The economy under AI

What the next 10 years might look like

Two years ago, I wrote about how I used GPT to create a simple webgame, and subsequently commented that

First, the infrastructure for hosting & deploying code is awful. Now that LLMs make it extremely easy for non-technical people to write code, there is a huge opportunity in helping people bring their ideas to life with better, simpler tools.

(Quoting myself — classic power play).

It hasn’t taken long for people smarter and more motivated than me to grasp that opportunity. Companies like Replit make it incredibly easy and cheap for people to launch apps just by describing them to AI, with no coding whatsoever. I launched Verso and Cadavre Exquis in an afternoon each.

This drives home that it’s a matter of time before AI transforms the economy. So, like many others, I want to speculate a bit on what the world might look like. TL;DR: I think

• We’ll see an increase in unemployment, which (contra AI optimists) will likely not be offset by increases in demand

• The cost of some services will decrease a lot, but most people won’t feel a big benefit from this, because the things we spend most of our money on won’t be affected so much in the short term

• There’ll be a large increase towards freelancing and self-employment, as big companies will lose most of their competitive advantages (their lower coordination costs and distribution moats)

• The impact on regulation is harder to predict: on one hand, AI can help us reduce or eliminate most rules that target process instead of outcome; on the other, this requires us giving too much power to AI (which we may not want to do) and big corporates may resist such reform, since regulation will remain one of their competitive advantages.

Utopia

Assuming AI doesn’t kill us, it will eventually be so transformative that speculating about it is pointless: the only limits to what AI will be able to produce are those imposed by physics, and since physics permit the creation of universes and stars, we’ll, uh, have plenty of leeway.

What we or the AI choose to do with such power is anyone’s guess. But I want to be clear that notions that have been floated about permanent inequality and whatnot are astonishingly unimaginative.

Near-utopia

There will be a period where AI will lead to incredible advances that nevertheless fall short of Godlike power. I’m talking here about the stuff of science fiction: nanobots that can erect buildings in weeks if not days and so on, things just a step behind alchemic material transmutation.

This state is also hard to think about, but it is also unlikely to be one of class struggle and inequality. The kind of advances I’m describing here are those that result in the end of scarcity, with extremely few exceptions — mainly land, works of art or other unique non-fungible valuable objects, and IP (though note that IP protection will probably be unenforceable, unless governments (the existence of which may be moot by this point anyway) can prevent individuals from creating their own versions of protected material. This is probably not going to happen, the same way that no-one will come after you if you draw a Mickey Mouse comic book in your own home. (Yes, large companies will not be able to use copyrighted material; but will large companies even exist in this world? I don’t think so — see the next section for more on this.))

The value of land will be kind of a problem, in that those who happen to own land before this level of AI arrives get to keep it forever, and no-one will be able to produce anything to trade with them — except other landlords or perhaps artists. But I don’t think this is such a dire issue as it is now. There is a lot of unused land in the world: today, not all land is equally valuable, but in a world where AI can be put to work to improve land in ways that are unimaginable now, even Siberia might become a nice place to build villas.

Some people worry that this power will be closely guarded by its inventors, and so all these benefits will not be enjoyed by everyone. Again, this shows a remarkable lack of imagination: first, why would anyone hoard this power? Many of today’s billionaires are committed to giving away their fortune, and the speed at which they’re doing so is not constrained by greed, but by their ability to do so effectively. And second, even granting that the inventors of powerful AI will want to keep it for themselves, all it’d take is one rogue employee, one benevolent capitalist, one rebellious heir to share these advances with the world.

(This all assumes most people will still want to operate primarily in the physical world. But it’s not crazy to assume a large number of people will prefer to live their lives in extremely realistic virtual reality.)

The near future

Before we reach the level where AI can interact meaningfully and transformationally with the material world, AI will start to automate and replace a number of white collar jobs. This has begun to happen, and it’ll accelerate over the next few years. So the big question is what does this world look like?

I’m too lazy and under-resourced to answer this question, but I can at least provide a framework for doing so. I think AI will reinforce several factors that often pull towards opposing directions.

Automation Vs Increased demand

The first thing we’d want to look at is which sectors will see increases in unemployment, accounting for the fact that advances in technology sometimes lead to increases in demand, which more than offsets the effort automation.

