Could AI Really Govern Us Better Than Humans?

The idea of artificial intelligence governing human beings sounds, at first, like the plot of a dystopian science-fiction film. We tend to assume that government is fundamentally a human activity: complicated, emotional, political and ultimately dependent on human judgement. But there is an awkward question sitting underneath that assumption: how well are humans actually doing at governing themselves?

Humans in charge: the results have been decidedly mixed

For thousands of years, societies have experimented with kings, emperors, dictatorships, democracies, revolutions, parliaments and countless political systems in between. The results have been decidedly mixed. Governments have produced extraordinary achievements, but they have also produced wars, corruption, inequality, economic crises and decisions that benefit short-term political interests at the expense of future generations. Human beings are intelligent, but we are also tribal, emotional, easily influenced and remarkably good at putting our own interests ahead of everyone else's.

An advanced AI could potentially offer some interesting alternatives. Unlike a politician, an AI would not need to win an election every four or five years. It would not need campaign donations, flattering headlines or popularity ratings. In principle, it could analyse enormous quantities of economic, environmental and social data simultaneously and model the likely consequences of different policies over decades rather than electoral cycles.

In principle, AI could analyse enormous quantities of economic, environmental and social data simultaneously and model the likely consequences of different policies over decades rather than electoral cycles

It could also potentially identify contradictions in our political systems. Governments frequently announce ambitious objectives while pursuing policies that make those objectives harder to achieve. An AI might be able to continuously evaluate whether policies are actually producing their intended results and recommend adjustments accordingly. Imagine a government that effectively had an extraordinarily powerful, impartial research department available 24 hours a day.

Co-governance may be more interesting than outright AI rule

Humans could retain democratic authority — deciding what kind of society they want — while AI systems could help determine how best to achieve those goals. Politicians might set the destination while AI provides the maps, simulations and warnings. Crucially, humans would remain responsible for deciding questions that are fundamentally about values rather than optimisation.

But there is a huge catch. Who decides what the AI is trying to optimise? An AI could potentially design a highly efficient society while pursuing goals that many humans would consider unacceptable. Freedom, equality, privacy, creativity and happiness cannot simply be reduced to numbers without making assumptions about what matters.

There is also the danger of handing responsibility to a machine because humans are tired of making difficult decisions. "The AI says so" could become a remarkably convenient excuse for politicians — and citizens — to avoid responsibility. And if an AI were controlled by a government, corporation or small group of powerful individuals, its apparent neutrality could conceal very human interests behind the technology.

"Because the AI says so" could become a remarkably convenient excuse for politicians

Perhaps the most realistic future is therefore not AI versus democracy, but AI becoming a new layer of democratic decision-making. Humans would still argue, vote and disagree about what kind of world they want. AI could help us understand the consequences of those choices far better than we currently can.

After all, humanity's greatest political problem may not be that we lack intelligence. It may be that we have never before had access to anything remotely intelligent enough to help us deal with the extraordinary complexity of the world we have created.