Getting machines to see more like humans.

Computer science is making progress in reducing AI’s propensity to make seemingly random mistakes in image recognition. Whilst the stories of such incidents can be funny, the potential for disasters is all too real. Think medical diagnostics. AI picks up on certain features in certain images, which we humans do not perceive, and sometimes draws the wrong conclusions. At MIT, they have found that it is possible to train recognition algorithms not to take such features into account. As a result, image recognition stats improve substantially. It is probably not the whole answer to the problem: understanding context better might also have a part to play. But it is a good start.

Link to article:

You may also like to browse other AI articles: