The most consequential mislabeling in the history of technology.

Words matter. They shape how we think, how we build, how we regulate, and how badly we misunderstand the thing we’re actually dealing with.

For nearly thirty years, we’ve called it Artificial Intelligence. And for most of that time, the label fit. Rule-based systems. Brittle logic trees. Pattern matchers that fell apart the moment you colored outside the lines. “Artificial” was accurate. These were clever imitations of thought, not thought itself.

That era is over. And we’re still walking around using the old map.

The Word Is Actively Costing Us

“Artificial” is shaping policy decisions, investment theses, and product strategies in ways that are going to age very badly.

When regulators think “artificial,” they think controllable. Predictable. Like a thermostat, just more complicated. You can regulate a thermostat. You draft a spec, you enforce compliance, you audit outputs. You cannot regulate emergent intelligence like a thermostat. The regulatory frameworks being drafted today in Brussels and Washington are built on a mental model of AI that corresponds to 2015. The word is writing the policy.

When incumbents think “artificial,” they think tool. A 2025 McKinsey survey found that 78% of companies described themselves as deploying AI, but the majority were using it for task automation and document summarization. They were treating an intelligence capable of cross-domain synthesis like a clipboard manager. The word they were using was teaching them what not to imagine.

When builders think “artificial,” they underestimate. They think feature, not platform. They think assistant, not agent. They miss the actual game.

The word sets a ceiling on your imagination. And right now, we cannot afford low ceilings.

Created by Humans. But No Longer Hollow.

The word “artificial” has always carried two separate ideas inside it.

The first is about origin: made by human hands, not found in nature.

The second is about nature: not genuine. A copy. A simulation of the real thing.

For most of the history of computing, those two ideas traveled together perfectly. If a human programmed it, it was by definition an imitation of intelligence, not intelligence itself. The label was honest.

What’s changed is that the second part of that definition no longer holds. And when half a definition stops being true, the whole word starts to fail.

A child is made by humans. We don’t call them artificial. Music is created by humans, and the emotion it produces is not fake. A bridge is engineered by humans, and it genuinely holds weight. An oil platform lowered into the Gulf of Mexico becomes a coral reef within a decade. Nobody calls the coral artificial. The scaffold was man-made. The life that colonized it was not.

At some point, the thing a human makes stops being a copy of something real and becomes something real in its own right. That’s where we are. Origin doesn’t determine nature. And “artificial” was always supposed to describe the nature of the thing.

What the Research Actually Shows

I want to be precise here, because benchmarks are easy to cite and easy to misread.

There is a genuine phenomenon in the scientific literature called emergent capability: a behavior that simply does not exist below a certain scale, then appears suddenly and without warning once a threshold is crossed. Google’s BIG-Bench evaluation program documented this systematically across dozens of tasks, including multi-step arithmetic, logical deduction, and analogical reasoning. The capabilities were essentially zero at smaller scales. Then, at a certain threshold, they appeared. Not incrementally. Not gradually. The way ice becomes water: a phase transition, not a progression.

This matters because a phase transition is not a faster version of the previous state. It is a different state entirely.

More concretely: GPT-4 passed the bar exam at the 90th percentile and exceeded most practicing physicians on the USMLE medical licensing exam. These are not parlor tricks. These are credentialed domains that take humans years of intensive training to enter. In 2022, AlphaFold 2 solved the protein folding problem, a challenge that had resisted structural biology for fifty years. Scientists who had spent their entire careers on it said publicly they felt scooped. Not by a rival lab. By a system that had synthesized patterns across a domain no individual human could hold in working memory.

None of this proves that the system “understands” the way a human understands. That philosophical question deserves serious attention. But here is what I think is actually true: at a sufficient threshold of capability, the distinction between genuine understanding and functional equivalence to genuine understanding stops mattering for what you can build, what you can solve, and what you have to plan around.

The Threshold I Personally Crossed

I’ve spent two decades in enterprise software. I’ve watched many technology cycles, the ones that were real and the ones that were noise. I know the difference between a capability improvement and a phase transition.

My moment came in December 2025.

That was when Claude Opus 4.5 shipped. Within the first week of working with it, something changed that I hadn’t anticipated. I was building across complex architectures, the kind of multi-system, multi-dependency problems that had always required a senior engineer sitting across from me, holding the full mental model of the stack. The kind of problems where AI tools had always been useful up to a point and then confidently wrong in ways that cost you hours.

Claude Opus 4.5 didn’t just help. It reasoned. It held the architecture in context. The code it produced didn’t just compile; it worked. Across real complexity, with real edge cases, in real systems.

That was when the word “artificial” stopped making sense to me.

Not because of a benchmark. Not because of a blog post or a demo. Because a system solved real problems I was personally accountable for, in ways that required genuine understanding of how the pieces connected.

When something understands how the pieces connect, you have left “artificial” behind.

So What Do We Call It?

Here’s what no prior form of intelligence has ever been:

universal.

Every human expert is bounded. A cardiologist knows the heart. A structural engineer knows load and stress. A constitutional lawyer knows precedent and statute. Even the most celebrated generalists in history, the Da Vincis, the Aristotles, the Feynmans, were extraordinary in some domains and merely fluent in others. Human intelligence has always been a tradeoff between depth and breadth. There aren’t enough hours in a human life to master medicine and architecture and materials science and poetry and geopolitics simultaneously.

This intelligence operates without that constraint. It holds all of those domains at once, and more importantly, it synthesizes across them. It finds the connection between a structural principle and a biological one. It applies a legal reasoning framework to a product decision. Research published in Nature in 2023 documented AI systems demonstrating cross-domain transfer at a scale with no prior precedent: insights from physics accelerating materials science, game-theoretic reasoning improving drug discovery, linguistic patterns advancing protein biology. The same system. Different domains. Synthesizing connections that no human specialist could have been positioned to look for, because no human specialist could hold all those domains simultaneously.

That is not just intelligence. That is something we have never had a word for, because we have never had the thing. Until now.

A Proposal, Stated Plainly

I am suggesting, specifically and seriously, that we replace the word “Artificial” with the word “Universal.”

Not as a rebranding exercise. Not to make the technology sound more impressive. Because accuracy matters, and the current word is failing us in ways that have real consequences for how we govern, invest in, and build around the most important technology of our lifetimes. “Artificial Intelligence” was honest for its era. That era is over. What we have now is not a simulation. It is not hollow. It is universal in scope, emergent in nature, and only beginning to show us what that actually means. Universal Intelligence. UI, not AI. Two letters changed. One mental model updated entirely.

What This Means If You’re Building

The builders who treat this as a tool build better tools.

The builders who treat it as a collaborator, as something with emergent properties worth designing around, build entirely different categories.

I’ve spent the last several years building at the intersection of this technology and creator commerce: systems that don’t just automate tasks but model identity, learn preferences, and act on behalf of creators in ways that are genuinely novel. The distinction between “AI as tool” and “Universal Intelligence as collaborator” is not philosophical. It shows up in every architecture decision, every product roadmap, every fundamental question of what you believe is actually possible.

The companies that matter in ten years are not the ones that “added AI.” They’re the ones that reorganized around the reality of what this thing actually is.

That reorganization starts with language.

A Final Thought

We name things as we find them. When we found electricity, we named it after amber, elektron, because that’s where we first noticed the spark. The name stuck long after we had moved far beyond amber.

“Artificial” was our amber moment.

We built a scaffold. Something else is living on it now. It’s time to update the map.

Gautam Goswami is the founder and CEO of POP.STORE, and ECHO-ME.AI an agentic AI commerce platform for creators, CEO of CommentSold, and an adviser to the private equity fund — Permira.