It’s a brave thing to try to make AI accessible. This is a valiant attempt that doesn’t (in my opinion) carry it off.

The problem with any discussion of AI is in the language, the inevitable tendency that the work “intelligence” brings to anthropomorphise the software. AI doesn’t get “confused”, doesn’t acquire “experience” – at least, not as these words suggest.

I think the biggest omission, however, is in what isn’t said: that AI is built on the assumption that the future will be like the past. While there are ways to avoid this – known as “lifelong learning”, which replace the initial training phase with an on-going re-learning from new examples – these remain in their infancy and haven’t yet translated into practice. Until they do, AI techniques are very much at the mercy of both their training set and the rate of evolution of their inputs.

2/5. Finished Wednesday 17 June, 2020.

(Originally published on Goodreads.)