This year’s Nobel Prize for Physics went to John J. Hopfield and Geoffrey E. Hinton who researched fundamental methods of computer science that made AI and machine learning possible in the first place. The Nobel Prize for Chemistry went to David Baker, Demis Hassabis and John M. Jumper for their AI models that can predict protein structures and thus open the door to huge advances in biology and medicine, among other things.
Why should AI be regulated anyway?
With the help of AI, great things can be achieved for the benefit of all. We need more AI, not less.
The EU AI Act is a done deal, the Council of Europe (COE) has published a document on AI safety, also signed by the US and UK.
At the same time, advocates of the free market are fighting fiercely against any form of AI regulation, with a large VC company (despite a translation library as the name sponsor, not very pre-Babylonian) leading the way.
They are afraid of communism and fear that innovation will be hampered by (over)regulation, and that (evil?) players like China will race past us (who exactly?) unchecked (on the way to world domination via AI?).
Large AI companies such as Meta are writing high-profile letters to the EU and even larger would-be AI companies such as Apple are fighting with lobbyists and, more recently, on the backs of users to persuade the EU’s regulatory folks to give in.
The main argument is that the EU directives, especially on data protection, are too complicated and – chaotic as they appear – untrustworthy. Therefore, according to Meta and Apple, their AI models cannot be published in the EU.
The desired effect is there, EU doomsayers are crying out, and even the AI ethics community is reacting with desperate appeals to the freedom of research.
But of course, this is above all a political power play. These players will not miss out on the market. Understanding the EU rules (and complying with them) is no fun – from my own painful experience, I will sign letters on this at any time (as long as the signature is not digital, which apparently is still not acceptable in many areas in the EU. But no, we don’t have a fax machine in our office).
However, compliance is not impossible, and the main requirement is, put very simply, transparency. The fact that this is not desired is as clear (transparent?) as it is understandable. Perhaps, Meta, the term open source should be discussed in more detail for your llamas.
Personally, I’m a big fan of AI (and of free market economics, but that’s not very relevant here). We at Daiki are also big fans of AI. We want to help teams use AI as successfully as possible everywhere. In a sensible, responsible, and (above all in an economic sense) sustainable way. That’s why we developed our AI governance software. I am therefore against regulation. In the case of AI, it is nevertheless necessary.
In this text, I don’t want to go too much into the big question of the medium to long-term social and possibly existential consequences of Nick Bostrom-like super AI. These are exciting and important questions.
The ultra-short version of my answer: Firstly, yes, AI solutions will bring about far more serious changes in most areas of society than those currently making the rounds in public discourse, but overall I am quite confident that this can happen positively, for the benefit of people and humanity.
Secondly: yes, AI can cause existential crises. At some point, there will be software that masters all cognitive functions better and faster than a human – than all humans.
These will not be large language models, but complex machine learning models. Whether these models are intelligent is another exciting philosophical question, whose winding paths through Chinese rooms I will not follow here.
The more important question for now is whether these models can be controlled. I think that is possible. The opposite is also a possibility.
The effects of using AI solutions are enormous and becoming even more enormous.
There are many significant challenges that AI presents us with. As the great Stephen Fry describes it:
“Machines are capable of bias, hallucination, drift and overfitting on their own, but a greater and more urgent problem in my view is their use, abuse and misuse by the three Cs . They are Countries with their specific ambitions, paranoias, enmities and pride; Corporations with their unaccountable rapacity and of course Criminals. All of them united by one deadly sin: greed.“
So let’s turn to this misuse as one of several challenges of AI that call for regulation.
AI and trust
In fact, probably the biggest problem with (generative) AI is that it can destroy trust in information. How are we supposed to trust anything and anyone (in the digital world) if we can no longer be sure whether a piece of information – a text, an image, an audio message, a video – is trustworthy, i.e. has a verifiable source which can provide further information about it?
My kids are – unfortunately / of course / ironically? – also TikTok users. This attention machine is constantly producing content that is faked to show cool tricks. Faked by humans, using mostly simple means of video editing, cuts, and repeats for double flips, reverse playback for a perfectly bouncing back bouncy ball, and so on.
As long as it is clear that tricks are being used, this is not a problem. Illusion and magic shows can be exciting, with deception and distraction.
But if it’s not clear whether what you’re seeing is real or not, it gets tricky. And in the case of TikTok, even dangerous.
Many people use this platform as a source of news and information. Also for political (mis)education. The content there cannot be trusted. And this crisis of trust will spread drastically and affect all digital media as soon as (generative) AI enters the playing field – and it has long since been substituted.
