AI is learning to forget “catastrophic forgetting”

Sep 06, 2018, 7:31 AM EDT
(Source: ITU Pictures/flickr)
(Source: ITU Pictures/flickr)

A common limitation with artificial intelligence systems is “catastrophic forgetting,” their tendency to forget previously learned responses on being fed with new information. The behavior keeps AI learning from its surroundings, which means teaching them something like a past lesson would require starting all over again from the scratch.

Irina Higgins, a senior research scientist at Google DeepMind, seems to have taken the first steps towards overcoming this problem, notes Futurism. Higgins has developed an AI system that can apply its existing knowledge to a new environment and learn new things without flushing out what it already has.

The ability to get rid of “catastrophic forgetting” marks a beginning towards building artificial general intelligence, a broad spectrum of empathetic and imaginative AI as portrayed in films.