AI Episode 1: What is it?
A lot of concepts are thrown around in an attempt to define what is Artificial Intelligence. This could be quite confusing so to start our series on AI let's try to agree on a few terms.
We don't really know what it means for software to be intelligent. The Turing Test, the first attempt to define AI, boils down to this: if people think it is intelligent, then it is. So many problems have been documented about the Turing Test that it is rarely worth mentioning outside history books. A better definition is that for software to be considered AI, it needs to perform one or more cognitive tasks such as planning, extracting information from texts, images, audio or video, making predictions about the future, choosing, optimizing, generating speech, and so on.
In health, many tasks are cognitive and are therefore prime candidates to be automated or supported by AI. Other tasks, such as providing care and empathy are not.
In order to adopt AI in medicine, it is important not to lose sight of evidence. The same rigour used to adopt new therapies should be applied to software. Is it effective in performing the tasks it says it can perform? Is it safe? does it improve care?