The AI Power Paradox

Can States Learn to Govern Artificial Intelligence—Before It’s Too Late?

It’s 2035, and artificial intelligence is everywhere. AI systems run hospitals, operate airlines, and battle each other in the courtroom. Productivity has spiked to unprecedented levels, and countless previously unimaginable businesses have scaled at blistering speed, generating immense advances in well-being. New products, cures, and innovations hit the market daily, as science and technology kick into overdrive. And yet the world is growing both more unpredictable and more fragile, as terrorists find new ways to menace societies with intelligent, evolving cyberweapons and white-collar workers lose their jobs en masse.

Just a year ago, that scenario would have seemed purely fictional; today, it seems nearly inevitable. Generative AI systems can already write more clearly and persuasively than most humans and can produce original images, art, and even computer code based on simple language prompts. And generative AI is only the tip of the iceberg. Its arrival marks a Big Bang moment, the beginning of a world-changing technological revolution that will remake politics, economies, and societies.

Like past technological waves, AI will pair extraordinary growth and opportunity with immense disruption and risk. But unlike previous waves, it will also initiate a seismic shift in the structure and balance of global power as it threatens the status of nation-states as the world’s primary geopolitical actors. Whether they admit it or not, AI’s creators are themselves geopolitical actors, and their sovereignty over AI further entrenches the emerging “technopolar” order—one in which technology companies wield the kind of power in their domains once reserved for nation-states. For the past decade, big technology firms have effectively become independent, sovereign actors in the digital realms they have created. AI accelerates this trend and extends it far beyond the digital world. The technology’s complexity and the speed of its advancement will make it almost impossible for governments to make relevant rules at a reasonable pace. If governments do not catch up soon, it is possible they never will.

Thankfully, policymakers around the world have begun to wake up to the challenges posed by AI and wrestle with how to govern it. In May 2023, the G-7 launched the “Hiroshima AI process,” a forum devoted to harmonizing AI governance. In June, the European Parliament passed a draft of the EU’s AI Act, the first comprehensive attempt by the European Union to erect safeguards around the AI industry. And in July, UN Secretary-General Antonio Guterres called for the establishment of a global AI regulatory watchdog. Meanwhile, in the United States, politicians on both sides of the aisle are calling for regulatory action. But many agree with Ted Cruz, the Republican senator from Texas, who concluded in June that Congress “doesn’t know what the hell it’s doing.”

Unfortunately, too much of the debate about AI governance remains trapped in a dangerous false dilemma: leverage artificial intelligence to expand national power or stifle it to avoid its risks. Even those who accurately diagnose the problem are trying to solve it by shoehorning AI into existing or historical governance frameworks. Yet AI cannot be governed like any previous technology, and it is already shifting traditional notions of geopolitical power.

The challenge is clear: to design a new governance framework fit for this unique technology. If global governance of AI is to become possible, the international system must move past traditional conceptions of sovereignty and welcome technology companies to the table. These actors may not derive legitimacy from a social contract, democracy, or the provision of public goods, but without them, effective AI governance will not stand a chance. This is one example of how the international community will need to rethink basic assumptions about the geopolitical order. But it is not the only one.

A challenge as unusual and pressing as AI demands an original solution. Before policymakers can begin to hash out an appropriate regulatory structure, they will need to agree on basic principles for how to govern AI. For starters, any governance framework will need to be precautionary, agile, inclusive, impermeable, and targeted. Building on these principles, policymakers should create at least three overlapping governance regimes: one for establishing facts and advising governments on the risks posed by AI, one for preventing an all-out arms race between them, and one for managing the disruptive forces of a technology unlike anything the world has seen.

Like it or not, 2035 is coming. Whether it is defined by the positive advances enabled by AI or the negative disruptions caused by it depends on what policymakers do now.

FASTER, HIGHER, STRONGER
AI is different—different from other technologies and different in its effect on power. It does not just pose policy challenges; its hyper-evolutionary nature also makes solving those challenges progressively harder. That is the AI power paradox.

