The EU wants to regulate statistics

Using bad terminology like "artificial intelligence" is far more damaging than just confusing people

Yesterday I complained that “artificial intelligence” is a vague term used inconsistently. I made this introduction before going on to show the latest example of how far reaching the effects of bad terminology can be.

A week ago, the European Commission published its proposal for an Artificial Intelligence Act, which they proudly advertise as the “first ever legal framework on AI”. It has provisions such as prohibiting “an AI system that deploys subliminal techniques beyond a person’s consciousness in order to materially distort a person’s behaviour in a manner that causes or is likely to cause that person or another person physical or psychological harm.”

But what does the proposal mean with “AI system”? It means software developed with one or more of these techniques:

(a) Machine learning approaches, including supervised, unsupervised and reinforcement learning, using a wide variety of methods including deep learning;

(b) Logic- and knowledge-based approaches, including knowledge representation, inductive (logic) programming, knowledge bases, inference and deductive engines, (symbolic) reasoning and expert systems;

(c) Statistical approaches, Bayesian estimation, search and optimization methods.

To be fair, the proposal’s definition of an AI system also seems to require that the software can influence its environment; however it isn’t clear what this means. Does an early warning system for floods that automatically sends SMS to citizens influence its environment? What if it only turns on a big red light in a control centre and lets a human press the button that will send the SMS’s? Where do you draw the line of “influencing the environment”?

In any case, if you somehow cause a person to harm himself or another, isn’t that already illegal? Why does it matter if you used an “AI system” to do this? Doesn’t your car’s ABS system need to go through rigid inspection rules? Why does it matter whether it uses statistical approaches?

The proposal is more than 100 pages long. A huge amount of effort and money has been spent in its preparation, and it is dwarfed by the amount of effort and money that will be spent in the EU’s lawmaking process. If, eventually, it is passed, it will cause, like the GDPR, a huge amount of confusion and damage.

None of this would happen if, instead of the misleading “artificial intelligence”, we used “probability modelling”.