Sign In

Communications of the ACM

ACM TechNews

Google Tries to Make Machine Learning a Little More Human

View as: Print Mobile App Share:
Star Trek: The Next Generation characters Data (Brent Spiner), left, and Counselor Troi (Marina Sirtis).

Google research engineer Pete Warden said the company's current machine-learning algorithms are comparable to Star Trek: The Next Generation's emotionless android, Data, but the company wants them to be more like the empathetic Counselor Troi.

Credit: Paramount Television

Google CEO Sundar Pichai says advances in machine learning would soon affect every product and service the company provides, and the company's researchers discussed some of these advances in a briefing at its headquarters last week.

Research engineer Pete Warden used characters from the TV show "Star Trek: The Next Generation" to explain the company's progress, comparing Google's current machine-learning algorithms to the show's emotionless android, Data, and saying the company wants them to be more like the empathic Counselor Troi. For example, they want the company's computer-vision algorithms to be able to see a scene of a cooked turkey and plates taken in late November, and make the human leap that this must be a scene of a Thanksgiving dinner.

Another researcher, Maya Gupta, is working on a project to help machine-learning algorithms process outlying data points that could trip them up. For example, a person might look at a sample of images of houses of various values and conclude that value relates to size, but an outlier of a small house in a very expensive market might cause a machine-learning algorithm to conclude value correlates to some other factor, such as color.

From Technology Review
View Full Article


Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA


No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account