The pace of progress is staggering. Take Moore’s Law, which has successfully predicted the doubling of computing power every two years. The new wave of AI makes that rate of progress seem quaint. When OpenAI launched its first large language model, known as GPT-1, in 2018, it had 117 million parameters—a measure of the system’s scale and complexity. Five years later, the company’s fourth-generation model, GPT-4, is thought to have over a trillion. The amount of computation used to train the most powerful AI models has increased by a factor of ten every year for the last ten years. Put another way, today’s most advanced AI models—also known as “frontier” models—use five billion times the computing power of cutting-edge models from a decade ago. Processing that once took weeks now happens in seconds. Models that can handle tens of trillions of parameters are coming in the next couple of years. “Brain scale” models with more than 100 trillion parameters—roughly the number of synapses in the human brain—will be viable within five years.

With each new order of magnitude, unexpected capabilities emerge. Few predicted that training on raw text would enable large language models to produce coherent, novel, and even creative sentences. Fewer still expected language models to be able to compose music or solve scientific problems, as some now can. Soon, AI developers will likely succeed in creating systems with self-improving capabilities—a critical juncture in the trajectory of this technology that should give everyone pause.

AI models are also doing more with less. Yesterday’s cutting-edge capabilities are running on smaller, cheaper, and more accessible systems today. Just three years after OpenAI released GPT-3, open-source teams have created models capable of the same level of performance that are less than one-sixtieth of its size—that is, 60 times cheaper to run in production, entirely free, and available to everyone on the Internet. Future large language models will probably follow this efficiency trajectory, becoming available in open-source form just two or three years after leading AI labs spend hundreds of millions of dollars developing them.

As with any software or code, AI algorithms are much easier and cheaper to copy and share (or steal) than physical assets. Proliferation risks are obvious. Meta’s powerful Llama-1 large language model, for instance, leaked to the Internet within days of debuting in March. Although the most powerful models still require sophisticated hardware to work, midrange versions can run on computers that can be rented for a few dollars an hour. Soon, such models will run on smartphones. No technology this powerful has become so accessible, so widely, so quickly.

AI also differs from older technologies in that almost all of it can be characterized as “dual use”—having both military and civilian applications. Many systems are inherently general, and indeed, generality is the primary goal of many AI companies. They want their applications to help as many people in as many ways as possible. But the same systems that drive cars can drive tanks. An AI application built to diagnose diseases might be able to create—and weaponize—a new one. The boundaries between the safely civilian and the militarily destructive are inherently blurred, which partly explains why the United States has restricted the export of the most advanced semiconductors to China.

All this plays out on a global field: once released, AI models can and will be everywhere. And it will take just one malign or “breakout” model to wreak havoc. For that reason, regulating AI cannot be done in a patchwork manner. There is little use in regulating AI in some countries if it remains unregulated in others. Because AI can proliferate so easily, its governance can have no gaps.

What is more, the damage AI might do has no obvious cap, even as the incentives to build it (and the benefits of doing so) continue to grow. AI could be used to generate and spread toxic misinformation, eroding social trust and democracy; to surveil, manipulate, and subdue citizens, undermining individual and collective freedom; or to create powerful digital or physical weapons that threaten human lives. AI could also destroy millions of jobs, worsening existing inequalities and creating new ones; entrench discriminatory patterns and distort decision-making by amplifying bad information feedback loops; or spark unintended and uncontrollable military escalations that lead to war.

Nor is the time frame clear for the biggest risks. Online misinformation is an obvious short-term threat, just as autonomous warfare seems plausible in the medium term. Farther out on the horizon lurks the promise of artificial general intelligence, the still uncertain point where AI exceeds human performance at any given task, and the (admittedly speculative) peril that AGI could become self-directed, self-replicating, and self-improving beyond human control. All these dangers need to be factored into governance architecture from the outset.