The simple video tricks mentioned above are nothing compared to the current and (in the very near future) upcoming text-to-video, image-to-video and video-to-video AI tools. It will very soon be much easier to create cool TikTok videos of yourself and others and other others using AI. The digital world will be flooded with generated fakes that look completely real.
Even the US Department of Defense wants to create fake people to deceive real people.
In one of his last texts, “The Problem With Conterfeit People”, Daniel C. Dennett talks about AI-generated people that make us believe the truth. According to Dennett, anyone who creates and disseminates such “fakes” should be prosecuted.
Dennett compares AI to money in terms of its social significance. He writes: “Money has existed for several thousand years, and from the outset counterfeiting was recognized to be a very serious crime, one that in many cases calls for capital punishment because it undermines the trust on which society depends. Today, for the first time in history, thanks to artificial intelligence, it is possible for anybody to make counterfeit people who can pass for real in many of the new digital environments we have created. These counterfeit people are the most dangerous artifacts in human history, capable of destroying not just economies but human freedom itself.”
The AI / money analogy is shaky (like any comparison), but it is actually very convincing for our trust argument.
Money works because everyone trusts the value of money and because states use laws and law enforcement to ensure that this trust endures.
Dennett continues: “Democracy depends on the informed (not misinformed) consent of the governed. By allowing the most economically and politically powerful people, corporations, and governments to control our attention, these systems will control us.”
No one in their right mind would think of putting the value proposition of money in the hands of corporations. (As a side note, I believe that most crypto enthusiasts are of sound mind. Even if many in this group go to great lengths to counteract this impression. Bitcoin is an exciting experiment, and based on an even more exciting technology, but it is worthless for our argument because it does not – yet – carry its own independent value foundation).
Money is controlled by the state everywhere in the world. Unfortunately, not always “by the people” (which is definitely a problem), but always by state institutions.
Counterfeiting is strictly prosecuted. And great efforts are being made to make counterfeiting more difficult. The EURion constellation, for example, is a pattern based on (semi-)secret algorithms that can be found on many banknotes worldwide.
If it were easy to counterfeit money, trust in money would disappear very quickly. The consequences would be catastrophic. If it were easy to counterfeit people, trust in people would disappear very quickly. The consequences would be catastrophic. The former is not easy, the latter is just getting way easier.
We should not relinquish the basis of trust in money. Unfortunately, this is exactly what we are doing with AI.
If fake content that hurts, defames, humiliates, deceives, etc. can no longer be distinguished from real content, then the only thing left to do is complete digital detox – but without a how-to from TikTok.
Open source and its limits
It is a charming idea to simply make AI models available to everyone and thus distribute control.
Yann LeCun from Meta, for example, argues that open source is important for trustworthy and sovereign AI. Meta invests a lot of money and makes its large models available under (something like) an open-source license. This is above all a USP, but still very laudable.
According to LeCun, the same applies to AI as to freedom of the press: many sources of information are only available with many players and thus open-source models that can be adapted by many other AI researchers. That is true.
And because this is so important, there should be binding rules. Meta is now riding the open-source bandwagon. For how much longer? Will EU citizens stay on the sidelines (because the EU wants to protect its citizens’ data – in an admittedly overly bureaucratic way)?
Open source is good, but it only solves part of the problem, namely that of the AI manufacturers; the problem of fake and dangerous content remains (and could become even bigger with free models).
Fair play can only take place where clear rules apply and are enforced. Voluntary commitments have not worked for banks (that could have been clear, but humans, like AI, like to learn from examples). So why should they work for AI?
The only effective measure against the loss of trust is to regulate AI – or more precisely – certain AI systems. And the consistent prosecution of the misuse of this technology.
When AI is regulated as it is in the EU, this can be very annoying – for small and large companies that want to use AI, for researchers, and, above all, for AI developers.
Fortunately, there are companies and solutions that enable AI and ensure compliance with these regulations almost incidentally.
There are also technical options for marking AI-generated content as such (the C2PA standard initiated by Adobe or Google’s SynthID are examples of this) and for registering AI models in a tamper-proof manner. They just have to be used.
Preferably voluntarily, more preferably with a little regulatory pressure.
The EU AI Act, for example, promotes a “human-centric approach to AI [… ] in the development of secure, trustworthy and ethical AI” based on fundamental ethical principles like human oversight, transparency, and fairness.
AI is great and great things can be achieved with it, as long as we can trust it and its developers.
