This year a Nobel Prize was given to artificial intelligence pioneer Demis Hassabis, whose AI software cracked an impossible chemistry problem—predicting the structure of every protein known to science. When Hassabis was finishing his PhD and I was starting mine, we shared an office in London at the neuroscience research institutes known as Queen Square. And Queen Square is also where AI pioneer Geoffrey Hinton, who won a Nobel Prize the day before in a different field, founded the computational neuroscience unit where Hassabis and I spent time. Queen Square has been little known outside neuroscience, but it’s impossible to understand American AI capabilities without the international networks running through it. In fact, such hidden networks illustrate why and how the United States can get an innovation edge, in AI and a lot else besides.
As Robert Louis Stevenson wrote in the 19th century, Queen Square “is a little enclosure of tall trees and comely old brick houses.” It remains an elegant London square with a pub on the corner, tucked away in the district of Bloomsbury. Associated for over two centuries with brain doctors, today it is home to a neurology hospital, and neuroscience labs have taken over many of its buildings. When Hassabis and I were there in the mid-to-late 2000s, it was an extraordinary environment for basic research on the brain: Our functional brain imaging lab alone had three of the world’s top 10 most influentially cited neuroscientists. We both benefitted from the unit Hinton had founded there a few years before. Queen Square is also part of University College London, which, before this year, already had three neuroscience Nobel Prizes, the latest in 2014 for work on the hippocampus: the same brain region Hassabis researched for his PhD. Brilliant ideas about how the brain works flow from and around Queen Square—ideas that matter far beyond neuroscience.
This year a Nobel Prize was given to artificial intelligence pioneer Demis Hassabis, whose AI software cracked an impossible chemistry problem—predicting the structure of every protein known to science. When Hassabis was finishing his PhD and I was starting mine, we shared an office in London at the neuroscience research institutes known as Queen Square. And Queen Square is also where AI pioneer Geoffrey Hinton, who won a Nobel Prize the day before in a different field, founded the computational neuroscience unit where Hassabis and I spent time. Queen Square has been little known outside neuroscience, but it’s impossible to understand American AI capabilities without the international networks running through it. In fact, such hidden networks illustrate why and how the United States can get an innovation edge, in AI and a lot else besides.
As Robert Louis Stevenson wrote in the 19th century, Queen Square “is a little enclosure of tall trees and comely old brick houses.” It remains an elegant London square with a pub on the corner, tucked away in the district of Bloomsbury. Associated for over two centuries with brain doctors, today it is home to a neurology hospital, and neuroscience labs have taken over many of its buildings. When Hassabis and I were there in the mid-to-late 2000s, it was an extraordinary environment for basic research on the brain: Our functional brain imaging lab alone had three of the world’s top 10 most influentially cited neuroscientists. We both benefitted from the unit Hinton had founded there a few years before. Queen Square is also part of University College London, which, before this year, already had three neuroscience Nobel Prizes, the latest in 2014 for work on the hippocampus: the same brain region Hassabis researched for his PhD. Brilliant ideas about how the brain works flow from and around Queen Square—ideas that matter far beyond neuroscience.
AI is all the rage now and moves trillions of dollars in stock prices, but for two decades after failures in the early 1990s, AI was a backwater. That continued until 2012 when Hinton, who had moved to Toronto, published revolutionary research on artificial brain networks for computer vision. Hassabis had recently founded the start-up DeepMind—now owned by Google but still a 25-minute walk from Queen Square—and Hassabis’ team built better artificial neural networks which in 2016 beat the human world champion at the game Go. That was a Sputnik moment for China’s leaders, a huge technological surprise like the one delivered by a pioneering Cold War Soviet satellite to shocked Americans. The rest is history.
A leafy London square’s hubbub of ideas was one wellspring in the networks that supplied the intellectual fuel for our current AI boom. The process of innovation goes from basic research (which seeks laws of nature, usually in universities like at Queen Square) to “deep tech” (like at the Defense Advanced Research Projects Agency or DeepMind, typically with long time horizons, frontier science, and big potential commercial or security value), to proprietary research (Apple’s designing the iPhone) and scaling (Apple building the iPhone at scale). Understanding where ideas arise, and how they evolve, matters for democracies today because of China’s rise as a peer innovator.
Chinese science has now almost caught up with the best U.S. and U.K. universities, and in 2022 it overtook the United States to publish more highly cited papers. China leads some fields. Its brilliant Peking University, with which I’ve collaborated, now ranks globally above any European Union university. Chinese companies lead innovation in areas like low-cost electric vehicle manufacturing. And the United States needs an innovative edge to offset China’s advantage as our era’s manufacturing superpower. China today manufactures more than the next nine largest countries put together, and is increasingly sophisticated: In 2022 China installed over half the world’s industrial robots.