AI is not the first technology with some of these potent characteristics, but it is the first to combine them all. AI systems are not like cars or airplanes, which are built on hardware amenable to incremental improvements and whose most costly failures come in the form of individual accidents. They are not like chemical or nuclear weapons, which are difficult and expensive to develop and store, let alone secretly share or deploy. As their enormous benefits become self-evident, AI systems will only grow bigger, better, cheaper, and more ubiquitous. They will even become capable of quasi autonomy—able to achieve concrete goals with minimal human oversight—and, potentially, of self-improvement. Any one of these features would challenge traditional governance models; all of them together render these models hopelessly inadequate.

TOO POWERFUL TO PAUSE
As if that were not enough, by shifting the structure and balance of global power, AI complicates the very political context in which it is governed. AI is not just software development as usual; it is an entirely new means of projecting power. In some cases, it will upend existing authorities; in others, it will entrench them. Moreover, its advancement is being propelled by irresistible incentives: every nation, corporation, and individual will want some version of it.

Within countries, AI will empower those who wield it to surveil, deceive, and even control populations—supercharging the collection and commercial use of personal data in democracies and sharpening the tools of repression authoritarian governments use to subdue their societies. Across countries, AI will be the focus of intense geopolitical competition. Whether for its repressive capabilities, economic potential, or military advantage, AI supremacy will be a strategic objective of every government with the resources to compete. The least imaginative strategies will pump money into homegrown AI champions or attempt to build and control supercomputers and algorithms. More nuanced strategies will foster specific competitive advantages, as France seeks to do by directly supporting AI startups; the United Kingdom, by capitalizing on its world-class universities and venture capital ecosystem; and the EU, by shaping the global conversation on regulation and norms.

The vast majority of countries have neither the money nor the technological know-how to compete for AI leadership. Their access to frontier AI will instead be determined by their relationships with a handful of already rich and powerful corporations and states. This dependence threatens to aggravate current geopolitical power imbalances. The most powerful governments will vie to control the world’s most valuable resource while, once again, countries in the global South will be left behind. This is not to say that only the richest will benefit from the AI revolution. Like the Internet and smartphones, AI will proliferate without respect for borders, as will the productivity gains it unleashes. And like energy and green technology, AI will benefit many countries that do not control it, including those that contribute to producing AI inputs such as semiconductors.

At the other end of the geopolitical spectrum, however, the competition for AI supremacy will be fierce. At the end of the Cold War, powerful countries might have cooperated to allay one another’s fears and arrest a potentially destabilizing technological arms race. But today’s tense geopolitical environment makes such cooperation much harder. AI is not just another tool or weapon that can bring prestige, power, or wealth. It has the potential to enable a significant military and economic advantage over adversaries. Rightly or wrongly, the two players that matter most—China and the United States—both see AI development as a zero-sum game that will give the winner a decisive strategic edge in the decades to come.

China and the United States both see AI development as a zero-sum game.
From the vantage point of Washington and Beijing, the risk that the other side will gain an edge in AI is greater than any theoretical risk the technology might pose to society or to their own domestic political authority. For that reason, both the U.S. and Chinese governments are pouring immense resources into developing AI capabilities while working to deprive each other of the inputs needed for next-generation breakthroughs. (So far, the United States has been far more successful than China in doing the latter, especially with its export controls on advanced semiconductors.) This zero-sum dynamic—and the lack of trust on both sides—means that Beijing and Washington are focused on accelerating AI development, rather than slowing it down. In their view, a “pause” in development to assess risks, as some AI industry leaders have called for, would amount to foolish unilateral disarmament.

But this perspective assumes that states can assert and maintain at least some control over AI. This may be the case in China, which has integrated its tech companies into the fabric of the state. Yet in the West and elsewhere, AI is more likely to undermine state power than to bolster it. Outside China, a handful of large, specialist AI companies currently control every aspect of this new technological wave: what AI models can do, who can access them, how they can be used, and where they can be deployed. And because these companies jealously guard their computing power and algorithms, they alone understand (most of) what they are creating and (most of) what those creations can do. These few firms may retain their advantage for the foreseeable future—or they may be eclipsed by a raft of smaller players as low barriers to entry, open-source development, and near-zero marginal costs lead to uncontrolled proliferation of AI. Either way, the AI revolution will take place outside government.

