The hyperinflation of arguments

by Miguel Lucas

Not long ago, a solid argument came at a price: hours, reading, discarded ideas. Today it’s generated for free, in five seconds, in whatever ideological color you need. Welcome to the hyperinflation of arguments.

The data is blunt. A study by EPFL, Princeton, and the Fondazione Bruno Kessler published in Nature Human Behaviour pitted GPT-4 against real human debaters 1. On a level playing field, a technical draw. But when the AI was given basic information about its opponent —age, education, ideology— it beat the humans in 64.4% of the debates. It didn’t win by being smarter: it won by adapting faster than a human can process in real time.

The effect is even more brutal in the labor market. A study using data from Freelancer.com shows that before generative AI, writing a carefully crafted, personalized proposal predicted whether a candidate was any good, and clients paid a premium for it: $25.67 on average 2. Once anyone could generate that same proposal with AI in seconds, the premium fell to $14.85, and the proposal stopped predicting anything at all. The result: the best candidates were hired 19% less often, and the worst ones, armed with the perfect AI-made cover letter, 14% more. The quality of the text stopped meaning the quality of the person.

If an intern handed you a flawless analysis five seconds after you asked, without having read the file, you’d be suspicious. When AI does it, we call it productivity. We still treat “having an argument” as proof that someone has thought, when all it proves now is that someone pressed a button. Meanwhile, nearly half of all new content on the internet is already written by a machine 3.

This is the flip side of the authenticity economy: if what gets cheap there is the copy, what gets cheap here is reasoning itself. And what’s left scarce isn’t the argument — it’s the signature behind it: whoever stands by it with something to lose if they’re wrong.

So the question is no longer whether you have an argument. Anyone has one now. The question is: if producing an argument costs nothing, what are you going to use to decide who deserves your trust?

Related theses

References

  1. Gizmodo — AI Gets a Lot Better at Debating When It Knows Who You Are, Study Finds (EPFL, Princeton & FBK, Nature Human Behaviour)
  2. Jesse Silbert — Making Talk Cheap: Generative AI and Labor Market Signaling
  3. Graphite — AI Now Writes as Many Online Articles as Humans Do