China is strong all along the process of innovation for AI. It now publishes more AI research papers than the United States and is catching up fast in highly cited AI papers. In 2022, China produced 61% of the world’s granted AI patents. China has innovative AI start-ups, as well as tech titans like Tencent. And Chinese companies are innovating furiously in downstream applications like driverless vehicles. Even a highly innovative United States with 333 million citizens will struggle to out-innovate 1,412 million increasingly well-educated Chinese. Allies will be crucial, but how?
Collaboration is often easier earlier in the process because it helps avoid minefields that can bedevil later stages, such as intellectual property rules, patent arguments, or protecting national champions. That suggests a focus on basic science—like at Queen Square—and the deep tech that emerges from it—like at DeepMind.
In basic science, which mostly happens at places like universities, an obvious answer for the highest impact is to turn to the “five eyes” nations: the United States, the U.K., Canada, Australia, and New Zealand. The United States currently has only 16 of the global top 50 ranked universities—but adding the five nations’ universities together they dominate the global rankings, with 33 of the global top 50 ranked universities, including eight of the top 10 (four American and four British). Together that’s a dominant chunk of the world’s top basic research that fuels innovation. The rest of the G7 plus South Korea would add only six more. The U.K. alone produces more highly cited science than any country except the United States and China. Australia and Canada occupy fifth and sixth places globally for the most highly cited researchers. From 2017 to 2021, Britain ranked third and Australia fifth globally in top-cited AI research articles. Such allies provide the mass of excellence vital for today’s global competition.
In deep tech, the five eyes nations also bring a lot to the table. London is the world’s sixth most valuable tech city after three U.S. and two Chinese cities. U.K. and Canadian tech funds, including venture capital and so on, rank next after the United States. The U.K. produces twice as many tech unicorns (that is, companies worth over $1 billion) as France or Germany. This doesn’t exclude other allies and partners. But in the process of innovation, the United States should nurture the hidden networks that generate real power in cutting edge tech like AI.
The challenge is how to knit together innovation ecosystems so the five nations’ combined might adds up. For basic science, the five nations should create a collaborative, multilateral, truly five-nation funding scheme for university-led research. It should require researchers from at least two nations and focus on key security and dual-use areas like AI, space, quantum, and genetics. The five nations can also award fellowships that require spending time in at least two nations—following in Hinton’s footsteps doesn’t sound bad, for instance, and would entail working in Britain, the United States, and Canada. And money awarded to each nation’s researchers can match each country’s inputs, so there’s no overall funding of overseas jobs.
For deep tech that follows in the innovation process, like at DARPA or DeepMind, the five governments should jointly map capabilities and critical supply chains, providing targets around which to organize joint missions such as in rare earths or biological threats. They should collaborate on deep tech venture capital for security and dual-use areas, building on organizations like Australia’s Main Sequence Ventures, the U.K.’s National Security Strategic Investment Fund, or America’s In-Q-Tel. They can build on the research pillar of AUKUS, the security agreement between the United States, the U.K., and Australia. The five nations should build an allied fleet of interoperable DARPAs for which the five nations’ long-established, trusted relationships provide a secure scaffolding to grow bottom-up connections between ecosystems of researchers, investors, and entrepreneurs.
DARPA has invented revolutionary deep tech for decades, such as stealth technology and miniature GPS. Key ingredients for DARPA’s success include institutional autonomy—and, above all, the empowered program managers who run DARPA projects and leave when they end (typically after three to five years), which maintains a relentless pace. The U.K. has a new agency modelled after DARPA copying that key method. The “five eyes” nations share the most intimate collaboration to produce intelligence, which has survived 75 years because it operated in distinct areas and worked for all five nations. Given the future’s inevitable political vicissitudes, this could produce a durable innovative edge for the next 75.
Such advantage in allied innovation will likely last decades, even if China seeks its own bloc through BRICS or the global south. Building science and deep tech capacities take decades of sustained effort, as China’s history shows, and only two countries can move the dial sooner, but, except in a few niches, Russia languishes far behind Britain scientifically, and India currently views China warily.
Hassabis and Hinton won their Nobels for “deep learning,” and that rests on the “hidden layers” in artificial neural networks. Those hidden layers sit between the network’s “input layer” (the layer that gets inputs like visual data) and the “output layer” (the layer that informs you it’s a picture of a cat). It’s the hidden layers that learn, and that get the job done. For U.S. innovations in AI, it’s ideas from little known networks through places like London’s leafy Queen Square that help get the job done too. The United States and its allies should nurture the hidden networks that generate real power.
The post The United States Owes Its Edge in AI to a London Landmark appeared first on Foreign Policy.
