Is This Really Artificial Intelligence? Three Questions to Test a Claim

On 18 March 2024 the US Securities and Exchange Commission announced settlements with two investment advisers over false or misleading statements about their use of artificial intelligence, with penalties totalling $400,000. The firms were not accused of having weak AI, but of having AI that was not what they said it was. The practical question comes before the philosophical one: when may a product carry this name?

Why Is the Name So Hard to Pin Down?

In our article "What Is Artificial Intelligence, Really? A Plain Explanation Without the Hype" we dealt with the limits of the concept and the common misreadings. Here we ask the buyer's and the regulator's question: what is the minimum that makes a system "artificial intelligence" in legal usage?

Two close definitions answer it. The OECD, in its version revised in 2023, describes an AI system as a machine-based system that, for explicit or implicit objectives, infers from the input it receives how to generate outputs such as predictions, content, recommendations or decisions, which can influence physical or virtual environments. Article 3 of the EU's AI Act nearly repeats this and adds that such a system operates with varying levels of autonomy and may adapt after deployment.

The key word in both is "infers". On our practical reading, automation alone is not intelligence: a spreadsheet that applies a rule a person wrote line by line does useful work, but it infers nothing by itself. The exact boundaries remain debated among specialists, so we should not accuse every program on that basis.

Three Questions Before You Believe the Label

What goes in and what comes out? If the seller cannot say plainly what the system receives and what it produces, the label is still a slogan.

Does it infer or merely execute? Ask what the system learned from data and what was written by hand. Both can coexist in one product, and there is no fault in that so long as the advertising is honest.

Who works behind the screen? In April 2025 the US Department of Justice accused the founder of a shopping app of leading investors to believe in automation with no human hand in it, whereas the app, according to press reports of the indictment, relied on hundreds of contractors in a call centre in the Philippines who completed purchases manually. These are allegations that no court has yet decided. Using people is not a fault in itself; concealing it is.

When a Description Becomes a Legal Matter

On 25 September 2024 the US Federal Trade Commission announced "Operation AI Comply", five cases concerning exaggerated claims about AI. One was DoNotPay, which marketed itself as the "world's first robot lawyer"; the Commission said it had not tested the service against human lawyers and employed none, and the company agreed to a $193,000 settlement and a ban on unsupported claims. In other cases, schemes selling "AI-powered" online stores were alleged to have promised passive income that did not materialise. All of these are allegations or proposed settlements, not final judgments.

The lesson is not confined to one country: regulators measure an advertisement against evidence, not against its label. That matters to every organisation buying "smart" services, including charities and non-profits that often contract on tight budgets.

What to Do Before Signing

• Ask for a trial on your own data and cases, not a recorded demonstration.
• Ask for the performance measure, how it was measured, and what happens when the system errs.
• Require in the contract a written description of what the tool does and what people do.
• Be wary of a flawless result presented without stated limits.

None of this means suspecting every product; the starting point is good faith, with verification. Honesty in describing goods is a principle of fair dealing, and whoever is honest in the description will find the three questions easy to answer.