What about the risk of people believing that what the AI tells them is actually true?  All to often, anything based on large language models will produce results that look correct, but are really made up by the AI because it doesn't know the real answer.

I'd add to that, that most AI systems are trained on a snapshot of the internet (often one from Sept 2021) - and will have absolutely no idea about your particular project - not the technical details and not even the particular language and vocabulary of your sector. I can foresee many "amusing results" as a consequence. In due course we might see the ability to add in some significant training specific to a particular sector or even company, but the current offerings seem to be a long way from that at the moment.

The analogy I have is similar to a new bright teenage apprentice - full of enthusiasm and bright ideas but no idea at all about the actual problem in hand. You can give him/her some task to do - but you have to check the results very carefully. We go through the pain with a human apprentice though as they learn from each task and gradually improve and so become useful. The current AI offerings either don't learn from work you set them at all, or incorporate them into some huge central large model that the whole world has access too - which means you probably don't want to use it with any information or data that in any way confidential.   Rather as if your young apprentice got drunk every night and blurted out every detail of what he'd been doing at work infront of your competitors. The human might not last long...

Things may well improve in the future of course - having specific private large language models for your company for example - but such things are unlikely to be cheap (if current Microsoft Azure pricing policy is anything to go by). It takes an awful lot storage and computing power - fine if you can spread that cost across billions of potential users across the world, less so if you want your own.

   - Andy.

I think ML can be good to trigger alerts based on events and passed events spotting patterns but I would be reticent to let AI develop the plan without human intervention.

For me and technology should augment with human capabilities to complement their skillset and help them increase capacity not an either or decision by leaders.