AI will not morph into Terminator

Why AI Won't Take Over

Unlike what popular culture insists, AI is not going to take over mankind and drive us extinct. And it's not because we'll be smart enough to stop it — in narrow domains AI has already outperformed the best humans alive. The reason is much less dramatic, and much more boring: AI is a technology, and every technology in human history has followed the same shape.

That shape is a sigmoid.

The three phases of every technology A sigmoidal curve has three phases.

Phase one — the start.

Growth is slow. The foundational ideas exist, but the surrounding ecosystem doesn't. There aren't enough specialists, the supporting hardware is primitive, and the capital required to push the field forward isn't there yet.

Phase two — the acceleration.

Growth turns exponential, sometimes super-exponential. Specialized materials get manufactured at scale. Entire career fields spin up around the technology. Capital floods in. Progress that used to take a decade now takes a quarter.

Phase three — the ceiling.

Growth slows again. Not because the ideas dry up, but because the resources do. Energy, raw materials, talent, physics — something becomes the binding constraint, and no amount of investment can buy past it.

AI is firmly in phase two right now.

So what happens after phase three?

A new technology gets introduced — one that supersedes the old one — and the curve restarts. Sailing ships gave way to steam. Steam gave way to internal combustion. Vacuum tubes gave way to transistors. Every "world-ending" technology eventually hits its ceiling and gets folded into history while something else takes the lead.

This is the part that pop culture gets wrong. The narrative of AI takeover assumes phase two never ends. It does. It always does.

What this looks like for AI

AI has actually run through this curve before. It's on its second one right now.

Curve one: 1950s through the 1980s.

The pioneers — Alan Turing, John McCarthy, Marvin Minsky — asked whether a machine could think. They built symbolic AI systems, expert systems, the first neural networks on hardware that couldn't support them. Phase one was slow because the compute, the data, and the theory weren't there yet. Phase two had real bursts of progress (chess engines, expert systems in industry), but it crashed into the resource ceiling of the era: not enough compute, not enough data, and a theory of intelligence that turned out to be wrong. The result was the AI winters of the late '80s and early '90s, where funding evaporated and the field nearly died.

That's a complete sigmoid. Start, acceleration, ceiling.

Curve two: 2012 to now.

The new technology that restarted the curve was deep learning on GPUs, kicked off when AlexNet won ImageNet in 2012. Phase one was short — the ecosystem from curve one didn't have to be rebuilt from scratch. Phase two is what we're living in. Transformers in 2017, GPT-3 in 2020, ChatGPT in 2022, frontier models that can reason, code, and generate video by the mid-2020s. The acceleration is real, and it's still going.

But the ceiling is already visible. Training runs are bottlenecked on power. Datacenters are bottlenecked on land and grid capacity. The supply of high-quality training data is finite and we're approaching the end of it. Scaling laws are showing diminishing returns at the frontier. None of this means AI stops being useful — it means the *exponential* part stops, and we hit a new equilibrium.

That's phase three. It's coming.

Why this kills the takeover narrative

The takeover scenario requires AI to keep getting smarter, faster, forever. The sigmoid says no. AI will plateau, the same way every technology before it has plateaued. When it does, two things happen:

1.

Society absorbs it.

The way society absorbed electricity, the internet, and smartphones. The world adjusts, the technology becomes infrastructure, and the existential anxiety fades.
2.

Someone invents the next thing.

And the curve starts over with a different technology — quantum computing, biological computing, something we haven't named yet — and *that* becomes the new "this is going to end us" story.

AI is not the final technology. It's a technology. The fact that it feels final is just what phase two feels like from the inside.

As far as I'm concerned, we should stop cowering from AI and start shoveling resources into it the way railroad engineers shoveled coal during the industrial boom — as fast as humanly possible. The boom doesn't last forever — ride it!

The takeaway

Don't bet against AI being world-changing. It already is. But don't bet on it being world-ending either, because the math of how technologies actually grow doesn't support it. Curves bend. Resources run out. Successors arrive.

However the question should be more of, what exactly is AI's successor? That may of be more concern.