Europe's artificial intelligence (AI) industry is not in a good state and is becoming desperate. The latest indicator of this desperation of an industry that has at best been mediocre as measured against achievements in the USA and China, is the call for "a European large-scale initiative on Artificial Intelligence", which include the call for a €100 billion public subsidy lifeline from European taxpayers. This  latter call has been through CLAIRE and euRobotics, industry support-and-lobbying organizations, arguing that an AI Moonshot is needed in the EU to throw €100 billion between 2024 and 2029 at creating a "competitive AI industry" underpinned by a CERN for AI, and which would use this funding to deliver a Made in Europe Generative AI (GenAI) by 2029 that is "trustworthy." The AI Moonshot is justified, amongst others regarding the  EU's public support for CERN and Airbus in response to industrial developments in the USA. 

The European Association for AI (EurAI) and the Centre for European Policy Studies (CEPS),  recently  reiterated and supported the call for this Moonshot. According to CEPS  it is needed for "restoring competitiveness and achieving technological sovereignty" in Europe. 

With European and perhaps global AI in the doldrums, the European AI community would like nothing better than access to (even more) public money.  Organizations like CLAIRE, euRobotics, and EurAI , whose members would hugely benefit from their proposed AI moonshot, are, therefore, far from being dispassionate, scientifically independent voices. The proposal, therefore, needs hard, cold scrutiny. This is because the objective of getting state subsidies for one's industry is as old as business itself.

There are probably around 30,000 lobbyists in Brussels wanting to cash in on taxpayers' money and influence regulations. Rent-seeking and regulatory capture are ubiquitous in Europe. Hence, the proposal for an AI moonshot by CLAIRE and euRobotics is unsurprising. What is surprising is the audacity of the claim.

First, despite funding for AI in Europe surging, the authors of the call do not see this as sufficient and propose an ambitious "Moonshot." A popular definition of a Moonshot is a "groundbreaking project undertaken without the assurance of near-term profitability or benefit and, perhaps, without a full investigation of potential risks and benefits." The original moonshot was the Apollo program of the US government, launched by President John F. Kennedy in 1961. This was essentially a political vanity project for the USA to save face after the Soviet Sputnik launch in 1957 and to distract attention from its war in Vietnam.

Once the mission was accomplished in 1969, further moon missions were abandoned. Its aim was to show the competitiveness of the USA's technological prowess and nothing more - from a scientific point of view, the program "did little to directly advance scientific understanding." It cost around 4.4% of the US Federal Budget, and while it did return technological benefits, it has been argued that "the moonshot was not a cost-effective way to boost technology. Giving a quarter of the money on the National Science Foundation would surely have accomplished more, as would directing NASA to spend it on satellites and unmanned space probes."

By labelling its call for €100 billion public investment in EU AI a Moonshot, the authors of the call are audacious by playing on the political pride of Europe. To invoke Europe's competitiveness and technological sovereignty and label it a Moonshot, it is suggested that it is about nationalism and vanity - to show that Europe can play along with the USA and China - irrespective of whether the economics make sense. Indeed, the authors are likely attracted by the idea of a moonshot precisely because it does not need the assurance of economic payback nor a full cost-benefit analysis or investigation of the potential risks and benefits.

Such analyses are thus completely lacking from the call. And while the proposal wants to paint the case for AI in the same frame as the EU's public funding support for CERN and Airbus, these are inappropriate examples. CERN was not a moonshot and has a modest budget of €1 billion per year - its recent plans to invest €17 billion in a supercollider are running into opposition, including from Germany, which has called such a budget unaffordable. If CERN's projects of a mere €17 billion are unaffordable, what hope is there for a €100 billion call and an AI-CERN? The comparison with Airbus is even more problematic: if the AI moonshot is to be anything like the EU's support for Airbus, it would be illegal in terms of WTO rules. When Airbus recently mooted asking the EU for subsidies, it raised the spectre of a transatlantic trade war. Hence, in the case of any EU Moonshot subsidies for its AI industry, there are likely to be repercussions from the US, flaming an unproductive trade and technology war.

The second reason why the call is audacious is now clear in light of the above. The CLAIRE and euRobotics claim €100 billion, amounting to around 10% of the EU's budget. This is more than double the Apollo moonshot share in the US Federal Budget (4,4%). And what do they offer in exchange for this? EU-based GenAI that is trustworthy. Really? Are Europeans so desperate for a "Made in Europe" GenAI that is "trustworthy" that they will fork out 10% of the EU's budget for it? 

The authors of the AI moonshot call seem to think that GenAI is a very big deal. This is surprising. As Jim Covello of Goldman Sachs recently pointed out, "Eighteen months after the introduction of generative AI to the world, not one truly transformative - let alone cost-effective- application has been found." According to economist Daren Acemoglu, AI, including GenAI, will automate less than 5% of tasks in the next decade and will, at most, contribute a slight 0.9% to GDP growth in the USA. And as Benedict Evans asks, "Hundreds of millions of people have tried ChatGPT, but most of them haven't been back. Every big company has done a pilot, but far fewer are in deployment.[...]most people who tried it didn't see how it was useful [....] why do most people say, in effect, 'very clever, but not for me' and wander off, with a shrug?"

