Article Published March 16th, 2021
by Dr Hugh Lawson-Tancred, Associate Research Fellow at Birkbeck, University of London
Henry C.W. Price, Doctoral Researcher, Imperial College London
Mankind’s last invention
At the present critical juncture of world history, it is of vital strategic interest for the European Union to develop the capacity to match and perhaps outstrip the progress made in the USA and the People’s Republic. To achieve this, a crucial role should naturally fall to London.
Information travels through silicon (let alone graphene or quantum carbon nanotubes) 10 million times faster than through organic tissue. The brain, on the other hand, consists of 100 billion neurons, with roughly as many as the stars in the Milky Way, with perhaps a quadrillion connections. Neuro architecture is fantastically complex but amazingly we are beginning to understand it. Once the engineering concept of artificial replication of dynamic network storage of information is established, as it has already been, the process whereby silicon brains overtake carbon ones is unstoppable. In many areas, such as formal games, natural language translation, the extraction of objects from the visual array, suprahuman levels of achievement have already been reached or are imminent. It is true that human-transcendence is, so far, confined to specific skills. It is far from obvious how, or whether, these advances can be extrapolated to the modelling of general artificial intelligence (AGI). However, there is clearly a spectrum of performance between an AI mastering the game of Go and full AGI, and each step along this spectrum will have profound implications.
Vladimir Putin is surely right to say that whoever controls A(G)I will control the world. If we can invent anything like a general AI, then it will indeed be our last invention. Everything else that we need, it will invent for us, at inconceivable speed. It will also, at the same rate, increase its own capacity, thereby raising the alarming prospect of an intelligence explosion, by which artificial intelligence vastly and uncontrollably exceeds human, as graphically forewarned by Nick Bostrom.
The road along the spectrum to AGI is fraught with peril. As Bostrom also stresses, the strategic case for a first strike against any power perceived to be on the threshold of AGI will be extraordinarily strong. In such circumstances, the development of AI would ideally be governed by international treaties comparable to those regulating nuclear energy, than which it is arguably still more dangerous. However, there is currently little realistic prospect of such an accord. Instead, as we enter the third decade of the century, research in AI is dominated by two major powers, America and China. All other players, including Russia, India and, especially, Europe (with by far the world’s largest concentration of basic science) currently lag far behind.
What it takes to “get AI done”
What, then, is required to become a major player in AI? There are four crucial requirements. The first is scientific talent. Current progress in AI is driven by an army of researchers with a number of conspicuous rock stars. They are, of course, concentrated in Silicon Valley but are also distributed around the world. From the European perspective, there certainly seems to be no reason to fear a fundamental shortage of the necessary technical abilities. The second requirement is hardware. The progress in the last two decades has largely been driven by the arrival of ever higher powered processors, made commercially viable by the appetite for gaming and sufficiently robust to cope with the network structures required for deep learning. These continue to be developed at warp speed and, crucially, they are about to be joined by the still greater potential of quantum computing. Here again, Europe is not at a disadvantage, as much important research on quantum computers is conducted in our continent. The third crucial ingredient is funding and political will, on which more below, while the final and in some ways most vital precondition is an abundance of suitable and available data. Here, Europe stands between China with its billion data subjects and the US, with access both to its own 300 million and to many other originators of data around the world. The European Union, with a first-world population of slightly under 500 million, is in a competitive position in this respect as well. Crucially, the stricter European data protection regime is not an impediment, indeed possibly a stimulus, to the effective development of AI.
The current leaders
In order to see how the EU could emerge as an AI giant, we need to look briefly at the background to the current dominance of the USA and China. The American case is clearest. At the dawn of the computer age, the two leading players were the USA and the UK. Both played an equal role in the deployment of computing in the course of the Second World War. What happened in the 15 years after that, however, could not have been more different. While the British were driving Alan Turing to suicide and cyber visionary Donald Michie to despair, the Americans, under President Eisenhower, were pouring vast funds into the newly created Defence Advanced Research Projects Agency (DARPA). Much of this money went to the West Coast and effectively seeded Silicon Valley. American readiness to write blank cheques for fundamental research paid off handsomely.
This is not a lesson which has been wasted on the Chinese. They too have developed an industrial strategy with a clear emphasis on advanced technology. The Chinese academic infrastructure both on the mainland and in Hong Kong (with an impressive five of the world’s top 100 universities) has risen at astonishing speed to rival the much older American centres. The Chinese authorities have encouraged the collection of the data generated by their citizens and imposed a liberal regulatory burden on the use of the data harvest. There is a similar willingness to devote effectively limitless resources to what is regarded, rightly, as a key strategic area. There are indeed also signs that the Chinese have ambitions to move into one area of particular interest to London, namely fintech.
