The Un-AI: Futurism Almost Gets It

*“Servotron Robot Allegiance. Join us or Die!”*

The Futurism web site is surprisingly skeptical of the generative “AI”/LLM industry, for a web site called “Futurism”.

But it’s not skeptical enough.

For example, go read today’s piece at Futurism with the probably-too-long title:

Bosses Are Becoming Obsessed With AI, Using It to Make Every Decision, Barraging Their Employees With Nonsensical ChatGPT Directives, and Even Asking It Who to Fire

It discusses the fall into drooling “AI” psychosis in several bosses via interviews with the employees, and wow do those bosses sound terrible.

And extremely foolish.

Though mostly terrible.

But the piece makes the usual modern-journalism error of trying to both-sides the issue:

Most of the people we spoke to noted that they do use AI in their work and personal lives, and believe it has value when used with caution and self-awareness.

And quotes several of the employees saying, basically, that it’s not the “AI”s fault; the bosses are just using it wrong.

This is nonsense.

This is just a variation on the error I wrote about 6 months ago in Artificial Stupidity: It Only Seems Right Because You Don’t Know It’s Wrong: the employees, the people who actually know what the work of the company is, can see that the boss’ “AI” is wrong because they actually know what the company’s work is.

But.

They can’t see when it makes an error when they’re using it because they don’t know what responses it should be giving to them.

If you only ask the chatbot questions you don’t know the answer to, then you are not in a position to evaluate whether it’s responses are any good.

No one has any good reason to think these systems will ever give reliably useful results.

The fact that they’re statistical language-forming machines certainly argues that the default expectation should be that their responses would only coincidentally be useful or accurately reflect the real world in any way.

And what little research has been done into chatbot accuracy does seem to support this: generative “AI” is not particularly accurate about the real world.

Which should be no surprise to anyone.

So thanks, Futurism, that was a good and horrifying read but it didn’t need the “these people are using it wrong, but we use AI correctly” bit at the end.

As the terms of service on the chatbot says: if you’re not using it “for entertainment purposes only”, you are actually using it wrong.