To a limited degree, some of these challenges resemble those of earlier digital technologies. Internet platforms, social media, and even devices such as smartphones all operate, to some extent, within sandboxes controlled by their creators. When governments have summoned the political will, they have been able to implement regulatory regimes for these technologies, such as the EU’s General Data Protection Regulation, Digital Markets Act, and Digital Services Act. But such regulation took a decade or more to materialize in the EU, and it still has not fully materialized in the United States. AI moves far too quickly for policymakers to respond at their usual pace. Moreover, social media and other older digital technologies do not help create themselves, and the commercial and strategic interests driving them never dovetailed in quite the same way: Twitter and TikTok are powerful, but few think they could transform the global economy.

This all means that at least for the next few years, AI’s trajectory will be largely determined by the decisions of a handful of private businesses, regardless of what policymakers in Brussels or Washington do. In other words, technologists, not policymakers or bureaucrats, will exercise authority over a force that could profoundly alter both the power of nation-states and how they relate to each other. That makes the challenge of governing AI unlike anything governments have faced before, a regulatory balancing act more delicate—and more high stakes—than any policymakers have attempted.

MOVING TARGET, EVOLVING WEAPON
Governments are already behind the curve. Most proposals for governing AI treat it as a conventional problem amenable to the state-centric solutions of the twentieth century: compromises over rules hashed out by political leaders sitting around a table. But that will not work for AI.

Regulatory efforts to date are in their infancy and still inadequate. The EU’s AI Act is the most ambitious attempt to govern AI in any jurisdiction, but it will apply in full only beginning in 2026, by which time AI models will have advanced beyond recognition. The United Kingdom has proposed an even looser, voluntary approach to regulating AI, but it lacks the teeth to be effective. Neither initiative attempts to govern AI development and deployment at the global level—something that will be necessary for AI governance to succeed. And while voluntary pledges to respect AI safety guidelines, such as those made in July by seven leading AI developers, including Inflection AI, led by one of us (Suleyman), are welcome, they are no substitute for legally binding national and international regulation.

Advocates for international-level agreements to tame AI tend to reach for the model of nuclear arms control. But AI systems are not only infinitely easier to develop, steal, and copy than nuclear weapons; they are controlled by private companies, not governments. As the new generation of AI models diffuses faster than ever, the nuclear comparison looks ever more out of date. Even if governments can successfully control access to the materials needed to build the most advanced models—as the Biden administration is attempting to do by preventing China from acquiring advanced chips—they can do little to stop the proliferation of those models once they are trained and therefore require far fewer chips to operate.

For global AI governance to work, it must be tailored to the specific nature of the technology, the challenges it poses, and the structure and balance of power in which it operates. But because the evolution, uses, risks, and rewards of AI are unpredictable, AI governance cannot be fully specified at the outset—or at any point in time, for that matter. It must be as innovative and evolutionary as the technology it seeks to govern, sharing some of the characteristics that make AI such a powerful force in the first place. That means starting from scratch, rethinking and rebuilding a new regulatory framework from the ground up.

The overarching goal of any global AI regulatory architecture should be to identify and mitigate risks to global stability without choking off AI innovation and the opportunities that flow from it. Call this approach “technoprudentialism,” a mandate rather like the macroprudential role played by global financial institutions such as the Financial Stability Board, the Bank of International Settlements, and the International Monetary Fund. Their objective is to identify and mitigate risks to global financial stability without jeopardizing economic growth.

