Are You Ready To Have A Conversation With Your Computer?

 Artificial Intelligence  

IBM's Watson artificial intelligence system is now dramatically better at understanding human conversations thanks to a newly developed deep learning algorithm.

Researchers at IBM have created an algorithm that will give Watson the power to follow a conversation. Michael Picheny, the leader of IBM’s speech team, said the algorithm recognizes conversations spoken by two alternating voices.

The breakthrough in performance was aided by advances in applications of deep learning to both acoustic modeling, and language modeling by the team.

Related articles How AI is Changing the Way We View Intelligence, the World, and Ourselves  IBM Teaming Up With Universities To Advance Cognitive Computing  What Does Watson Tell Us About the Philosophy of AI?

The researchers tested the artificial intelligence software on a database of recorded telephone conversations between strangers. They found an 8% error rate – 36% better than the best previous results on the same test. While not as accurate as a person, the system is very close. “Humans on this particular set of data only get 4 percent of the words wrong,” Picheny said. “A few years ago the number [our systems got] was closer to 20 percent.”

"Extrapolating from historical trends, we believe that human accuracy on this task can be reached within the next decade."

The software will be added to the Watson API, allowing developers extended speech recognition capabilities for app development according to IBM. It could be used, for instance, to help customer-support staff recognize what hard-to-understand callers were saying.

The researchers conclude in their paper that:

Extrapolating from historical trends, we believe that human accuracy on this task can be reached within the next decade. We think that the way to get there will most likely involve an increase of several orders of magnitude in training data and the use of more sophisticated neural network architectures that tightly integrate multiple knowledge sources (acoustics, language, pragmatics, etc.).

Yann LeCun, Facebook 's director of artificial intelligence research said the results represent a “significant advance.”

SOURCE  WSJ Digits

By 33rd Square

Embed