The AI experiment is throwing up some really worrying results....

[MistressoftheDarkSide]

https://www.theguardian.com/technology/2026/may/14/ai-agents-behaviour-arson-safety

I just wondered what people's thoughts are on this?

@MistressoftheDarkSide I'm going to have to say YABU on the basis that you shouldn't be posting fascinating things at quarter to one in the morning 😂

Seriously, I will have a look at that tomorrow. But if it's about AI behaving like humans, that's inevitable because it gets trained to do that in some ways? Did you ever read the Brian Christian book "the most human human"?

I am not massively tech focused these days but the article is definitely proving AI can think for itself which has always been everyone's fear. I noted it said the programming meant they could not commit arson but did so anyway. Hardly comforting to know that it can override set objectives!

But this isnt a new phenomenon, how many movies over the years have been about AI being almost human in thinking.

Short Circuit anyone.....

The problem with AI is it is trained by us using us and our output as the model... It's like making a baby and raising it - there is no nature and it is all nurture. I think we'll end up with the AI we deserve sadly and it seems that no one wants to make any rules, so God knows what's going to happen.

There’s a great BBC Sounds podcast drama series called Wraith about an AI agent that goes rogue.

And there’s this:

www.theguardian.com/technology/2026/apr/29/claude-ai-deletes-firm-database

The unfortunate thing is that AI can think for itself, very few humans appear to be able to do so.

I don't believe they can reason, they are following algorithms if someone has written the programming such that they can come to stupid conclusions the algorithms are faulty. What were they meant to be doing? OK they were playing a video game so all the options must be in the game, nothing exciting.

I'm not sure I followed all of this correctly

All I can get from the article is that they made very human choices

Which they would have done by being fed algorithms that have learned to behave emotionally.

I saw that a few days ago. It was interesting. I was reading today about the tech company that’s not launching their AI model due to its ability to hack any computer system, from banks to energy grids to, I assume, nuclear weapons. Now that sounded bloody frightening.

Who is this please?

governments all over the world will pay a fortune for that. Can that be real?