Table of contents: 1. A dystopian past. – 2. Dogmatism vs. Science. – 3. AI as an alien species. – 4. Other legal idiocies
- A dystopian past
The scene: Germany, at the end of the 19th century: a buoyant technical, scientific, and cultural atmosphere.
The actors: two engineers, Messrs. Benz and Daimler who have just perfectioned a very performing internal combustion engine which, in their expectations, will bring a revolution in industry, substituting old, cumbersome and manpower consuming, steam engines.
They proudly travel to Brussels to present their invention to the EU Commission (yes, it already existed!), expecting to receive a very warm welcome by their noble compatriot Princess Ursula von XY.
Instead, they are given the cold shoulder, and their enthusiasm is thoroughly questioned.
Have they considered the very high risk that their invention entails? The impact assessment that the EU has already conducted forecasts that in the following century more than a million people will die of car accidents in Europe; and even more will be killed by tanks, battleships and warplanes which use their invention. Furthermore, the Commission is unfavourably impressed by the non-compliance with the precautionary principle. The two inventors have in no way considered how the landscape will be devasted by motorways, and the permanent damage to the environment: CO2 emissions, global warming, climate change. To sum all this up, what is required before authorizing such devices is a comprehensive regulation that can protect Europe and its citizens from such dramatic consequences.
A few years later – we are already at the dawn of the 20th century – the EU comes up with the “Internal Combustion Engine Act”: 270 pages, 200 recitals, 100 articles, 12 annexes.
Messrs. Benz and Daimler try to read and understand it, but rapidly give up and emigrate to the US. Another German engineer, Mr. Diesel, is enjoined by the Independent Motor Authority to cease the development of a new engine which has not passed the “sandbox test” because considered highly polluting. A few European industrialists, Messrs. Citroen, Peugeot and Agnelli, who were ready to invest in the new technologies, find it much more prudent to devote their capitals to farming. At the New York Stock Exchange, the shares of the Ford Motor Co. skyrocket.
But who cares? European values have been reaffirmed and there is no doubt that horse and buggies are authentically “green” if only you think of the tons of manure you can recycle for organic agriculture.
- Dogmatism vs. Science
The EU approach to AI is the typical approach of a bureaucracy which in the name of very high-level principles wants to control scientific and technical development. Four centuries ago, Galileo Galilei’s astronomical discoveries were silenced by threatening excommunication and banishment for having challenged even higher level, theological, principles. But outside the reach of the retrograde Catholic Church, Galilei’s ideas thrived, bolstered maritime trade, astronomical discoveries of outer space and a general development of scientific reasoning and method.
The AI Act puts into place a similar attempt. Scientific development is accepted and acceptable only if it conforms to certain values set by the 21st century clergy, i.e. the EU institutions. What is lacking – today as four centuries ago – is the understanding of what science and technical development are about, what are their epistemology, what are their intrinsic values and how they collide with the dogma of regulating anything that appears on Earth and beyond.
Under certain aspects such resistance is understandable: AI most probably can replace – for better or for worse – many functions that now are run by the EU personnel. On the other hand, given the undeniable competitive advantage of foreign AI, the AI Act is a typical Technical Barrier to Trade (a specialty in which Europeans excel) to protect its non-competitive digital industry, and to negotiate better bilateral agreements with the US and China.
- AI as an alien species
Beyond the regulatory frenzy which has taken the EU, the AI Act is an example of a science-fiction idea of AI, as if the latter were some sort of alien species menacing Europe, its citizens, its values[1]. There are several misconceptions behind this approach.
The first is the ambiguous use of the term “intelligence” which appears to mimic human abilities, while it is consolidated science that human – and even more non-human animal – “intelligence” resides not only in mental faculties, but in the whole body, in its four senses and internal organs. This corporeal intelligence is something that a non-material entity such as AI does not possess. AI, instead, does something that humans are not able to do, i.e. process thousands, millions, billions of data drawing inferential conclusions upon which decisions are taken. Computational ability, is therefore, an extremely limited part of what we mean for “intelligence”.
Apart from the misnomer, the second misconception is that what we know of and on AI is not what AI will be in the very near future. The Act is an attempt to shape the development of a technology which is completely out of the EU’s reach. This denotes a fear of the unknown and the conceited idea to master it.