The Cummings message
Dominic Cummings is a man now thought to need burial rather than praise. However, it is important to distinguish what was right in his vision from what was absurd. Cummings, albeit in a vainglorious and amateurish manner, understood the need for an industrial strategy focused on AI. The absurdity of his position was the fantasy that the UK would be able to achieve a global position by adopting such a strategy on its own. It is to be hoped that the departure of Cummings will put an end to that delusion. However, the need for a systematic strategy across multiple agencies to foster the development and deployment of AI remains critical and has still been insufficiently grasped in the corridors of Brussels power. Europe is behind in the game, but it is by no means impossible for it to catch up and indeed in the vital interests both of the continent and of the whole world that it does so.
Where then, we should ask, is European AI at the moment and what needs to be done? This question can be answered from both a bottom-up and a top-down perspective.
From the bottom-up perspective, the position is looking pretty promising. There are many advanced centres of AI research around the continent, and scholars from such institutions have now created the hugely promising initiative of the European Laboratory for Learning and Intelligent Systems (Ellis) to coordinate the activities needed to launch a truly Europe-wide platform. Ellis is still very young, but already it has shown a capacity to dynamise the huge latent talent available to create a European AI giant. Europe has the lion’s share of the scientific talent and funding in the world. Yet its performance in using it for Big Science has been mediocre. Whereas the Americans have repeatedly followed the logistical triumph of the Manhattan Project, nothing similar has occurred in large-scale scientific coordination in Europe, for all the eye-catching but sporadic successes of the European Space Agency. The shortcomings of European Big Science were only too painfully revealed, for example, by the Human Brain Project. It is vitally necessary that the bureaucracy listens to the scientists. If the funding agencies can respond appropriately to the burgeoning initiatives from within the European scientific community, there is absolutely no reason why European AI could not rapidly rise to match and indeed surpass what is happening in China and even America.
The top-down view is dominated by the Commission White Paper of January 2020, but there are also some practical initiatives that the Union is taking to promote AI directly. The White Paper was of course unluckily timed, discussion of it immediately drowned out by the pandemic. In some ways, however, this is not a bad thing. For the White Paper was a document conspicuous for its modesty. The Commission is open to the criticism of concentrating more on the moral hazard of AI than on its industrial potential. (Even so the loudest objections came mainly from those feeling that it took insufficient account of ethical considerations.) It is indeed a crucial advantage for the EU to be able to seize the moral high ground in AI as it has already done in data. This ethical premium is likely to be a vital competitive edge in the coming decades. However, an ethical premium on its own is not enough. It needs to be flanked by a systematic policy of creating structures and systems both to comply with and embody the ethical principles and to achieve state-of-the-art results. It is in this area that the Commission White Paper is particularly disappointing. There is certainly nothing in it to hint at a willingness to make the massive deployment of resources needed to create an environment for the evolution of AI in Europe comparable to what has been achieved in America for two generations and is rapidly emerging in China. This urgently needs to be addressed.
It is true that the Commission has taken certain practical initiatives. For example, the Elise programme is a systematic attempt to integrate research on AI across the continent. This is certainly a step in the right direction, but unfortunately, it also is a timid step. The total budget for Elise is €12 million, whereas the provision should really be measured in billions. Obviously, we currently live in an age of fiscal stress, but the long-term vision must be for a much greater orientation towards the infrastructure of AI. Crucially, this is not in any way at odds with the green revolution, indeed it could be argued that a meaningful green revolution will only achieve its goals if bolstered by AI. There is huge scope to use AI to identify the best ways in which the difficult balance can be struck between economic well-being and ecological sustainability. Expanding the currently timid AI initiatives and harnessing them to the ambitious green ones will be a key success factor for Europe in the next two decades.
The British wildcard
How, then, in the current dispensation do the British fit into this European initiative? It is certainly true that among the individual European nations the British are the leaders in AI. Even leaving aside the special position of London, the universities of Edinburgh, Sussex, Oxford, Sheffield, Birmingham, Cambridge and Warwick are all at the cutting edge, as is the Alan Turing Institute established and driven by those universities, the London colleges and many other UK institutions. Indeed, of the medium-sized European nations it is only the British who could even harbour the fantasy of becoming an AI superpower on their own. It is only one of the most damaging consequences of Brexit that this enormous UK pool of talent is at risk of being excluded from the European AI initiative. It should, of course, hardly need to be mentioned that, although these are British institutions, a huger proportion of the actual scholars come themselves from the European Union. Were research to focus on a separate British initiative, it is quite likely that these European scholars would revert to their home member states, still further undermining the British effort. The continuation of the extraordinarily fertile collaboration of European academics with British institutions which has been such a feature of the last 40 years should obviously be prioritised wherever possible.
