The Last Moat is Capital

The AI buildout might be the last era in which humans have a moat.

There’s a techno-optimistic view of AI: that humans will be able to reinvent themselves, switching from performing intellectual task themselves to orchestrating AIs that do.

However this view relies on the assumption that AI doesn’t get smarter than us. What if it does? AI arguably is, already, smarter than most humans. More over, if its current pace holds, the gap will soon widen too much for us to communicate with it. It’s studied that humans with a >20 IQ point difference struggle to communicate with each other, as if they lived in different worlds or were different species.

Do we reasonably think that we will be able to orchestrate and oversee beings that are significantly smarter than us? Do chimpanzees oversee our work as humans? Do dogs? Chimpanzees might be jumpy and dogs fluffy, but that isn’t a moat that allows them to control beings smarter than them.

Thus the only other viable route for humans to keep at the helm is mixing with AI: implanting chips in our brains that augment us in novel ways. There are many risks with this approach, the most important being that we would have a second brain that’s potentially smarter than our biological one, therefore taking us back to the control question. But even if it works, we would cease to be humans, and thus we become something new.

There’s of course the physical world, and we humans do have a defensible moat there. Robotics is emergent but the world of atoms is so far harder to navigate for AI than the world of bits. But that’s a matter of time. Will it take one year? Two? Five? Ten?

We humans are really bad at understanding exponentials. We see robots that might clean a house today and think they are cute. But if the advancement of physical AI remotely follows the advancement of digital AI, our moat over atoms will be gone too.

What moat remains then? Until AI can fully manufacture its own data centers, its own robots and produce its own energy, capital is the only defensible moat.

Short term, AI companies take capital and transform it into labor cuts.

An analyst making $100,000/year can now be automated for $1,000/year, a housekeeper making $40,000/year can be replaced by a robot to be rented for $4,000/year. Notice that paradoxically the most skilled work (intellectual) is the one that’s the most optimized by AI.

There are two sides of this coin: a) the regular worker who doesn’t have tons of assets but has a job, and b) the capitalist who has assets but whose net worth is being constantly debased (thanks to different kinds of entitlements to seniors, immigrants, unproductive people, etc.)

Your regular worker will lose their job. Your regular capitalist will thus have its capital debased via money printing and higher taxes, which are necessary to subsidize the workers, for them not to revolt.

Thus paradoxically the only rational thing the capitalist can do is further invest in the AI buildout to protect their capital.

Any kind of national crackdown on AI will lead to an unfavorable outcome: a slowdown in AI progress means an adversarial country (e.g. China to the US) can win the AI race. If a country has an AI that’s even 30-40 IQ points higher, they might be able to conquer another country (not only in the traditional sense, but via subconscious manipulation of its population, hacks, etc.)

This situation reminds me of Meditations on Moloch, where it’s in the collective best interest to do one thing, but individual interests make us all sacrifice ourselves.

I’m not sure there’s an alternative (more than praying). But while the AI buildout is in full effect, from now up until the point in which AI can build its own datacenters and produce its own energy, capital is the last moat.