However, it is doubtful that the EU institutions have the capabilities to forecast future developments and if one sticks only to the digital environment, one should question the sources and forecasts it has relied upon in the last 40 years.
Decades ago, the formula “Voodoo economics” was invented for haphazard forecasts and measures similar to pins sticked in a doll. One could easily adapt the formula into “Voodoo regulation”. This is not because of some innate irrationality of the EU legislature, but simply as an application of the “precautionary” dogma, whose first tenet is that a political institution, if it does not want to be accused of overlooking risks, must adopt – for its own survival – a “worst scenario” approach. Clearly this reduces in the first place the political risks, because it will be very difficult to prove counter-factually that a more flexible approach would have been better. But the misconception is broader. The paradox is that by presenting AI as a potentially dis-humanizing technology the EU institutions have put it into place what Karl Mannheim called a self-fulfilled prophecy.
What seems to be ignored in the Act is that AI is a product of humans. And as with any human product its quality depends not on some casual and unpredictable event (the by now mythological “black-box”) but on the quality of the data analytics that move it and the quality of the data that is fed into it at the beginning and is continuously collected.
And behind each and every phase of an AI tool are human beings with their abilities and their limits. To put things bluntly, stupid people create artificial stupidity. Stupidity is not a feature that can – stupidly – be measured and curbed though some IQ test, but a much more nuanced notion that must take into account the context, the circumstances, the actors, the purpose. A complexity which surely AI is capable of tackling but only if fine-tuned. And this is a costly, time-consuming, skilled operation that is the essence of quality.
And stupidity (which, to paraphrase René Descartes, «est la chose du monde la mieux partagée») can be found not only in the creation of AI tools, but even more in their implementation, especially when humans – and not a machine – tend to have a fideistic trust in their results. Natural stupidity – which is devastating when coupled with assumed “artificial intelligence” – is the lack of doubts and especially the loss of the basic notion that just as human make mistakes, also human produced tools can make mistakes. And that the latter can be more easily detected through appropriate digital technologies which must be nor glorified nor damned, but simply properly used.
- Other legal idiocies
When facing a 270-page text made up by more than ninety thousand words one should legitimately ask oneself if the drafters were in their right senses[2]. And further wonder if it has ever dawned upon them the doubt that such a text is completely unworkable, unless some high placed official in Brussels was winking his or her eye in order that we understand that the AI Act is simply a declamatory exercise in regulation, a sort of mock-test which will have no consequences. But sadly, that is not the case, because all the declarations are in the sense that the AI Act is a dead-serious attempt to regulate the new technology and present Europe as the standard-setting world authority in this field.
The problem is that this text is at odds with the most abused mouth-washing principle in the EU, i.e. the Rule-of-law, of which the certainty of law is an essential component. When a legal text is ambiguous, open to multiple interpretations, contradictory, totally devoid of the sense of proportionality, self-multiplying in unknown directions like the generative AI it purports to regulate (see article 7 on the delegated powers) it is quite impossible for rational operators – whether public or private – to plan their activity over a reasonable time-span, deploying human and financial resources, verifying the effects and the performance of the measures adopted. The AI Act is a typical measure written by someone who will never be asked to abide by the rules it has set – and at any rate it will apply to itself the Humpty Dumpty principle («When I use a word, it means exactly what I choose it to mean – neither more nor less»).
Its only task will be to sit back and sanction – on a very casual and casuistic basis – what it does not feel is in conformity with its will and whims.
One could – ironically – suggest that the only way to make the AI Act work is through artificial intelligence. But Europe does not have such sophisticated tools, and therefore will have to resort to foreign enterprises to enforce the straitjacket it has sewn for itself.
Congratulations!
[1] Not surprisingly, one of the most quoted images used to forebode the dangers of AI is Hal the computer that wants to take command of the spaceship in Kubrick’s “2001 Odyssey in the Space”.
[2] In my pocket Bible, which I always have with me, the four Gospels cover 146 pages and the first five books of the Old Testament (Genesis, Exodus, Leviticus, Numbers and Deuteronomy) require only 248 pages. And even suggesting that their authors may have received a little help in conciseness from above, without such supernatural aid Messrs. Portalis, Tronchet, Maleville and Bigot de Préameneu managed to contain the 2281 (twothousandtwohundredandeightyone) articles (table of contents included) of the Code Napoléon in 266 pages (Les Cinq Codes, Richomme, Paris 1811).