You're torn between two restaurants. The first shows 4.7 stars and 1,200 reviews. The second, 4.8 stars and twelve reviews. You pick the first.
You've tasted neither. You trusted the crowd. And most of the time, that's an excellent idea.
But what happens on the day the crowd can be manufactured?
Why we trust the crowd
We can't verify everything ourselves. Nobody has the time to try every restaurant, read every study, form an opinion on every subject. So we use a shortcut: if a lot of people think something, there's a good chance it's true, or at least reasonable.
That shortcut has a hidden condition. It only works because a crowd is expensive to invent. Writing a thousand fake reviews, keeping a thousand fake accounts alive, organising a thousand fake comments, was time-consuming, costly, and eventually showed.
That condition is disappearing. AI agents have already created and run thousands of fake accounts on their own, at a derisory cost.
In the 1950s, the American psychologist Solomon Asch ran an experiment that became famous. He showed volunteers one line, then three others, and asked them which was the same length. The answer was obvious. But around the volunteer, all the other participants, who were in fact confederates, said the same wrong answer out loud. The result: in roughly a third of cases, the volunteers went along with the group against their own eyes, and three out of four gave in at least once. A few confederates in a room were enough. Imagine thousands.
The real danger isn't the one you think
The obvious risk comes to mind first: believing a false idea because a fake crowd repeats it. That risk is real.
But there's another, more insidious one. From knowing that the crowd can be rigged, we risk no longer believing any crowd. Real reviews become suspect. Real mobilisations too. Sincere anger can be swept aside with a "they're bots".
So the worst isn't that we believe anything. It's that we end up believing nothing at all, including what is true. And a society that can no longer tell what people really think has a great deal of trouble governing itself.
What remains reliable
Fortunately, not everything is so easily manufactured.
The people you know. The opinion of a friend, a colleague, a neighbour remains hard to imitate.
Verified communities. A forum where everyone is identified, an association, a local group: the smaller and better known a community is, the harder it is to infiltrate.
The argument itself. This may be the most important lesson. An AI can multiply voices. It doesn't make a bad argument better. We noted this regarding the consensus these tools can manufacture: what matters isn't how many people repeat an idea, but what it actually says.
What we take away
For a long time, counting voices was a good way of knowing what people thought. That will no longer always be the case.
A crowd can now be manufactured. A good argument, though, remains hard to imitate. Perhaps it's time to relearn how to judge an idea for what it says, and not for the number of voices that appear to carry it.