Thankfully minimal. I was very glad to see a firm-wide email saying "it should be obvious, but you should under no circumstances put any client data into any AI system, because that is a data breach under GDPR"...

Aside from the answer I gave about my current job... when I worked in a hospital, that was a suggestion to do a pilot scheme of using AI to prepare/type clinical correspondence, and then use medical secretaries just to 'tidy it up' before sending, without the secretaries having actually listened to the clinicians' dictations.

I asked the managers who seemed to be the decision makers about this pilot whether this was just fancy text-to-speech (i.e. they were paying AI prices for a version of something that had existed for years), or if it was genuinely AI. They did not seem to understand the question.

I asked instead whether they knew whether it was generative AI, and, if so, asked if there had been any consideration of potential clinical risk given the fact that generative AI can often just hallucinate things (thinking that it particularly does this when it thinks there is a pattern of what typically happens, but the whole point of going to a doctor is that something is not working in the typical way). They did not seem to know what the term generative AI meant, or to have heard of AI hallucination.

I asked whether the AI model used would retain the patient data to increase its own data set and effectively refine/retrain itself, and asked about how patients could opt out of their letters being prepared by the AI. I tried to explain that patients might want to opt out for a number of reasons, but one obvious reason was that any data retained to retrain the model would be an asset of financial value to the AI company, and patients might not want their data to financially benefit the AI company. They said there would be a chance to opt-out, but didn't really seem to understand the point about the patient medical data collectively having quite significant value to a company trying to sell a medical AI model, or understand how AI 'retrains' its future answers by increasing its dataset from information that's input.

Finally, I said that I could potentially see that there might be a use for AI in spotting patterns in scans or trends in test results quicker than humans can do, so I wasn't fully opposed to AI in medical settings, but I didn't understand how they could possibly get AI use for something that was currently done well by (fairly low-waged) humans approved as something that was ethically in line with the hospital's environmental policies generally, and in particular because of the health impacts on people living close to AI data centres. They didn't seem to know that there were any environmental downsides to AI.

I feel strongly that a lot of people buying into AI on behalf of their organisations are doing so without actually understanding the basics of what it is or the risks it can pose, because they're too scared of looking silly for not knowing all about it already or for looking old-fashioned if they don't get on board.