanted tasks that any human who can read can answer," says Facebook researcher Jason Weston.
The researchers used the exam to test several algorithms, and found none received a perfect score.
Allen Institute for Artificial Intelligence researcher Peter Clark says this type of testing is problematic because the AI does not need to understand to which real-world objects the words relate.
Facebook wants to use the AI technology to help filter news feeds. "People have a limited amount of time to spend on Facebook, so we have to curate that somehow," says Facebook AI research director Yann LeCun. "For that you need to understand content and you need to understand people."
From New Scientist
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA