At present, AI products are often marketed using terms such as “reasoning,” “judgment,” and “understanding,” and in some cases by framing these systems as functional equivalents to human roles such as assistants or junior staff. What's seldom spoken aloud is that these systems don't do any of those things the way a human thinks of them, so the ways they can fail come as a surprise to us.
This leads to more than just confusion, it's a consumer protection failure driven by marketing practices.
Advertising descriptions can convey a level of reliability, comprehension, and consistency that they do not demonstrate in practice. The implication that an AI system "comprehends" leads the user to rely on its comprehension. So when the system fails to deliver the user is unprepared to handle the consequences.
In part I believe this comes from a general belief that computers are reliable. We've all had bad days trying to connect a headset to a Zoom meeting, but overall you turn it on, type your email and send it without any hiccups. However the reason computers are reliable is the exact reason that AI is not. Computers insist on properly formatted input, and when that fails they stop working. So their failure is noisy and obvious. An AI model is a tool that interprets imprecise language. The input and output still look the same (I type stuff into a box/stuff gets typed back into a box for me to read) but now when I type something into a box, the model interprets it and gets text (right or wrong). Instead of stopping, it continues. The text may be incorrect and not tell you. You'll likely never get the famous "blue screen of death" from Microsoft windows because that's not the kind of error it produces.
Some argue that these systems are getting better all the time and can eliminate this risk. They're half-right. The systems are getting better all the time, but they actually intensify the risk. As systems become more useful, users place greater trust in them, and the impact of errors can become less visible but more consequential. Without clear and accurate descriptions of how these systems function and where their limits lie, this dynamic is likely to increase the risk of over-reliance rather than reduce it.
In my work, I consistently encounter both over- and under-inflated expectations of these systems, often stemming from the same underlying misrepresentations. We actually already have laws that should deal with this. Canada's Competition Bureau handles misleading claims in advertising. If we are to safely and effectively adopt AI as a technology, enforcement of accuracy in marketing is necessary to support the required trust.