Google recently rolled out an experimental search feature, using AI to summarise results. And while it's generally very effective, sometimes it recommended putting glue in your pizza base.

Legal research tools like Westlaw have implemented AI into their products, but a recent Stanford research paper found that up to a third of the results returned by the new tools were hallucinations.

These powerful tools are becoming increasingly common, but, like any tool, they have their strengths and weaknesses.

Dr Ali Intezari and Professor Toby Walsh examine how AI should they used, and what happens when we ask them to do more than we should?