Agent washing

Even more evidence that artificial intelligence is becoming more like human intelligence: we have to worry about which AIs are faking it.

We now have a concept called “agent washing.” If you haven’t heard of this, it means someone faked an AI’s resume. The company that sold you the program claimed the AI had agency — that it could do stuff and make decisions on its own — but what you really got was plain old software that needs guidance.

Washed AI programs are just like humans who don’t have a clue but pretend they can do their jobs.

A tech consulting company called Gartner has said that “only about 130 of the thousands of agentic AI vendors are real.”

The vendors are fictional? Does it matter if the fake vendors have agency? Did AI — with or without agency — write the Gartner report? Do I need AI to make some sense of any of this?

Regardless, there’s no end to the ingenuity of people who want to sell you things.

So what should law firms do about this?

Well, first off, you need some clarity about the problem. If your AI is making up case law and lying to judges, it may well have agency — just not the kind of agency you want.

On the other hand, if your AI never makes anything up, then it quite possibly doesn’t have agency.

So would you really rather have agency or a reliable robot? Why isn’t anyone talking about robot washing?

You wouldn’t think spotting AI with agency would be all that hard. If the AI can pick out the tractors among the cars and bicycles, that should be enough.

Apparently, it isn’t.

There’s some strange advice out there for determining whether the AI sold to you has been agent washed.

Forbes, for example, recently ran an article telling us that “if the system performs the same way no matter how many times it runs, that’s your sign you’ve been snared by a marketing gimmick.”

Yes, then you’ve got one of those darn reliable programs that doesn’t come up with different answers to the same problem.

I do understand that AI agents are supposed to learn and adapt and that can be a good thing for a regular business. Law may be a tad different. The best-researched most logical argument provided by a top-notch AI agent won’t help when you come up against a judge who can’t learn and adapt.

And what happens when opposing law firms both use AI agents? Would they come up with the same answers? Would they look at each other and explode? Will we need AI agent judges to referee?

The technological future in law is going to be so much fun.

Jury trial . You may have read about the University of North Carolina School of Law experiment using AI as jurors in a fake trial. If not, head to your computer and check out the news coverage — the variety of reactions is fascinating.

My favorite headline comes from Above the Law: “Law School Runs Mock Trial Before Jury of AI Chatbots as Dystopian Nightmare Accelerates.”

I’m pretty sure AI didn’t write that one.