Why Humans Are Selfish: What Science Actually Says

Put four strangers in a room. Give each of them ten pounds. Anything they put in a shared pot gets multiplied and split between everyone.

So if we all put in everything, we all come out ahead.

Right. And if you put in nothing while everyone else gives, you do best of all. That's a public goods game, and it's been run thousands of times.

Let me guess, everyone's a saint for about two rounds.

Almost exactly. People start out generous, and cooperation collapses as the rounds go on. That collapse is the thing the whole debate about human selfishness hangs on.

Why? Doesn't collapsing just mean we're selfish and slow to realise it?

That's one reading. The other is that we really are kind, and we only stop when we feel exploited. Same data, two completely opposite stories about human nature.

Okay, so how do you tell them apart?

There's a trick. You let cooperation die, then you announce a surprise restart. Same people, same game, fresh set of rounds. What do you think happens?

Nothing? They've already learned the lesson.

Cooperation jumps back up. It's called the restart effect, and for years it was treated as proof that something warm and non-selfish is in there.

Huh. That is a bit odd.

So a team published in PNAS in twenty twenty-two tested it two ways. First, they made contributions invisible, so you couldn't influence anyone. The restart effect vanished.

Meaning people were only being generous to nudge others into being generous back.

Strategically, yes. Then the second test. They had people play with computer players, and told them they were computers.

And people still restarted? Against machines?

They did. They tracked the computers' cooperation up and down. Only about twenty-two percent of people played the rational thing and contributed zero every round.

So most people were being nice to a program. That's not altruism, that's just... not following the instructions.

The authors call it confusion. Their conclusion is that these games have often been measuring self-interested but confused behaviour, not some uniquely human generosity.

Honestly, that's the part I'd have got wrong. I'd have said the games prove we're kind deep down.

And yet, don't swing all the way the other way. A meta-study of ninety-six prisoner's dilemma studies, over three and a half thousand people, found cooperation rates averaging about thirty-five percent.

Thirty-five is a lot more than zero.

And it swings wildly, from four percent to eighty-four. What predicts it isn't how nice the people are. In one-shot games it's risk, meaning how much you lose if you cooperate and they don't.

So the same person is generous or stingy depending on what's at stake.

Change the structure, change the human. A huge meta-analysis across six decades, over two hundred thousand participants, found the two things that lift cooperation most are being able to punish or reward people, and being allowed to talk to each other.

Talking. Just talking.

Just talking.

But hang on. Does any of this survive outside a lab with money on a screen? People jump into rivers for strangers.

Which brings in the biology, and that's where the word altruism means something stricter. To a biologist, altruism is a behaviour that lowers your own fitness and raises someone else's.

Fitness being, what, number of descendants?

Roughly, yes. And that's a genuine puzzle, because cheats who take the benefit without paying the cost should wipe altruists out.

So why hasn't that happened?

Hamilton's rule. Altruism is favoured when r b minus c is greater than zero. R is how related you are, b is the benefit to them, c is the cost to you.

So helping your sister is partly helping copies of your own genes.

From the gene's point of view, the altruistic act is selfish. Which is why "selfish gene" never meant selfish people. The gene's accounting and your feelings are two different ledgers.

That distinction's been muddled my whole life.

Try one. You're in a one-shot game with a stranger, no names, no chance to talk, and if you cooperate and they don't, you lose a lot. Predict the cooperation rate.

High risk, no communication, no punishment... low. Near the bottom of that range. Ten, fifteen percent?

About right for that end. Add a chat window and the ability to punish defectors, and you climb towards the top of it. Same species, different room.

So the honest answer to "are humans selfish" is... it depends on the room.

Partly. The lab games look less altruistic than people claimed, once you strip out strategy and confusion. The biology explains real self-sacrifice through relatedness and cost. And the structure around you moves the number more than your character does.

I'm going to be insufferable next time someone says the selfish gene thing at dinner.

Just let them finish first.