AI, AI, oh!

This week we have seen some of the biggest names in tech – including Jensen Huang of Nvidia and Microsoft’s Satya Nadella – announce massive inward investments to make Britain an “AI superpower”. While we all know that Artificial Intelligence is the future – what sort of future will it be?

As someone who has written for a living for the last four decades, I’m no fan of AI services like Copilot and ChatGPT trying to “improve” my writing. However, given the quality of writing I’ve seen from supposed professionals, maybe I’m the wrong person to offer help to. I’m sorry Satya – but I’ve removed Copilot from my Microsoft subscription (so saving about a third of the cost too).

Earlier this week I became aware of a troubling article in the Washington Post about the accuracy of AI search provided by Google, ChatGPT, Grok et al. It found that the most accurate – Google AI Mode – was correct just 60.2% of the time. That’s two out of every five results that are wrong. Imagine if that was a map – you want to go to Manchester and 40% of the time it sends you to Leeds or Cardiff? Imagine if the BBC was only right 60% of the time? Or even The Sun?

However it is more insidious. If you ask a traditional search site – Google for instance – for advice on say a health issue or a legal matter, it would link you to a site that should be an expert on the matter. But AI search brings up an answer for you, which mean a massive proportion of people don’t then click through to the sites where the information comes from.

This creates two big problems:

1.       The people who are the experts don’t get the click through to their sites. I must declare an interest – I’m a trustee of a charity Family Rights Group that provides an advice service for families trying to stop kids being taken into care (among other things). We’re seeing a big drop in traffic and we’re hearing anecdotally that other charities are having the same issue;

2.       If the AI search is only 60% accurate at best – and 33.7% accurate for the worst (Meta AI) – then a big proportion of the population are being given duff info.

So before we give headlong into AI, should we not work out the accuracy and attribution problems or we will have a ill-informed future.