A technoprudential mandate would work similarly, necessitating the creation of institutional mechanisms to address the various aspects of AI that could threaten geopolitical stability. These mechanisms, in turn, would be guided by common principles that are both tailored to AI’s unique features and reflect the new technological balance of power that has put tech companies in the driver’s seat. These principles would help policymakers draw up more granular regulatory frameworks to govern AI as it evolves and becomes a more pervasive force.

The first and perhaps most vital principle for AI governance is precaution. As the term implies, technoprudentialism is at its core guided by the precautionary credo: first, do no harm. Maximally constraining AI would mean forgoing its life-altering upsides, but maximally liberating it would mean risking all its potentially catastrophic downsides. In other words, the risk-reward profile for AI is asymmetric. Given the radical uncertainty about the scale and irreversibility of some of AI’s potential harms, AI governance must aim to prevent these risks before they materialize rather than mitigate them after the fact. This is especially important because AI could weaken democracy in some countries and make it harder for them to enact regulations. Moreover, the burden of proving an AI system is safe above some reasonable threshold should rest on the developer and owner; it should not be solely up to governments to deal with problems once they arise.

AI governance must also be agile so that it can adapt and correct course as AI evolves and improves itself. Public institutions often calcify to the point of being unable to adapt to change. And in the case of AI, the sheer velocity of technological progress will quickly overwhelm the ability of existing governance structures to catch up and keep up. This does not mean that AI governance should adopt the “move fast and break things” ethos of Silicon Valley, but it should more closely mirror the nature of the technology it seeks to contain.

In addition to being precautionary and agile, AI governance must be inclusive, inviting the participation of all actors needed to regulate AI in practice. That means AI governance cannot be exclusively state centered, since governments neither understand nor control AI. Private technology companies may lack sovereignty in the traditional sense, but they wield real—even sovereign—power and agency in the digital spaces they have created and effectively govern. These nonstate actors should not be granted the same rights and privileges as states, which are internationally recognized as acting on behalf of their citizens. But they should be parties to international summits and signatories to any agreements on AI.

Such a broadening of governance is necessary because any regulatory structure that excludes the real agents of AI power is doomed to fail. In previous waves of tech regulation, companies were often afforded so much leeway that they overstepped, leading policymakers and regulators to react harshly to their excesses. But this dynamic benefited neither tech companies nor the public. Inviting AI developers to participate in the rule-making process from the outset would help establish a more collaborative culture of AI governance, reducing the need to rein in these companies after the fact with costly and adversarial regulation.

AI is a problem of the global commons, not just the preserve of two superpowers.
Tech companies should not always have a say; some aspects of AI governance are best left to governments, and it goes without saying that states should always retain final veto power over policy decisions. Governments must also guard against regulatory capture to ensure that tech companies do not use their influence within political systems to advance their interests at the expense of the public good. But an inclusive, multistakeholder governance model would ensure that the actors who will determine the fate of AI are involved in—and bound by—the rule-making processes. In addition to governments (especially but not limited to China and the United States) and tech companies (especially but not limited to the Big Tech players), scientists, ethicists, trade unions, civil society organizations, and other voices with knowledge of, power over, or a stake in AI outcomes should have a seat at the table. The Partnership on AI—a nonprofit group that convenes a range of large tech companies, research institutions, charities, and civil society organizations to promote responsible AI use—is a good example of the kind of mixed, inclusive forum that is needed.

AI governance must also be as impermeable as possible. Unlike climate change mitigation, where success will be determined by the sum of all individual efforts, AI safety is determined by the lowest common denominator: a single breakout algorithm could cause untold damage. Because global AI governance is only as good as the worst-governed country, company, or technology, it must be watertight everywhere—with entry easy enough to compel participation and exit costly enough to deter noncompliance. A single loophole, weak link, or rogue defector will open the door to widespread leakage, bad actors, or a regulatory race to the bottom.

In addition to covering the entire globe, AI governance must cover the entire supply chain—from manufacturing to hardware, software to services, and providers to users. This means technoprudential regulation and oversight along every node of the AI value chain, from AI chip production to data collection, model training to end use, and across the entire stack of technologies used in a given application. Such impermeability will ensure there are no regulatory gray areas to exploit.

