One essential ingredient for AI safety? Democracy

‘To avoid widespread political disempowerment, we will need to renew our traditions of self-government.’ Photograph: Joseph Prezioso/AFP/Getty ImagesView image in fullscreen‘To avoid widespread political disempowerment, we will need to renew our traditions of self-government.’ Photograph: Joseph Prezioso/AFP/Getty ImagesOne essential ingredient for AI safety? DemocracyMatthew BotvinickThe rise of AI makes the preservation of democracy even more urgent – especially as the technology threatens institutions

AI safety is having a moment. After years as a topic for techies and futurists, the potential risks from artificial intelligence systems – from cyberattacks to hyper-persuasion to rogue behavior – are suddenly under discussion both in the corridors of power and around kitchen tables.

In fact, the conversation now reaches beyond classical safety concerns, getting into AI’s impact on economic life and the environment, as well as possible responses, from alignment research to government regulation.

This expanding public discussion is coming none too soon, as recent news about AI security breaches makes clear. However, something important is still missing. For all its vibrancy, the public debate has not yet focused on one of the most important requirements for AI safety: constitutional democracy.

To see why democracy matters for AI safety, imagine that we have already built a powerful AI system that is “safe” in all the now-familiar senses: it cannot escape human control, cannot launch cyberattacks or make bioweapons, and cannot power killer robots. Now picture that system in two different political contexts, one democratic, the other authoritarian.

Placed inside a healthy democracy – a constitutional republic, with free and fair elections, protected civil liberties, and robust rule of law – AI could bring massive benefits, advancing science and medicine, expanding education and bolstering economic prosperity. Under the right conditions, AI could strengthen democracy itself, helping citizens make their voices heard, improving policymaking and making government services more efficient and accountable.

Read moreNow consider the authoritarian scenario. Here, everything flips. AI becomes a tool for government surveillance and censorship, for propaganda and for political suppression and entrenchment. No imagination is required to conjure this scenario. It is already playing out in China, Russia, Iran, Venezuela, several Gulf states and elsewhere.

Placing these two scenarios side by side reveals an elemental truth: there can be no such thing as safe AI unless we secure the future of constitutional democracy and the rule of law. It doesn’t matter how “safe” your AI is if you live under authoritarian rule; you will not be safe. In short, AI safety requires political freedom.

Once this fact sinks in, it becomes clear that we must reverse the retreat of democracy that has occurred around the world over the past two decades, including, here at home, the steady erosion of constitutional checks under administrations of both parties.

Of course, preserving democracy has always been a worthwhile goal. With the AI era looming, it has become an urgent one.

To complicate matters, AI itself will put new pressure on democratic institutions. As AI starts to affect jobs, the resulting economic insecurity may create openings for demagogues of every stripe.

By creating national security crises – cyberattacks, for example – AI will multiply opportunities for governments to invoke emergency powers, powers that, once claimed, are rarely relinquished. And as AI is integrated into executive-branch functions, traditional checks and balances may be disturbed, and the risk of political surveillance will grow.

In the longer term, if AI replaces a majority of jobs – an outcome that economists such as Daniel Susskind and Anton Korinek are concerned about – ordinary citizens will face a serious risk of losing their leverage over those who govern them. To see why, we need only look at present-day petrostates, where governments that do not depend on taxing their citizens’ work often disregard those same citizens’ rights.

A solid first step toward countering these risks would be to update our legal codes, setting limits on executive power by reforming emergency and surveillance law, and ensuring due process by narrowing the circumstances in which the state secrets privilege may be invoked.

In the US, this would mean, for example, revisiting the National Emergencies Act and International Emergency Economic Powers Act, and closing the “data broker” loophole in intelligence work. As executive agencies begin to leverage AI, we should also furnish Congress and the courts with equally powerful AI tools, so that their oversight can keep pace.

The principle here is not new. As James Madison wrote, “Ambition must be made to counteract ambition,” only here in a digital setting.

Longer-term, to avoid widespread political disempowerment, we will need to renew our traditions of self-government, involving citizens more directly in policymaking and government oversight. If we play our cards right, AI itself may help, making information easier to access and helping citizens reason together – town halls for a digital age.

Buttressing political freedom in these ways would be important in its own right. Today, however, it is also a form of AI safety work.

Who is best positioned to drive the needed reforms? Policymakers, civic organizations and ordinary citizens must all participate.

But there is another key player: the AI industry. While the major AI labs are willing to weigh in on specific policy issues, most have recently gone quiet on the topic of democracy itself.

Over the past year, a number of AI-industry leaders have justifiably stressed the need to equip military and intelligence agencies with AI, so that democracies can compete abroad. But the same leaders have been conspicuously absent from the debate over the health of democracy at home.

The tech sector was not always so reserved. In 2013, Apple, Google, Microsoft, Facebook, Twitter, Yahoo, LinkedIn and AOL joined forces to publish an open letter to the president and Congress, stating that government surveillance methods had “tipped too far in favor of the state and away from the rights of the individual – rights that are enshrined in our constitution”.

Read moreIn 2018, Microsoft publicly asked Congress to limit government use of facial recognition technology, writing: “We must ensure that the year 2024 doesn’t look like a page from the novel 1984.” And as recently as 2021, companies including Alphabet, Amazon, Apple, Facebook, IBM, Microsoft and Twitter published an open letter opposing proposed changes to state voting laws, declaring: “We stand for democracy.”

More recently, however, many tech companies have fallen silent. This is understandable. These companies answer to their investors, and in the current environment, speaking out can put shareholder value at risk.

That, however, should not be the end of the story. The leading AI labs all describe AI safety as core to their mission. If they mean what they say, then – for the reasons I’ve highlighted – they bear a duty to stand up for basic constitutional principles, despite the commercial risks.

Journalists, researchers and ordinary citizens can help them recognize that duty by insistently pointing out the connection between safety and freedom.

It is good that people outside the AI industry are suddenly talking about AI safety. AI will affect everyone, and everyone should be involved in shaping how it is built and used.

But we need to make political freedom a centerpiece of the discussion. We need to build public awareness that AI safety fundamentally depends on democracy. Only by bearing this point firmly in mind can we avoid the dangers of AI and reap its potentially immense benefits.

Matthew Botvinick is the author of the forthcoming book AI and Political Freedom and a senior fellow at Yale Law School. He has held multiple research positions in the AI industry, including at Google DeepMind and currently Anthropic, but writes here exclusively in his Yale role