Search Here

Computer scientists were able to create a computer that is able perform a vision-related task slightly better than humans in an important moment for the development of artificial intelligence.

Computer software the scientists developed could more easily and accurately identify and recreate handwritten symbols after only seeing a few examples, or even just one, The New York Times reports. The study was published in the journal Science by a trio of researchers from Massachusetts Institute of Technology, New York University, and the University of Toronto.

The software is a breakthrough because it operates differently than the most advanced AI at the moment. Currently, AI relies on systems called deep neural networks, that layer the learning a computer does over time in order to create deeper bases of reference to enable faster and broader understanding of stimuli.

This new process, Bayesian Program Learning, does not require these layers and quickly generalizes from what it has learned without previous exposure to other information.

The researchers had their new software look at a system of 1,623 characters from 50 languages called Omniglot. The software then had to perform 5 different tasks.

The new software has wide-reaching implications, as AI is used in a variety of industrial settings around the world, including healthcare, journalism, defense, and countless others.