For centuries we judged a claim partly by how well it was written. A clear, well-argued message counted as evidence that its author knew the subject. AI has ended that. Anyone can now produce a polished message in seconds, so polish says nothing about who wrote it. What is left is slower: understand the claim yourself, then check it yourself.
What Gorgias conceded
In Plato's Gorgias, Socrates puts a question to a celebrated rhetorician. Take a speaker who knows nothing about medicine, stand him next to an actual physician, and put both in front of a crowd of non-doctors. Who wins the room?
Gorgias concedes: the speaker.
That was 2,400 years ago and we never solved it. We built a workaround instead. If a message arrived well-structured and well-argued, we treated that as evidence about the person who sent it. Fluency vouched for competence. Competence vouched for the claim. And the way you wrote your message vouched for you.
Call it prestige. We have run on it ever since. Hiring, funding, boards, politics, the comment under your post.
Sounding credible now costs nothing
That workaround is dead.
A model now hands anyone a well-structured message in seconds. The price of sounding credible went to zero, and the proxy went with it.
Most people read this as a fraud problem. It is worse than that.
It does not only promote people who cannot stand behind their claims. It demotes everyone. If polish tells you nothing about the author, every polished message becomes suspect, including the ones written by people who genuinely know the subject. Twenty years of learning to write clearly used to be an asset. It has been repriced, and the people absorbing that loss did nothing wrong.
Full disclosure: I draft with AI. This post included. So you have no reason to extend me credit for how it reads. That is rather the point.
The two tests that are left
What replaces prestige is slower, and it is not optional.
You separate the message from the messenger. The messenger tells you nothing about himself now. Two tests are left, and both are work.
One. Do you understand it? Not recognise it. Not nod along to the shape of it. Understand it well enough to restate it in your own words to someone who disagrees.
Two. Can you corroborate it? One source that is not the author. That is why I left you a section number. Plato, Gorgias 459a. Go and look. If I have misrepresented what is on that page, discount everything else I just told you.
Call that reasonable fact-checking. Not suspicion of everyone, not trust in no one. A floor, applied to the message, regardless of who is holding it.
Why this is good news
I think this is the end of prestige, and I think it is good news that we will experience as exhausting.
Prestige was never a measure of truth. It was a shortcut that let us outsource judgment to whoever was deemed convincing enough to be an expert. It rewarded the articulate. It filtered out anyone who was right but inelegant, or right in their second language, or right without a title.
That filter is breaking. What comes next asks more of you, because it cannot be delegated: understand the thing yourself, then check it yourself.
The vigilance of the wise. It was always the job. We simply had a very convincing excuse not to do it.
Common questions
What does Plato's Gorgias say about expertise and persuasion?
At 459a, Socrates gets Gorgias to agree that a speaker who is not a doctor can be more convincing than a doctor in front of a crowd. A crowd here means people who do not know medicine. So the person who does not know wins over the people who do not know. Any edition that prints the standard section numbers lets you check the passage.
Why does AI make polished writing less trustworthy?
Polish used to take effort and skill, so it worked as a rough sign of competence. A language model now writes a clear, well-argued text for anyone in seconds. Once anyone can produce the sign, it stops telling you who knows the subject.
How do you check a claim without relying on who wrote it?
Apply two tests to the message itself. First, understand it well enough to restate it in your own words to someone who disagrees. Second, confirm it with at least one source other than the author. If a claim fails either test, discount it, whoever sent it.