Finally, AI governance will need to be targeted, rather than one-size-fits-all. Because AI is a general-purpose technology, it poses multidimensional threats. A single governance tool is not sufficient to address the various sources of AI risk. In practice, determining which tools are appropriate to target which risks will require developing a living and breathing taxonomy of all the possible effects AI could have—and how each can best be governed. For example, AI will be evolutionary in some applications, exacerbating current problems such as privacy violations, and revolutionary in others, creating entirely new harms. Sometimes, the best place to intervene will be where data is being collected. Other times, it will be the point at which advanced chips are sold—ensuring they do not fall into the wrong hands. Dealing with disinformation and misinformation will require different tools than dealing with the risks of AGI and other uncertain technologies with potentially existential ramifications. A light regulatory touch and voluntary guidance will work in some cases; in others, governments will need to strictly enforce compliance.

All of this requires deep understanding and up-to-date knowledge of the technologies in question. Regulators and other authorities will need oversight of and access to key AI models. In effect, they will need an audit system that can not only track capabilities at a distance but also directly access core technologies, which in turn will require the right talent. Only such measures can ensure that new AI applications are proactively assessed, both for obvious risks and for potentially disruptive second- and third-order consequences. Targeted governance, in other words, must be well-informed governance.

THE TECHNOPRUDENTIAL IMPERATIVE
Built atop these principles should be a minimum of three AI governance regimes, each with different mandates, levers, and participants. All will have to be novel in design, but each could look for inspiration to existing arrangements for addressing other global challenges—namely, climate change, arms proliferation, and financial stability.

The first regime would focus on fact-finding and would take the form of a global scientific body to objectively advise governments and international bodies on questions as basic as what AI is and what kinds of policy challenges it poses. If no one can agree on the definition of AI or the possible scope of its harms, effective policymaking will be impossible. Here, climate change is instructive. To create a baseline of shared knowledge for climate negotiations, the United Nations established the Intergovernmental Panel on Climate Change and gave it a simple mandate: provide policymakers with “regular assessments of the scientific basis of climate change, its impacts and future risks, and options for adaptation and mitigation.” AI needs a similar body to regularly evaluate the state of AI, impartially assess its risks and potential impacts, forecast scenarios, and consider technical policy solutions to protect the global public interest. Like the IPCC, this body would have a global imprimatur and scientific (and geopolitical) independence. And its reports could inform multilateral and multistakeholder negotiations on AI, just as the IPCC’s reports inform UN climate negotiations.

The world also needs a way to manage tensions between the major AI powers and prevent the proliferation of dangerous advanced AI systems. The most important international relationship in AI is the one between the United States and China. Cooperation between the two rivals is difficult to achieve under the best of circumstances. But in the context of heightened geopolitical competition, an uncontrolled AI race could doom all hope of forging an international consensus on AI governance. One area where Washington and Beijing may find it advantageous to work together is in slowing the proliferation of powerful systems that could imperil the authority of nation-states. At the extreme, the threat of uncontrolled, self-replicating AGIs—should they be invented in the years to come—would provide strong incentives to coordinate on safety and containment.

On all these fronts, Washington and Beijing should aim to create areas of commonality and even guardrails proposed and policed by a third party. Here, the monitoring and verification approaches often found in arms control regimes might be applied to the most important AI inputs, specifically those related to computing hardware, including advanced semiconductors and data centers. Regulating key chokepoints helped contain a dangerous arms race during the Cold War, and it could help contain a potentially even more dangerous AI race now.

