How Not to Pass a Law: The EU's AI Act
Oh no, the Nazis are coming! I guess the way to stop them is to act very clueless, very quickly!
Hamburg, 11 February 2024 Filed under: Technology, Politics
After reaching an agreement on the proposed text of the legislation a week ago, the European Union is now set to pass its AI Act. This would be the first ever instance of a government passing wide ranging regulation concerned with the kind of technology people usually refer to by the name “artificial intelligence”. This legislation is a disaster.
Nomen est Omen
The problems with this law go from its intrinsic concept all the way through its implementation details to the way it is being passed politically. That it is called the AI Act is the first of many indicators of problems. An “act” is usually a piece of legislation that regulates specific technical and implementation details, whereas a “law” provides a general framework for desired behaviour in a society. Generally speaking, a law specifies how things should be, and an act how to do something correctly.
What society (in the EU and elsewhere) needs right now is an AI law. We need a general understanding of what “artificial intelligence” is, what it can do and what it should do. Once we are at that point — a point that most people, including the EU politicians passing these laws, are very far removed from right now — we can start worrying about implementation details. In effect, the EU is putting the cart before the horse. These people are legislating how a technology can and cannot be used, before even understanding what the technology even is. This is like deciding on how a drug can legally be used before having any idea what the drug even does.
Understanding “Artificial Intelligence”
By now, I’ve written quite a lot about how most people have no idea what AI even is. This is mostly the fault of journalists, who are supposed to explain this thing to people (including EU politicians), but are generally doing a horrible job, because they themselves do not understand the technology. Partly because they have been mislead by the propaganda — or advertisement, if you want to call it that — of the companies selling AI products, and partly because they don‘t have the time or patience to research this topic properly.
People have the idea that “artificial intelligence” is new, revolutionary technology. They almost equate it to magic sometimes. And they also think that there is actually intelligence involved in the process. All off this is wrong.
Artificial intelligence research has been going on since the ‘50s. The field has experienced a continued cycle of hype, investment and subsequent disappointment in the technology. This has occurred so frequently, in fact, that there is a dedicated term for this phenomenon: it is called the AI winter. Some basic research into this topic tells us that many of the concepts being bandied about today are far from new and many of the ideas and implementations heralded today as “revolutionary” are anything but. That is not to say that AI research hasn’t made incredible progress in the last decade or so. But this has been the result of a continuous process. The feeling that there's anything new or particularly revolutionary going on simply stems from not having paid attention to this process before now.
If you start looking into AI research as a scientific field that has been developing for over 70 years now, it also becomes obvious that the AI products of today are about as magical or surprising as, say, the development of the CPU. Which also sprang from decades of research that went on, unnoticed by the general public, beforehand. In technology, that’s always how it goes. There’s lots of boring background work going on for ages, which suddenly culminates into something cool that surprises the public and then gets marketed by Silicon Valley types basically as bottled lightning.
Like any other technology, things like machine learning and neural networks get quickly de-mystified once you start actually looking into how they work. AI can mostly be explained as large-scale application of statistics on huge amounts of data. Some of the techniques are quite creative, but they aren't magical. On the contrary, in hindsight, they seem to be a rather inevitable offshoot of the big data idea. In many ways, AI seems to be the long promised, long overdue, first proper use of big data. Finally, these humongous amounts of information collected on everything, and everyone, are producing some results.
“Artificial intelligence” doesn’t possess any kind of actual intelligence — this is something else that becomes obvious once you start understanding what AI actually is, beyond the buzzwords and sales pitches. It should also be obvious to anyone who’s studied what the term intelligence means. Whether that be from a biological, medical or even philosophical standpoint. We barely have a firm scientific grasp on what actually constitutes intelligence — and how it differs between species. We understand so little about our own brain, and the neurons and other cells that constitute our nervous system, that brain surgery, even today, involves a large amount of trial and error. New types of neurons are still being discovered and we don’t know how some of these work at all. And even when we understand individual cells, we have no scientifically reproducible concept of how all of these neurons interact to create what we call “intelligence”.
