Human Nature Reserves

Yesterday, as reported by Axios,¹ Bill Gates proposed creating jobs reserved for humans: work that society would deliberately leave to people even when AI or robots could perform it. He compared the idea to nature reserves, where we choose not to develop something because we believe its preservation matters. His examples include childcare and jury service, while on education and healthcare he argued against banning AI outright.²

»Perhaps the danger of automation isn’t that machines take the things humans are good at, but that we automate them before we understand what they are«

The inversion is useful. Before deciding which parts of human life should remain human, perhaps we should establish what humans are actually good at.

We know what machines are good at. Calculators beat us at arithmetic. Computers remember more than we do. Robots can repeat movements without getting tired. AI can search, compress, and recombine information at a scale no human can approach.

But the frontier is strange. Computer scientist Hans Moravec famously observed that the things humans find intellectually difficult can be surprisingly easy to automate, while apparently trivial things such as perception and movement can be extraordinarily difficult.³ Artificial intelligence doesn’t simply take over human abilities from easiest to hardest. It can be extraordinarily good at things we find difficult, while struggling with things we find trivial. So where is the human advantage?

One possibility is social intelligence. Anthropologist-psychologist Robin Dunbar’s social-brain hypothesis proposes that primate brain size is strongly related to the size of the social groups animals can maintain.⁴ This could mean that human intelligence evolved not primarily to calculate, but to keep track of other people: who is loyal to whom, who owes what to whom, who can be trusted, who is falling out with whom. Gossip may be one of the oldest information technologies.

Another possibility is the ability to create new spaces of possibility. AI researcher Margaret Boden distinguished between combinatorial and exploratory creativity, which find new combinations within an existing space, and transformational creativity, which changes the rules defining the space itself.⁵ A machine can be extraordinarily good at searching a landscape. It’s less obvious how we should ask it to decide which landscape to search through.

A third human advantage is long-horizon agency. Humans can spend years pursuing projects whose purpose changes as they pursue them. We don’t merely optimise towards a goal. Sometimes we discover what the goal was by trying to reach it.

But none of these is guaranteed to remain uniquely human. And that’s precisely the point: we should find out. But there are really two questions here. One is economic: what are humans better at? The other is philosophical: what do we want to remain human even if machines become better at it?

There’s another reason to be careful. Economics has long given us a comforting answer to technological displacement. Economist David Ricardo’s theory of comparative advantage shows why two parties can benefit from specialising even when one is better at producing everything.⁶ A displaced worker might not have to be better than a machine at anything; there only needs to be some activity in which human time is relatively more valuable.

But AI changes the geometry of that argument. Human labour is rivalrous: if I spend an hour doing one thing, I cannot spend the same hour doing another. Software is replicable. A successful AI system can potentially perform the same task thousands or millions of times in parallel. Its capacity can still be scarce, but it’s not scarce in quite the same way a human hour is.

If that difference becomes economically important, comparative advantage may no longer provide the safety net it once did. Which makes Gates’s nature-reserve metaphor more interesting. Perhaps we will eventually need reserves not because humans are sacred, but because some human capabilities are valuable and cannot easily be recovered once they disappear.

Maybe the danger of automation isn’t that machines take the things humans are good at, but that we automate them before we understand what they are.

There’s also a deeper possibility. Perhaps asking what humans are good at is actually the wrong question. Philosopher Hannah Arendt distinguished between labour, work, and action: between producing things, making a world, and acting with other people. Action, for Arendt, was not valuable because it produced something useful. It was part of what it meant to be human.⁷

Maybe that’s the real problem with a human-reserve approach. We could become so concerned with preserving uniquely human capabilities that we forget to preserve activities that still seem uniquely human to us: things people do not because they are better at them, but because doing them is part of what makes a human life worth living.

Perhaps we shouldn’t ask which human abilities machines should leave alone, but which parts of being human were never supposed to justify themselves by usefulness at all. I’m not sure that list is very long. I’m not even sure I can name three.

Can you?

References