Like rocks, like water

24 May 2024
24 May 2024
SoHo, New York
4 mins

For a long time, global supply of energy was limited by the total number of humans in the world. We could only put more energy to work by creating more humans, or by each human working more.

And then we mechanized it. As energy supply grew exponentially, we came up with lots of new things we could do now that we could spend a thousand or a million times more energy on something valuable. Things that didn’t make sense to even try before when every joule of energy came from human labor, like invent trains and electric lights and build cities and industrialize farming. Before mechanized energy supply, we could have asked, “What would we ever do with a billion times more energy besides make a billion of the things we already make?” But it turns out there are a lot of things we could do with a billion times more energy. Much of the energy consumed today goes to power human activities that did not exist in current form before the mechanization of energy supply, like trade, transit, and computing.

If we are approaching the slow but certain mechanization of intellectual labor, it’s natural to ask, “What would we ever do with a billion times the intelligence?”

I think the vast majority of intelligence supply in the future will be consumed by use cases we can’t foresee yet. It won’t be doing a billion times the same intellectual work we do today, or doing it a billion times faster, but something structurally different.

Rocks and water

Manual supply, whether of energy or intelligence, is like having a bunch of rocks. You find small rocks, big rocks, sharp rocks, and round rocks, each for their own purpose. You can amass a huge rock collection to do different things, but each rock is kind of its own thing. You can’t just say “I have n kg of rocks. That enables XYZ.” You can’t combine small rocks to make a big rock, or turn a big sharp rock into a giant wheel of the same size. There are things you can do with huge rocks that you will never be able to do with a million pebbles.

Scaled, mechanized supply is like water.

Right now, people totally misunderstand what AI is. They see it as a tiger. A tiger is dangerous. It might eat me. It’s an adversary. And there’s danger in water, too — you can drown in it — but the danger of a flowing river of water is very different to the danger of a tiger. Water is dangerous, yes, but you can also swim in it, you can make boats, you can dam it and make electricity. Water is dangerous, but it’s also a driver of civilization, and we are better off as humans who know how to live with and work with water. It’s an opportunity. It has no will, it has no spite, and yes, you can drown in it, but that doesn’t mean we should ban water. And when you discover a new source of water, it’s a really good thing.

I think we, collectively as a species, have discovered a new source of water, and what Midjourney is trying to figure out is, okay, how do we use this for people? How do we teach people to swim? How do we make boats? How do we dam it up? How do we go from people who are scared of drowning to kids in the future who are surfing the wave? We’re making surfboards rather than making water. And I think there’s something profound about that.

David Holz, Midjourney

Water is water. Once you figure out how to rein in water to do useful work, you simply construct a way to channel the flow of water, and then go out to any river or ocean and find some water. All water is the same, and having twice the water gives you twice as much of whatever you want to use the water for. A billion times the water, a billion times the output. Water can flow constantly, continuously, forever, as long as the river flows and the tides come and go. Every drop of water costs the same, and every drop of water is like every other drop of water. There is only less, or more, and how you put it to work.

Water, given enough time, can chip and dissolve any rock into powder. But rocks held together can guide where water flows, and by doing so, carve rivers and canyons and even move coastlines.


Mapping interpretable features in a text embedding space

In the beginning… was the command line

I share new posts on my newsletter. If you liked this one, you should consider joining the list.

Have a comment or response? You can email me.