The third reason why the call for an EU AI Moonshot is audacious becomes clear from comparison with the original moonshot. The Apollo moonshot was successful in that it had a clear target. It was clear to everybody when the target was achieved. The moonshot could then end. The problem with the proposed AI Moonshot is that the eventual aim is vague. One problem is the definition of AI. As Max Tegmark has quipped, even intelligent intelligence researchers cannot agree on what intelligence is. And what is trustworthy? The difficulty is that "trustworthy AI is a nascent, multifaceted concept that is hard to bound."  The danger is that with no firm and unambiguous goal in sight against which progress can be measured, the AI moonshot and AI CERN would become a bottomless pit into which public monies will disappear. It could turn into a nightmarish version of the costly Cancer moonshot, launched in 1971 by President Nixon, and which fifty years later has achieved little success and remains mired in controversy.

Claiming nowadays that AI is the solution to climate change is about as credible as claiming that a wolf is a good shepherd

The fourth reason why the call for an EU AI moonshot is audacious is that it claims that GenAI is so remarkable (despite the poor prognoses coming from the leading GenAI country) that it will "make a major contribution to solve climate change." Yes, the authors play the climate change card. Unfortunately, claiming nowadays that AI is the solution to climate change is about as credible as claiming that a wolf is a good shepherd. A recent New York Times headline phrased it succinctly: "AI Could Soon Need as Much Electricity as an Entire Country." A single query on GenAI consumes 2,9 watt-hours, almost ten times as much energy as a traditional Google query. If the US, one of the world's largest energy producers - and self-sufficient in energy - considers the AI energy challenge a significant problem, what about energy-starved Europe, which is critically dependent on the US for its energy supplies? And it's not just energy - the water demands of AI data centres are of growing concern. It is estimated that by 2027, AI data centres will require almost 7 billion cubic meters of water, nearly two-thirds of what an advanced economy such as that of England uses in a year.

Unsurprisingly, the EU AI industry has joined the queue of rent-seekers and regulatory captors in Brussels. It is par for the course in the EU. What is surprising is the audacity of its claim. Another element of the call for an AI moonshot is not so much audacious as astonishing - because of its incoherence. This is the timing of the call, coming as it does at the end of 2023.

For more than a decade, the EU's broad AI community has been emphasizing the enormous negative externalities (harms) that AI poses, supporting arguments that the EU should comprehensively regulate AI for the safety of EU consumers. This campaign has resulted in the AI Act and related legislation. If one accepts this course of action as having been correct, then it is inconsistent to now ask for billions of subsidies for the very products that one has argued poses so much harm. 

The EU does not need "AI Made in Europe", just as it does not need coffee trees to grow in Europe for its consumers to enjoy the best brews daily

There is a simple alternative to the AI moonshot. It is to be found in the simple economics of comparative advantage and cooperation, not the mercantilist, conflict-infused view of the world in which "competitiveness" and other zero-sum notions dominate. The EU has no comparative advantage in making AI products and services. It does not need "AI Made in Europe", just as it does not need coffee trees to grow in Europe for its consumers to enjoy the best brews daily. And this should not matter. It is a fantasy to imagine that the EU can be better in everything than everyone else. By specializing in its actual comparative advantage and engaging in international trade, EU consumers could be better off. If the EU is concerned about the "trustworthiness" of AI, it can, by all means, impose strict requirements on imports of AI - just as it does presently in the case of food, medicines and many other products that it does not make itself.  The EU should certainly come down harder on Silicon Valley Tech giants - Technofeudalists - if they misuse and abuse their market power, and address the loopholes in the AI Act. And, for the scientific and educational community in the EU, there is plenty of R&D and other subsidies pouring into AI research.

An AI moonshot would be a case of mission-oriented industrial policies (MOIP), which would only lead to a race to the bottom as this would invite other countries to adopt similar programs. The whole fallacy of MOIP is exposed in the recent scholarly book by Magnus Henrekson and colleagues, "Moonshots and the New Industrial Policy: Questioning the Mission Economy." It is a recommended read for all those who may still be convinced at this point that there is merit in the AI moonshot call. 

More than ever, problems such as climate change and the impacts of technological disruption require international cooperation rather than the Cold War logic of Moonshots

Finally, more than ever, problems such as climate change, conflict, migration, and the impacts of technological disruption require international cooperation. These are tough collective action problems. The Moonshot approach worked when it had a single mission to land a person on the moon. This is not the type of challenge which AI poses. An editorial in Nature Magazine in 2019 emphasised that the Moonshot approach to the challenges faced by the world today is less than ideal for addressing complex problems. By resurrecting the Cold War logic of a moonshot, the EU AI industry is not only calling for an audacious misallocation of public resources but also calling for Europe to shoot itself in the foot.