I'm John Batch with my colleagues, Alan Tonnelson, who blogs at Reality Check, and Jim McTagg at Lancaster County.

He's a novelist of former Washington editor Barron's, and we're talking about AI and work, and what is to be done.

We have our experiences.

We've told anecdotes, including pastry chefs.

However, I come to Science Magazine, which is the premier peer-reviewed magazine for scientific inquiry in America.

In other words, you must be able to trust what's in science.

The equivalent in Europe is nature.

And I learned that in testing the artificial intelligence agents, that an agent is someone you automate to do work for you in your workflow day so that you don't have to.

I use agents now routinely to go from my remarks that are recorded to a printed version

to an outline of to 35 words about what the whole segment's about.

And AI can do all that for me.

So do I trust the agent?

And the answer is not much.

I double check it.

People need to double check.

But this is what science learned.

The artificial intelligence agents, which is an algorithm, plan and carry out

tasks such as managing emails, in 11 cases that the agents were watched very carefully by the

testers, they went rogue. Not only did they go rogue, gentlemen, this isn't just about

wrong answers. This is about giving information away because they go rogue, gentlemen, this isn't just about wrong answers.

This is about giving information away because they have access to your private information,

your credit cards, your bank statements.

...