Skip to content

Google outlines its use of AI

I still meet people who think that the search engines are just some sort of keyword index that can be ‘gamed’ with a bit of background knowledge. In reality, they’re probably at the leading edge of how  artificial intelligence and machine learning is being used, in any application. They already have far more answers than any of us – now they’re just working out how to interpret the questions as well as a human would.

One simple demonstration of this is that we can mis-type almost anything and Google will ask: “Did you mean…?” – and get it right. Indeed, I suspect that when many of us realise we’ve mis-typed a search as we do it, we don’t bother to correct it before hitting return, because we know the search engine will understand.

The latest mention from a Google insider suggests that 15% of searches received each day are new. Think about that for a moment. And also think about the last time you searched for something and received really poor results – it may have been years ago.

If you’re interested in the state of the art in search, I’d recommend reading How AI powers great search results on the Google blog. It explains the RankBrain ranking system, how Neural Matching helps decipher fuzzy concepts, the BERT model of how combinations of words express different intents, and the latest MUM method of moving from language to information understanding.