A problem with many different faces
Human verification isn't a single problem, it's several related problems disguised as one. There's the technical problem (how do we tell generated content apart from human content?). There's the identity problem (how do we confirm an account belongs to a real person and not to thousands of automated ones?). And there's the everyday-experience problem (how do we do this without turning every interaction into a tedious chore?).
Each of these problems has its own community of researchers, engineers, and designers working on partial solutions. But none of them, so far, has an answer that is simultaneously effective, privacy-respecting, and pleasant to use. Usually one of those three has to be sacrificed.
The paths being tried
Different corners of the tech industry are exploring very different paths. Some bet on cryptographic credentials that certify 'this account belongs to a verified person' without revealing exactly who that person is. Others bet on analyzing behavioral patterns (how we type, how we move, how we interact) to detect automation. Others, more directly, ask for identity documents or biometric verification.
Each path brings uncomfortable trade-offs. Cryptographic credentials are elegant in theory but hard to adopt at scale and easy to misunderstand. Behavioral analysis can fail with both false positives and false negatives, and raises its own surveillance questions. Verification through documents or biometric data solves the problem at the root, but at the cost of an amount of personal information many people, reasonably, aren't willing to hand over just to leave a comment on a website.
As far as we know, there's no path that solves this at zero cost. That doesn't mean it isn't worth trying. It means any real solution will have to accept explicit trade-offs, instead of promising there are none.
Why the problem will keep growing, not settle down
It's tempting to think this is a temporary problem, an uncomfortable phase that will resolve once detection technology matures. But there are reasons to doubt that expectation. Content generation and content detection are, structurally, two sides of the same race: every improvement in detection tends to feed, sooner or later, improvements in generation designed to evade it.
This doesn't mean detection is useless. It means we'll probably never reach a final, definitive point, but rather a shifting equilibrium that demands constant attention, the same way information security is never 'solved' once and for all, but sustained through continuous vigilance.
A question that's social too, not only technical
Beyond technology, there's an outstanding social question: what kind of spaces do we want to build while this problem has no complete solution? Spaces where we assume, by default, that anything could be generated, with all the emotional weight that implies? Or smaller spaces, deliberately designed to prioritize the human over scale and volume?
We don't think there's a single correct answer to that question. Large platforms need scale, and so they'll mostly have to bet on technical, large-scale solutions. But there can also be room for smaller spaces that solve the problem differently: by prioritizing what's ephemeral, direct, and moderated in real time over sheer volume.
Our modest, partial bet
We don't have the definitive technical solution to human verification, and it would be dishonest to suggest otherwise. What we did instead was make a modest bet: build a live conversation space, with ephemeral messages and pre-publication moderation, where we prioritize direct interaction over volume and permanence.
It isn't a solution to the general problem. It's simply our concrete way of trying to make sure that, at least in this small corner of the internet, talking to another person still feels like talking to another person.