This applies across the UK, but the political reality is that it will be easier to ensure the continued presence of London at the heart of European AI than that of Britain as a whole, at least for the moment.
What then of London? London has a unique combination of advantages, potentially putting it in pole position across the entire world. It has an unmanaged concentration of world-leading academic institutions, spearheaded by the four members of the Golden Triangle (ICL, UCL, KCL and the LSE). In addition, especially in the Knowledge Quarter, it has a unique ecosystem for academics and scientists to talk to entrepreneurs and venture capitalists. On top of this, London enjoys the benefit of gold-standard regulation in both finance and technology. Finally, London has a unique cultural environment making it extremely attractive to the inflow of creative talent. All these benefits have, of course, been hugely magnified by membership of the European Union and are now correspondingly put at risk.
It is vital that Brexit is not allowed to detach London from its European intellectual hinterland. London as the central hub of a Europe-wide AI initiative makes perfect sense. London, on the other hand, as the flagship of a separate British programme would create a deep structural imbalance and, if it achieved anything, merely render the British playing field still less level.
Indeed, London could prove to be the crucial catalyst in forging the link between the relatively unresponsive bureaucracy of the European Union and the energetic AI talent that it harbours. London can act as a kind of crucible. Nowhere else in Europe is it so easy for scientists from different member states to meet, discuss and create. The continued performance of this function by London is therefore crucial for the transition to the European age of AI. It is also, obviously, a vital factor for the continuation of the prosperity and well-being of the inhabitants of London and indeed for their ability to continue to support and foster the economy of the wider UK as a whole.
It is therefore an urgent political project to ensure that, in some form, the detachment of London from the continent in the scientific, technical and in particular AI sectors is minimised and contained as much as possible during the Brexit phase. At the very least, this will require an absence of meddling by the Westminster government in the development of AI policy in London. All attempts by Westminster politicians to lock the AI talent of London into purely British strategies should be resisted. (It is unclear how much this remains a threat after the departure of Cummings and the sheer chaos of the current administration may in itself diminish the risk, but constant vigilance is nevertheless still required.)
Going beyond this negative provision, it surely makes sense for London to institutionalise an ongoing connection with European AI. One obvious way for this to happen would be for London to have a direct and separate integration into the Horizon Europe programme. Would it not be possible for London to make a separate contribution to the central funding in order to be able to ensure that it is included in the programme whatever stance is adopted by the UK government? The overall coordination of this relationship would be handled by theoffice of the Mayor. An encouraging step in this direction has already been taken with the commissioning of a major survey of AI in London from the consultancy CognitionX.
A still further and more effective guarantee of continued London involvement in Europe would be for the declaration of a kind of AI Freeport, with a specific regime of both data compliance and strategic integration between a defined area of the UK and one aspect of the larger European economy. If such a Freeport London continues to contribute taxes to the UK economy, then all three parties seem to win. London taxes will be able to support at least some measure of “levelling up” in the North (and possibly even retain Scottish membership of the UK). London itself will be able to prevent or mitigate the impairment of its capacity to “prosper mightily”. And the European Union will have a crucial instrument for the project of achieving global stature for Europe in AI which, unlike the Cummings vision for Britain, is a real possibility.
What we need, then, is a European Cummings. We need a gadfly to sting the slumbering European beast into activity, and, we would suggest, the most natural form that this gadfly would take would be that of a dynamic London AI community closely integrated with the other great centres across the continent in terms of industrial strategy as it already is in terms of fundamental research. This does not require the creation of some ambitious new infrastructure, merely the funding and leadership to allow what already exists to continue on the natural trajectory which has been so recklessly put at risk by the destructive folly of Brexit.
Linking London into the EU in this way would form a crucial bridge to integrate and engage the other major AI centres around the British Isles as much as possible into the dynamic European project. Other major institutions could point to the success of London and organise themselves at local level to join it. Crucially, London would at all stages welcome and encourage such initiatives. London could form the spearhead for a more general re-engagement of the UK AI community with the European heartland, as a kind of first among equals. London is the only centre currently in a strong enough position to be able to lead this intellectual return and therefore play a crucial role in one important element of the British return to the European home. It must all start in London.
 https://www.mckinsey.com/~/media/McKinsey/Featured%20Insights/Artificial%20Intelligence/Artificial%20intelligence%20in%20the%20United%20Kingdom%20Prospects%20and%20challenges/Artificial-intelligence-in- the-United-Kingdom-VF2.ashx