Few powerful constituencies favor containing AI.
But since much of AI is already decentralized, it is a problem of the global commons rather than the preserve of two superpowers. The devolved nature of AI development and core characteristics of the technology, such as open-source proliferation, increase the likelihood that it will be weaponized by cybercriminals, state-sponsored actors, and lone wolves. That is why the world needs a third AI governance regime that can react when dangerous disruptions occur. For models, policymakers might look to the approach financial authorities have used to maintain global financial stability. The Financial Stability Board, composed of central bankers, ministries of finance, and supervisory and regulatory authorities from around the world, works to prevent global financial instability by assessing systemic vulnerabilities and coordinating the necessary actions to address them among national and international authorities. A similarly technocratic body for AI risk—call it the Geotechnology Stability Board—could work to maintain geopolitical stability amid rapid AI-driven change. Supported by national regulatory authorities and international standard-setting bodies, it would pool expertise and resources to preempt or respond to AI-related crises, reducing the risk of contagion. But it would also engage directly with the private sector, recognizing that key multinational technology actors play a critical role in maintaining geopolitical stability, just as systemically important banks do in maintaining financial stability.

Such a body, with authority rooted in government support, would be well positioned to prevent global tech players from engaging in regulatory arbitrage or hiding behind corporate domiciles. Recognizing that some technology companies are systemically important does not mean stifling start-ups or emerging innovators. On the contrary, creating a single, direct line from a global governance body to these tech behemoths would enhance the effectiveness of regulatory enforcement and crisis management—both of which benefit the whole ecosystem.

A regime designed to maintain geotechnological stability would also fill a dangerous void in the current regulatory landscape: responsibility for governing open-source AI. Some level of online censorship will be necessary. If someone uploads an extremely dangerous model, this body must have the clear authority—and ability—to take it down or direct national authorities to do so. This is another area for potential bilateral cooperation. China and the United States should want to work together to embed safety constraints in open-source software—for example, by limiting the extent to which models can instruct users on how to develop chemical or biological weapons or create pandemic pathogens. In addition, there may be room for Beijing and Washington to cooperate on global antiproliferation efforts, including through the use of interventionist cybertools.

Each of these regimes would have to operate universally, enjoying the buy-in of all major AI players. The regimes would need to be specialized enough to cope with real AI systems and dynamic enough to keep updating their knowledge of AI as it evolves. Working together, these institutions could take a decisive step toward technoprudential management of the emerging AI world. But they are by no means the only institutions that will be needed. Other regulatory mechanisms, such as “know your customer” transparency standards, licensing requirements, safety testing protocols, and product registration and approval processes, will need to be applied to AI in the next few years. The key across all these ideas will be to create flexible, multifaceted governance institutions that are not constrained by tradition or lack of imagination—after all, technologists will not be constrained by those things.

PROMOTE THE BEST, PREVENT THE WORST
None of these solutions will be easy to implement. Despite all the buzz and chatter coming from world leaders about the need to regulate AI, there is still a lack of political will to do so. Right now, few powerful constituencies favor containing AI—and all incentives point toward continued inaction. But designed well, an AI governance regime of the kind described here could suit all interested parties, enshrining principles and structures that promote the best in AI while preventing the worst. The alternative—uncontained AI—would not just pose unacceptable risks to global stability; it would also be bad for business and run counter to every country’s national interest.

A strong AI governance regime would both mitigate the societal risks posed by AI and ease tensions between China and the United States by reducing the extent to which AI is an arena—and a tool—of geopolitical competition. And such a regime would achieve something even more profound and long-lasting: it would establish a model for how to address other disruptive, emerging technologies. AI may be a unique catalyst for change, but it is by no means the last disruptive technology humanity will face. Quantum computing, biotechnology, nanotechnology, and robotics also have the potential to fundamentally reshape the world. Successfully governing AI will help the world successfully govern those technologies as well.

The twenty-first century will throw up few challenges as daunting or opportunities as promising as those presented by AI. In the last century, policymakers began to build a global governance architecture that, they hoped, would be equal to the tasks of the age. Now, they must build a new governance architecture to contain and harness the most formidable, and potentially defining, force of this era. The year 2035 is just around the corner. There is no time to waste.