Politicians and regulators fundamentally misunderstand AI as a category. Artificial intelligence is a process, not an object. It is not oil, shovels, or images. It is more like mathematical equations, programming languages, or speech. Artificial intelligence is a set of statistical methods used to turn information and energy into output, like new images or emails. It is a wide research area that includes simple methods a single software engineer could manually write in one night and billion-dollar models such as OpenAI’s ChatGPT.
Consequently, regulating or licensing “artificial intelligence” is like regulating statistics itself. It would be the equivalent of assigning government bureaucrats to going into each and every company where AI is used (virtually all of them in a few decades) and micromanaging what software they can download, what mathematical formulas they could write down, or what emails they could send. In other words, it is completely infeasible.
As Jim Pethokoukis put it in conversation with me, “If AI is used across the whole economy, then having a department manage AI is like having a department manage the economy”.
Within precedent, a targeted and consequence-based approach has been extremely successful. If a person writes software in C++ to hack or defraud another, that person is punished. There is no collective punishment doled out to every C++ programmer or the inventors of the C++ language. If a person uses email to scam another, that person is punished, not the email server or protocol inventor. The intuitive reaction to these statements is “of course, it would be devastating to punish programmers or email users as a whole”. That is the ubiquity AI tools will have within a decade.
Profound observations. They made me realize that if mathematics or natural languages (like English, as opposed to computer languages) were invented today, government regulators would regulate undoubtedly demand to control them to prevent them from being used for "unsafe" purposes. And such regulation would be a kind of universal control.
And yes, we are absolutely sleepwalking into Soviet-style "command" economy. And it may be inevitable. When an economic depression starts to happen (which is inevitable due to the national debt), Americans will almost certainly choose government control of the economy over the painful austerity of letting the real economy correct itself.
this does not take into account the much bigger problems of "AI used across the whole SOCIETY" (e.g. to influence politics) not just the economy, which is much less, and the smartness, and physical feasibility of spending on AI energy and raw resources that should be really spent to on other problems first. I wrote about this this very week: https://mfioretti.substack.com/p/the-only-thing-that-ai-tech-bros