21. duben 2021 / Jan Klesla
Brussels rules for artificial intelligence: are we finally regulating ourselves to prosperity?
The robot must not harm a human being, must obey his command and protect himself from damage. The three laws of robotics, written by science fiction legend Isaac Asimov almost eighty years ago, are still alive today, when artificial intelligence is already very real. In the full version three sentences, 61 words, 320 characters that give clear and understandable rules. In contrast, the draft AI regulation, which the European Commission will present on Wednesday and escaped last week, has 82 pages. Everyone in the business environment certainly knows that if something is eight times longer than the short story where Asimov presented his three laws, it will probably not be clear and simple rules. Indeed, the French trio-commissioner Thierry Breton and his people have decided to regulate all the technologies collectively referred to as artificial intelligence, as something extremely dangerous, endangering the health, life and, above all, the human rights of all of us. He almost seems to want to stop those terminators who are going to enslave humanity.
Affix the CE mark to the AI system
The new regulation, which is directly effective in all states without further implementation into national law, works with three categories of rules. The first prohibits, for example, social credit systems, algorithmic staffing, as tested in China. That is perfectly fine, and even today, in the EU, it would be contrary to the existing protection of fundamental rights and privacy, so it is a useless provision. On the contrary, it opens the door to the controversial use of facial recognition by repressive forces, which has recently been encountered by, for example, the Prague City Police. In the second category, it lays down general rules according to which, for example, a chatbot must present itself, which is still rather funny, especially in non-English languages. One can also think that developers will have less motivation to overcome the so-called Turing test, when it will not be possible to commercially offer a chatbot that is unrecognizable from humans.
The main problem and most of the massive text is in the third part of the so-called high-risk applications. That is, those where fundamental rights could be endangered, for example when checking the creditworthiness of a financial institution's client. An example is probably the most successful Czech fintech Twisto, which builds its business model on better AI than banks. It doesn't matter that it is already subject to financial regulation and the CNB's strict supervision, and if the regulation passes, the team that has recently expanded from Prague to Warsaw will stop, redesign internal systems, add more people and publish details of its algorithm and competitive advantage.
Last but not least, go through the process of obtaining CE, which must have, for example, gasoline, explosives, cable cars, transformer stations, or medical devices. Slightly inadvertently, the regulation literally says that the CE mark must be affixed to an AI system, such as a neural network in the cloud. All of this can take a long time and cost a lot of money, the Commission's estimate that a company will cost less than the average salary of an average euro official a year to meet new obligations is a testament to its contact with the real world. Above all, the problem is that Brussels officials themselves want to add other products to this list at will, according to vague rules. Everyone with an AI system, including their users, will be afraid that someone will sleep poorly and get an idea of where to protect humanity.
The best gift for brexit
For three years now, the European Union has been struggling to catch up with the United States, the United Kingdom and China in the "AI race." The Juncker Commission has thus come up with an ambitious plan to support the development of artificial intelligence, and the current one seems to be achieving the exact opposite. It almost looks like she's throwing a towel in the ring, thinking that if we can't win, we'll at least stop the infiltration of foreign technology into the EU, like the evil American IT giants it particularly falls for. It will probably not bother large European corporations, such as the semi-state French Athos, which until recently was headed by EU Commissioner Breton.
The victims will be startups. In an environment of re-regulation and legal uncertainty, the best can go, for example, closest to London, where they will enter the market and then return to obtain the CE in the EU. If it's still worth it at all. The Commission came up with the help of Solomon for them - it tasked nation states, but at least it does not give any relief for small and medium-sized companies. In addition, it will be a huge challenge to find scarce AI experts today to do all the required testing and supervision. It can easily happen that only large EU countries, such as France and Germany, will have the capacity to do so.
One page is enough
The worst thing about the new all-encompassing AI regulation is that it is completely unnecessary. We have a very robust protection of fundamental rights in the EU, and the rules for consumer protection and privacy. That they work is also evidenced by the fact that all the examples on which the Commission bases the need for regulation took place outside the EU, mostly in the US and the UK, in Europe it would not pass, or no one would have thought of. The whole list thus looks more like excerpts from the American and British issues of the popular IT magazine Wired. At the same time, it would really be necessary to address real issues such as what liability will be paid for when Tesla crashes you into an autopilot. If necessary, adjust sectoral regulations, for example in finance. However, political points are harder to collect.
In the area of AI ethics, voluntary codes, to which companies subscribe themselves and voluntarily, would work best, which is an area where the team of our European Commissioner Věra Jourová did a great job. Simple and clear rules, written by AI developers and experts themselves, do not need philosophers, archaeologists or, God forbid, officials. If that doesn't work, a few pages, clear and comprehensible barriers in a dynamically evolving industry, not enough unnecessary heavy-duty regulation with the new Brussels office and a lot of money from taxpayers' pockets, which will become obsolete before going through several years of legislation. By the way, the most controversial use of AI in the military is explicitly excluded from the proposal, so those terminators are allowed to develop. And the ban on injuring a person is also explicitly lacking in it, follows from the legislation already in force today.