The Singularity Is Not an Event

In 1997, a computer beat the world chess champion. Everyone noticed. It was front-page news, a cultural moment, the subject of documentaries and dinner table arguments. And then, almost immediately, the disqualification began. Chess has fixed rules. A computer can brute-force it. Interesting, but not the real thing. Not the Singularity.

Then in 2016, a computer beat the world Go champion. Everyone noticed again. Go is exponentially more complex than chess; the number of possible board positions exceeds the number of atoms in the observable universe. For decades, experts had insisted that Go required genuine intuition, something no algorithm could replicate. When AlphaGo won, the disqualification came anyway. It’s still a board game. Still a closed system. Still not the real Singularity.

This is the pattern. Each time the frontier crosses, we notice, we marvel, and then we move the goalposts. The crossing gets reclassified as a special case: impressive but limited, a narrow achievement that doesn’t really count. The real Singularity, we tell ourselves, is still to come.

I want to suggest the opposite: the Singularity has been happening for decades, domain by domain, crossing by crossing. The reason it doesn’t feel like the Singularity is that we’ve been waiting for a single dramatic moment. That moment will never come, because that is not how this works. The Singularity is not an event. It is a rolling frontier, crossed one domain at a time, at different rates, under different conditions. Some domains have already crossed. Others are crossing now. Others will cross in the years ahead. And a few may never cross at all.

The frontier crosses one domain at a time

The conventional framing of the Singularity imagines a threshold: on one side, humans are smarter; on the other, machines are. Cross the line and everything changes. This framing is clean, dramatic, and wrong. Intelligence is not a single dimension with a single score. It is a vast collection of capabilities, each with its own frontier.

Consider what has already crossed:

• 1997 — Chess. Kasparov versus Deep Blue. The frontier crossed: no human will ever reliably beat the best chess software again.
• 2016 — Go. Lee Sedol versus AlphaGo. The frontier crossed on a problem experts had insisted was beyond algorithmic reach.
• ~2020 — Medical imaging. AI systems began matching and then exceeding radiologists at detecting certain cancers in diagnostic images.
• ~2023 — Protein folding. AlphaFold solved a problem that had stumped structural biology for fifty years.
• Coming — Survey-taking. Legal research. Scientific hypothesis generation. Creative writing. The frontier has not yet been crossed.

Notice the pattern. Each crossing was preceded by experts insisting the domain was different, that it required something AI couldn’t replicate: intuition, creativity, judgment, contextual understanding. Each time, that argument eventually lost. The domains that haven’t crossed yet are not safe because they’re fundamentally different. They’re simply earlier on the same curve.

The frontier is not fixed; it can be engineered

Here is the insight that gets overlooked in most discussions of the Singularity: the frontier isn’t just something that happens to you. It’s something you can influence by design.

Consider survey-taking. In its simplest form (multiple choice questions, fixed responses, no follow-up) the frontier is close. An AI agent can handle this today with reasonable competence. But the moment you add open-ended questions with conversational follow-ups, the task becomes substantially harder. Add voice interaction, and the bar rises again. Add video. Add behavioral biometrics: the specific patterns of how a person moves a mouse, times their keystrokes, hesitates before answering. Add a wearable device with an accelerometer that verifies the respondent is physically present and moving like a human. Each layer pushes the frontier outward. Each one adds something that a pure AI agent cannot yet replicate reliably.

This is not a permanent solution. It is a deferral, a deliberate, engineering-based deferral of the moment the frontier crosses in this particular domain. The AI will eventually replicate mouse movements convincingly. It will eventually handle open-ended follow-ups without telltale patterns. It will eventually do all of it. But “eventually” is not “now”, and the gap between now and eventually is time that can be used productively.

The question is never simply “when will AI surpass humans?” The right question is: in this specific domain, with this specific architecture, where is the line right now, and how far can it be pushed?

What the data looks like at the frontier

The frontier can be observed empirically, not just theorized. When measured rigorously in survey research, one of the domains where the Singularity has not yet been crossed, human respondents (scoring near 0) and autonomous bots (scoring near 1.0) are separated by a 0.54 gap. A detection threshold at 0.5 sits in dead space with zero false positives in either direction. For this domain, with this detection architecture, the frontier has not been crossed.

