Sign In

Communications of the ACM

ACM TechNews

Microsoft's AI Is Learning to Write Code By Itself, Not Steal It

View as: Print Mobile App Share:
The algorithm does not take code from existing software.

Boston-based startup Gamalon has developed an algorithm that utilizes Bayesian program synthesis to develop solutions for problems in code.

Credit: Gamalon/MIT Technology Review

Researchers at Microsoft and the University of Cambridge in the U.K. have developed an algorithm that can create code to solve simple math problems by searching through potential code combinations for how a problem could be solved.

The DeepCoder algorithm's creators stress it does not steal code from existing software, but instead when asked to solve a problem, it predicts what code would have been applied toward the solution to similar problems it has encountered before, and in what order.

The team produced a domain-specific programming language, and DeepCoder also can sift through potential code for workable solutions.

"We're targeting the people who can't or don't want to code, but can specify what their problem is," says Microsoft Research's Marc Brockschmidt.

The team says its future ambitions for DeepCoder include understanding the subtleties of complete programming languages, and recognizing good code online.

View Full Article


Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA


No entries found

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