If you understand this, it becomes obvious that “artificial intelligence” as a term is quite bullshit. How can you artificially reproduce something that you don’t understand? We are like a cargo cult that once saw an automobile and then decides to built their own out of wood without understanding modern manufacturing methods, materials or principles like the internal combustion engine. We might call this thing we built a “car”, but that doesn’t mean it will work like one.
Now, if we started by calling this thing we built “large-scale statistical models” or something like that, people would understand its limitations and impact on society better. But alas, “artificial intelligence” is a better term if your goal is to produce boardroom hard-ons and make a lot of money in a short amount of time.
Clueless is an Understatement
I don´t particularly blame the Silicon Valley types for hyping the shit they’re trying to sell. That’s how business works. But the journalists writing about these things should know better. Sadly, most of them don’t. Which in turn means the public will be dragged into their follies. And this includes the politicians in Brussels, who then start thinking up ridiculous laws like the AI Act. Almost every single independent expert in the field who’s analyses I’ve read, or who I’ve talked to privately, agrees that this law is misguided at best and completely batshit insane at worst. They all seem to agree that the politicians in question have little idea of what they are regulating. Even parts of the press have now caught wind of this, but somehow that doesn’t prompt my colleagues to call for a stop to this insanity.
The only people who seem to have somewhat gotten their way with this legislation seem to be lobbyists from European AI startups. And even some of those people aren’t completely happy with the finished law. Still, lobbyists influencing EU legislation that might turn out disastrous for the general public, because politicians are clueless? Why does that sound familiar? Ah yes…
The more of this shit goes on, the more I am starting to think Brexit wasn’t actually such a horrible idea as I first thought.
How exactly is it a fringe opinion these days that politicians should understand what they are passing laws about? You’d think this was common sense. Why isn’t the press calling this out left and right? Why do you have to go on Substack to read this kind of opinion at all?
Is a Bad Law Preferable to No Law at All?
Well, part of the answer becomes clear when you look at how this law was passed. You see, there’s an election for the European Parliament this summer. And the powers that be are under assault. Everywhere in Europe, populist movements are on the rise because a large number of voters are dissatisfied with the current governments and their decisions. Gee, I wonder why?
Turns out that if you have no idea what you are doing when you’re passing laws, people will soon or later start to notice. And then they will eventually get tired of your bullshit and start to vote you out.
That is exactly what the currently governing coalition in the European Parliament is worried about. In their words: The populist right-wing Nazis are coming!!! I mean, I don´t like many of the ideas that these people are advancing either. But it seems pretty clear to me that a lot of this unhappiness stems from laws like the one I am writing about here. And from politicians who have no clue what they are doing, but who are doing it anyway. The examples are everywhere, from misguided foreign policy to politicians and judges not understanding the technology they are ruling on to idiotic reforms of parts of the criminal justice system.
So what does the EU Parliament do? Faced with the threat of being voted out of office as a result for passing stupid laws, they decided to quickly ram the AI Act through before that can happen. So quickly in fact, that some of the IT and legal experts who were consulted on the law say the process didn’t even allow for them to read all of the changes in the legal text in time to debate them. Never mind understanding these changes. In an interview with heise online, an expert working with conservative MEP Axel Voss on the consultations of the law called the situation surrounding the process “chaotic”. The European Commission was apparently changing the text of the law constantly, giving experts only minutes to review and comment on changes.
Can you imagine a worse way to decide on a far-reaching, revolutionary law with consequences on all of society in Europe? I can’t. At least not in a democracy. And what makes it all the much worse is that this law concerns highly complex technology that only a few people understand at all. An obvious recipe for disaster.
How hard is it to understand that this kind of rushed, idiotic decision-making process is exactly what caused the political backlash you are so afraid of in the first place? The prevailing wisdom in Brussels seems to be that passing a bad law is better than passing no law at all. I would suggest that it’s probably the other way around. Here’s to hoping that reaping what they sowed will teach these politicians the error of their ways. But somehow, I don’t think it will.