The 0.54 separation gap is the frontier made visible. Humans are on one side, bots are on the other. The detection threshold sits in empty space, not because the engineering is perfect, but because autonomous AI agents and authentic human respondents currently produce systematically different behavioral signatures. The frontier, in this domain, holds.

What makes this instructive is not the gap itself but what it tells us about the future. Today the separation holds because autonomous agents produce behavioral signatures that are systematically different from humans. These patterns are detectable, as demonstrated in a peer-reviewed randomized trial. But AI capabilities do not stand still. The agents will get progressively better at mimicking the behavioral signatures of humans. When the two populations overlap, the gap closes. That is the domain-specific Singularity crossing, made visible in advance.

The question is not whether the gap will narrow. It will. A paper published in the Proceedings of the National Academy of Sciences already demonstrated that AI agents can evade standard survey detection methods 99.8% of the time when deployed against conventional attention checks. The gap above holds because the detection goes far deeper than attention checks. But the arms race is real and the direction is clear.

The question is never simply “when will AI surpass humans?” The right question is: in this specific domain, with this specific architecture, where is the line right now, and how far can it be pushed?

Where the frontier may never fully cross

The domain-by-domain framing has one important corollary: there may be dimensions of human experience where the question of AI surpassing humans is not simply a matter of timing, but of category. Not because the technology won’t get good enough, but because the value of those interactions is constituted by their human origin.

A study published in Nature Human Behaviour, involving over 6,200 participants across nine experiments, found that people rated identical supportive messages as significantly more caring and empathetic when they believed the message came from a human rather than an AI, even when the actual words were the same. The content didn’t change. The perceived source did. And the perceived source changed everything.

This is not a capability gap that better AI will close. It’s a social fact about what human connection means. People don’t just want to receive warmth; they want to know that another person chose to offer it. AI can simulate the words of camaraderie. It cannot be a member of the social club that gives those words their meaning. It has no skin in the game, no history with you, no stake in the outcome. When people know they’re talking to an AI (and increasingly, they will know) that knowledge changes the nature of the exchange regardless of how fluent the AI becomes.

This doesn’t mean AI can’t be useful in social contexts, or that people won’t form attachments to AI systems. They already do. But it suggests the Singularity framing, the idea that AI will eventually surpass humans at everything in every way that matters, misses something important. For cognitive tasks, the frontier is almost certainly crossable in time. For the domains where human presence is itself the point, the question may not be whether AI can match the performance but whether matching the performance is sufficient.

What this means practically

Dario Amodei, CEO of Anthropic, has argued that we are currently in an era of augmented intelligence, with AI amplifying human capability rather than replacing it, and that this will gradually shift toward fuller automation over the coming years. That trajectory maps directly onto the domain-specific picture: most domains are still in the augmented phase, a smaller number have fully crossed, and a larger number are still ahead.

The useful question to ask about any domain is therefore not “will AI surpass humans here?” but “where is the frontier right now, and what choices move it?” That is an engineering question, and engineering questions have answers. Adding a layer of behavioral biometrics moves the frontier outward. Removing a quality control moves it inward. Every design decision is implicitly a position on the frontier, whether or not it is recognized as one.

The Singularity you’re already living in

The reason the Singularity never feels like it’s arrived is that we keep expecting it to announce itself all at once. It won’t. It’s arriving continuously, in specific places, at specific times, in a rolling progression that has been underway for decades and will continue for decades more. Chess arrived in 1997. Protein folding arrived in 2020. Survey-taking hasn’t arrived yet, but the gap will narrow as autonomous agents improve, and the direction of travel is not in question.

When it does, expect headlines. “The Singularity is here.” Cover stories. Urgent panels at conferences. And it will be true, for that domain, in that application, at that moment. But it will also be the same mistake in reverse. Just as we spent decades dismissing each crossing as not the real Singularity, we will spend the next era overclaiming each new crossing as the final one. The pattern runs in both directions: first we refuse to see what has already arrived, then we announce the arrival of everything when one more thing crosses.

The more useful response, now and when those headlines come, is the same: ask which domain, which application, which frontier. The Singularity is not a moment. It never was. It’s a question you have to keep asking, one domain at a time, rather than waiting for a dramatic announcement that will never come.