Building responsible AI models: The argument for less data

PERSPECTIVE © FRANCE 24

When it comes to building complex artificial intelligence (AI) models, received wisdom has always favoured bigger datasets. After all, the more data an AI model is exposed to, the more it can "learn". But Abeba Birhane, a cognitive scientist and Senior Fellow in Trustworthy AI at the Mozilla Foundation, warns against this approach. She says that by casting the net as wide as possible, data invariably contains content from parts of the internet we'd rather not be exposed to, in turn creating AI models that become racist, sexist and otherwise biased. She joined us for Perspective to tell us more.