Have you heard that AI will democratize intelligence? Everyone will have a genius - no, fifty geniuses! - snugly in their pockets. Tasks that formerly required an expert or professional will be doable by anyone, and each individual will be free to achieve their unfettered dreams.

Or have you heard that AI will usher in permanent, crushing totalitarianism? Whoever has the most powerful AI when the singularity comes online will have a permanent and decisive strategic advantage against the rest of us chumps, and will rule the world with an iron fist made out of diamandoid bacteria.

Surely you’ve heard that AI will move inequality to 0, or infinity, depending on how you look at it, by eradicating all complex life on the planet. You may have even seen this in a movie or two. Rest assured, the smartest people you’ve ever heard of have crunched the numbers, and it’s real. Or will be. Maybe.

All these scenarios are extrapolation. Will, will, will. I shall leave that to Daniel Kokotajlo and Ryan Greenblatt, who have more patience and skill.1

What about today and tomorrow? Straightforward. AI is extra awesome if you’re rich.

The Cheap Seats

I don’t know how good the free tier of AI services are, since I don’t use any. I have a $20 monthly subscription to Claude, and a $20 monthly subscription to ChatGPT. They’re pretty good. I ask most questions to both, since doing so is cheap, and get a useful answer from at least one about half the time.

When I’m more caffeinated, or excited, or otherwise ambitious, I use AI more, and get more out of it. But there is a ceiling. I can spin up a basic web extension as often as I feel like it, and create or maintain something actually useful once or twice a week, but if I started to go hard, I’d run out of tokens and have to upgrade. The next level of subscription up is $100 a month, and for me, that’s a noticeable line item. I’ll spend it on something I care about; currently, for example, I am spending an eyebrow raising $440 a month for the weight loss cheat code.2 But not on a lark.

Nor do I want to just pay for usage directly, which at my level of plan is the only way to access Claude’s best model, Fable. It’s one thing to pay $20 a month and spent $14 in theoretical tokens without ever having to think about it. It’s another to tool around on the computer for twenty minutes, and then discover that I’ve spent $13.11. If $200 or $1000 a month was nothing to me, I could mess around with AI all day (and get better at it, and have the best models earlier, and get a sense for exactly what was possible) without even thinking about it.

On the other hand, If $40 a month was a strain on my budget, I’d be locked into worse models with strict limits or ads. I’d definitely use AI less, and enjoy it less when I did.

Three Pecuniary Boons

I count three ways more powerful AI helps people with more money. One, they have more freedom even without AI, and so have more AI-friendly tasks. Like, if you have the money to build a custom house, AI can help you realize your vision at the architectural, logistical, and financial levels. If you own a modest home and have a decent budget, you can get AI help with a remodel. If you’re renting a lousy apartment, there’s just not much for AI to do; even if you fix something, you’re spending your own money and time to help the landlord. For a subtler example, if your house is already full of high quality LED smart lights, you can use AI to make a cool program to automatically redshift them as it gets late. But the smart lights cost hundreds of dollars in the first place.

Two, they can experiment without worry. Maybe in 2027 you’ll have to make 200 stupid one-shot AI-generated video games to get a feel of what Claude 7.5 can and can’t do, and then you can actually use your creative taste to make a really good game with its help. But those 200 stupid one-shots will cost $46 each. Hiring a team to make one would have cost you $200,000 a decade ago, and years off your precious life coordinating it all, so $46 is impossibly cheap. All the same, I’m unlikely to spend $9,200 for giggles. And if you can spend $50,000 for giggles, you can mess around indefinitely, without ever thinking about the price tag. But even at lower scales the same logic applies; being willing to spend $500 on a zany project gets you more practice than only spending $50.

Three, they sometimes get the good stuff earlier. Higher tier subscriptions often roll out the best models days or weeks before cheaper plans. Not that much of a head start, but it lets you cash in on the hype and feel like you’re really at the tip of the spear, which is intrinsically motivating. It’s probably easier to try to beat the market to a startup idea when you actually do have a head start, as opposed to when you’re slightly behind.

These things all go together, too, and feed off each other! Someone independently wealthy can spend all day messing around with dozens of simultaneous instances of the latest AI, feeling ahead of the curve and charting new territory, solving real problems in their already cushy life. I’m not shaking my fist at Elysium; I feel most of the way here just by virtue of being a middle class American, so I’m more grateful than grumpy about the returns to money.

Nor do I mean to say that only “the rich” will benefit, and “leave us behind” and it’s “so over” for “the permanent underclass.” That’s pathological extrapolation. Something can be a great option for most people, and an even greater option for a lucky few, and AI is such a thing.

But if you do want to know who benefits the most, it’s no mystery. Tech-savvy people. With money. Preferably lots.