/* ---- Google Analytics Code Below */

Wednesday, November 21, 2018

Teaching Machines like Children

A long known approach and goal.   Allow machines to have very basic learning capabilities, let them experience the world,  and result in general intelligence.   Seems we have taken a small step in this direction, but still in narrow contexts.   Based on statistical methods. How can we go further?

MIT-developed AI learns language like a child does

It could understand the world without the usual hassles of teaching AI.
By Jon Fingas, @jonfingas in Engadget.

When you teach language to AI systems, you typically use annotations that describe how words work. But that's not practical in many cases -- even if everybody agrees on those annotations, they often take a lot of time to produce and can still seem unnatural. MIT's solution? Have AI learn like a child. Its researchers have developed a parser that imitates kids' learning processes by observing scenes and making connections.

The system studies captioned videos and learns to link words to objects and actions by determining the accuracy of a description. It turns the potential meanings into logical mathematical expressions, picking the expression that most closely represents what it thinks is going on. While the AI may start with a vast range of potential meanings and little idea as to what it's seeing, it will gradually whittle down the possibilities. Annotations can help speed the process, but the technology doesn't need annotations to learn.  ..."

No comments: