FTC Warns Technicians: ‘Keep Your AI Claims Down’

Image credit: Andriy Onufrienko/Getty Images

The FTC just announced an all-new division to tackle technology’s “snake oil,” giving it another shot past the overzealous industry bow with a cheeky warning to “hold back on AI claims.” Sent.

I wrote a while ago (5 years ago) that “AI-powered” is the nonsensical equivalent of “all-natural”, but we’re progressing beyond cockiness. Almost every product out there seems to claim to implement some form of AI, but very few explain in detail how it works and how it works. Even fewer can explain exactly why.

The FTC doesn’t like it. Whatever someone means when they say “powered by artificial intelligence,” or a version of it, “one thing is for sure, it’s a marketing term,” the agency wrote. “And he said at the FTC, one thing we know about hot marketing terms is that some advertisers just can’t stop abusing them.”

Everyone says AI is reinventing everything, but a TED Talk does one thing. It is quite another to claim it as an official part of your product. I would like to.

So if your product uses AI, or your marketing team claims to use AI, the FTC asks you to consider:

• Am I exaggerating what AI products can do? If you’re making sci-fi claims that your product can’t back up, like reading emotions, boosting productivity, or predicting behavior, you might want to tone down.
• Are you promising that AI products are better than non-AI products? Sure, you can make weird claims like “4 out of 5 dentists prefer AI-powered toothbrushes,” but it’s better to put all four on the record. Claiming AI superiority requires evidence.
• Are you aware of the risks? “Reasonably foreseeable risks and consequences” sounds a bit vague, but lawyers can help you understand why limits should not be exceeded here. Your product will fail if certain people use it because you haven’t tried it, or if its results are biased because the dataset is poorly constructed. You can’t say it’s not responsible because it’s a ‘black box’ that you don’t understand or don’t know how to test,” the FTC adds. Do you provide instead of providing?
• Does your product actually use AI at all? As I pointed out a long time ago, one engineer used an ML-based tool to optimize the curve, which led to claims that something was “AI-powered” or that the product used AI. Not that it’s true, but for many people AI drops mean a whole bucket full of it. The FTC believes otherwise.

“If these claims are not supported, we don’t need machines to predict what the FTC will do,” it ominously concludes.

Agencies have already published common-sense guidelines for AI claims in 2021 (there were many “detect and predict COVID” guidelines at the time), so direct questions to that document, including citations and precedents. I’m here.