Published on May 8, 2023

The emergence of ChatGPT has led to a growing need to regulate the development of these systems as well as of the underlying data these systems rely on – which of course implies access to it. Regulation is now high on the policy and public agenda. This has inspired a significant group of scientists and businesspeople requesting a truce on the research around these algorithmic technologies.

This letter is quite remarkable and, apart from opposition to nuclear energy and its military capabilities, we have perhaps never seen such a high-level group calling a halt to research. There may of course be personal and commercial agendas driving it; most early signatories are based in the West; and the ask is only for a six month pause. But it is a remarkable initiative, nonetheless. The issues it raises should be taken very seriously and, in several ways, require deep reflection.

First, such an outcry to stop research has rarely, if ever, succeeded. When the first convergence of ICT and biotech became real, around the turn of this century, similar concerns were raised (as Yuval Harari has eloquently explained in his book ‘Homo Deus’). But they did not result in a stop on research. Why will this work now?

Second, this letter joins a succession, as many as five, concerted calls to stop proposed information and communication technologies becoming a widespread reality. With the emergence of cable and satellite TV, then of computers and the Internet, the world wide web and then web 2, we witnessed concerns that each would end the world as we know it – the end of jobs, of creativity, indeed of humanity. The end of analogue, of the bricks-and-mortar world, was predicted too. Nothing of that happened. But a lot has changed and mostly for the better.

The question arises, why will this call create a different outcome for data, AI, or ChatGPT? After all, the latter are ‘just’ technologies that automate what we as humans tell them to automate. As we trust a search engine to come up with non-malicious answers, we trust AI to do the same. So why worry now?

In my view, the answer is that data and the AI tools to interpret it, like ChatGPT, have become the fundamental building blocks of modern-day societies, much like energy. Without either, our societies come to a standstill and simply will not function as they do now. Neither data access nor AI are on Maslow’s hierarchy of needs, but they do enable those needs to be met in the 21st century. And of course, we could survive without data access and algorithms, if we are willing to go back to the ways in which we organized life and society in the 1950s. The same is true of modern energy production and use, in that we could survive if we went back to the Middle Ages.

Defining strategic assets

Data, and the AI tools to make sense of it, are now to be seen as strategic assets for any country or human activity, as much as energy production and use. And that may explain the present-day nervousness surrounding the debate on AI, not least because this discussion takes place in a changing geopolitical and geo-economic context. When the internet boomed, we lived in a time of ‘the end of ideologies,’ with a consensus that we all wanted the same kind of global order and similar ways of organizing societies. As the age of the ‘end of ideologies’ itself comes to an end, we need to look at global policymaking very differently today.

How? As I argue in my presentation to the Board on Research Data and Information at the National Academy of Science (linked below), the data policies of Europe can only be understood by considering the new European policy goal of “technological sovereignty.” This is seen as a drive to cut down dependency on the imports of those strategic assets Europe needs to keep its society going on. It implies being able to produce and control vaccines, for example. But primarily it implies being able to produce and control the strategic assets of modern-day society. Hence the substantial European investment boost in renewable energy, in data and in AI tech.

In record time, as I explain in my presentation, Europe has produced a significant body of regulation with the ambition of getting a grip on the use of data and ensuring the tech industry behind it pivots towards Europe. The US and China have similar policies in place, making the race for technological supremacy increasingly important in a changing geopolitical context.

Defining a global response

The same observation that data is a key strategic asset for societies can be applied to science. Science in the 21st century is data driven and the most relevant scientific tools available to make sense of that data are increasingly algorithmic. The possible benefits that AI offers science are immense. Chat GPT, for instance, offers the potential to eliminate the need to write the basics of a text, to find correlations that may be difficult to detect otherwise, to bring in new data sets, etc. The scientific community is therefore as concerned about the regulation of AI as society at large. And it should be.

Mutatis mutandis, the potential negative impact of any new geopolitical order on data access and use for science is just as significant for society as a whole. In a world where “technological sovereignty” is paramount, it is challenging to imagine that China, the US, or Europe will allow another power to control leading sense-making tools, such as ChatGPT. Nor is it imaginable that the three leading players in science will agree to mutualize access to their data and science without reciprocity, as that would imply less sovereignty. This creates a dilemma, as the internal logic of 21st-century science – which values openness, global collaboration, and fairness (FAIR) – is not compatible with the tech and economic ambitions of countries or continents striving to lead in data and AI.

If one accepts that the problem for open science, data access and the “correct” use of AI at large are entangled in a new geopolitical dynamic, then a solution for the “correct” use of data and AI must take that into account.

This is a complex task requiring a global approach. Attempting to impose an approach that lacks global support is likely to fail since the world is now multipolar. For instance, Europe had to accept being entirely pushed out of the internet and software race (there are no ‘European FAANGS’), and its present policy frenzy indicates it won’t tolerate that again. The US and China share the same concern about their position.

We need global regulation and a global agency

Bearing all that in mind, we need to consider at least four building blocks of a global regulatory approach, as follows:

All data and the algorithms underpinning them, particular in science, should be FAIR (findable, accessible, interoperable, and reusable), to make data a strategic asset and a common utility, like airspace for air traffic.
All AI systems should be obliged to make themselves known as AI whenever consulted by a human. “This text has been generated by ChatGPT” needs to become as mandatory as “any conflict of interest the author has been cleared.” Why not consider watermarking AI systems, just like food packaging mentions country of origin?
Regulation should be use-case-based and not intention-based. What is unacceptable in one country may be acceptable in another, making any ex-ante ‘one size fits all’ approach ineffective. (I analyze in my presentation why such an approach will not work in Europe).
Self-regulation is insufficient in the world of data and AI due to conflicting interests and high global ambitions. A global level playing field for AI and data use policy, along with an early warning and control mechanism, is needed. The upcoming G20 in India could address this, but the United Nations is likely the most suitable institution, as it includes all countries, and excludes no one.

This could lead to the creation of an “International Data and AI Agency” or IDAIA. IDAIA may not be so utopian as it seems (1). After World War II, as it was digesting the devasting effect of the first atomic bombs, and in the run up to the Cold War, the world agreed to set up an agency for what was then considered the riskiest thing on the planet: atomic energy and its potential uses and abuses. Just as the world fears now that AI might get out of hand, atomic energy was feared then. The International Atomic Energy Agency (IAEA) was created.

The reasons behind its creation are remarkably similar, as the following extracts from the IAEA account illustrate. I have added in brackets the words ‘data’ and ‘AI’:

“The IAEA was created in 1957 in response to the deep fears and expectations generated by the discoveries and diverse uses of nuclear technology (data).”

“The IAEA is strongly linked to nuclear technology (AI) and its controversial applications, either as a weapon or as a practical and useful tool.”

“The Agency was set up as the world’s ‘Atoms (data / AI) for Peace’ organization within the United Nations family. From the beginning, it was given the mandate to work with its Member States and multiple partners worldwide to promote safe, secure, and peaceful nuclear technologies (AI).”

“The Agency shall seek to accelerate and enlarge the contribution of atomic energy (data / AI) to peace, health, and prosperity throughout the world. It shall ensure, so far as it is able, that assistance provided by it or at its request or under its supervision or control is not used in such a way as to further any military purpose.”

Just as the IAEA played a crucial role in containing the use of a potentially self-destructive technology, IDAIA may play the same role for what is, probably rightfully so, seen as a new, potentially self-destructive tech.

This was presented at the National Academy of Sciences. See the presentation attached

(1) A similar proposal was recently made by two leading AI scientists.