For example, it’s been argued that innovation in retail finance in the 60s led to more demand for financial services, so that even though pessimists feared the advent of technology such as the ATM would cause unemployment, banks found they had to hire more bank tellers (it should be noted though that this wasn’t only because of increased demand, but also because of changes in regulation: competition amongst banks used to be heavily restricted, and banks weren’t allowed to have branches in more than one state; some states even limited bank branches to one per bank.)

On this one I side more with the pessimists: even though some fields will see increases in demand, I think the productivity increase from AI is so huge that it cannot be offset. For instance, AI makes it much easier to create web apps, and yes, this means that non-technical people will start creating games, tools, even operating systems in the near future. But the simplest of those can be done without any help from a human expert; and so, it’s not like this surge in demand will translate to more demand for coding freelancers. Even if it did, those freelancers themselves will be far more productive than they are now. And by the way, 99% of this new software will have so few users that it won’t be commercially viable — so who will pay for freelancers?

Of course, you may disagree with me. To help you get a feel for the impact of AI on employment, I created a simple model using employment figures from the department of Labor. You can use it to estimate the impact of increased demand and automation on employment figures (it’d have been like 10 times faster to do this in Excel — but none of you would click on a link to an Excel model. So I used Replit).

Lower cost Vs lower purchasing power

AI will lower costs in several industries, which in competitive markets will lead to lower prices. This means more things will become more affordable to more people. So far so good — but as per the previous section, one of the costs that will decrease is very possibly that of labour. So, while the cost of goods and services may go down, so will some people’s purchasing power.

It’s hard to say which effect will be stronger. Looking at consumer expenditures data at a high level, I think that the sectors where consumers spend the most will see at best modest cost reductions in the short term — it’s hard to see how AI will reduce housing or transportation costs, which together account for >50% of annual expenditures (there probably will be a reduction, but it’ll be minimal: AI can bring down all the bureaucracy costs, such as legal fees &c. But these are marginal.).

Of course, looking at averages can be misleading when the effects of a change are concentrated on the margins. It’s fair to say that if AI leads to higher unemployment, the people impacted the most will be white collar workers in higher-paid industries, and so we may see disproportionate reduction in demand for more luxury goods and services (e.g. travel).

In short, in the short term, I think most people will be worst off: everyday expenses won’t drop much in price, whereas luxury goods will become out of reach for more people.

Self-employment Vs conglomeration

A question most lay persons never consider is why do companies exist. Entire literature exists to answer this, but the single-sentence answer is that companies exist to minimise transaction and coordination costs. Technology is very good at reducing these costs, which is why the number of freelancers has been increasing, and is projected to keep doing so:

AI will accelerate this, because it will not only further reduce coordination costs (you’ll no longer need to manage a few engineers and a designer to build an app: you’ll outsource the entire thing to an AI agent), but it will erode one of the most powerful moats large companies enjoy: distribution — at least, to begin with, in the digital space.

The tech giants owe their immense profitability to their ability to control distribution. Google and FB are taking advantage the need of goods and services producers to distribute their goods, and Amazon (and in bricks-and-mortar, Walmart) to consumers’ need for convenience to source goods. But, it’s already feasible, and in a few years it will be commonplace, for consumers to use AI buying agents. If we all have a personal shopper/valet/executive assistant, which can crawl the web, analyse millions of products, and find the best quality goods for a given budget (or the cheapest good for a specified quality level), then the power of advertising (and comparison websites, including Booking.com &c) will collapse.

In such a world, a niche artisan will be able to compete almost head to head with large multinationals. They will no longer need to strive for brand awareness and distribution. And so, talented creatives and innovators will have all the more reason to go into business for themselves.

Regulation

Laws and regulations come in two flavours: the one — and ideal — is that they prohibit something because it’s bad in itself. For example, murder is illegal because… it’s obvious why murder is illegal.

But the majority of regulations do not target results but process, because banning a specific result is impossible. You cannot make bankruptcy illegal, so what you do is enforce rules to minimise the odds of a devastating financial meltdown — hence capital ratios, GAAP, &c. These regulations are necessarily blunt tools: they often end up discouraging beneficial behaviours (which is why it’s harder for people who work in certain industries to open bank accounts).

In theory, sufficiently advanced AI could drastically reduce regulatory burden: instead of requiring companies to follow millions of pages of rules, you require them to share all their data with an AI-powered regulation, which scans the data and determines the likelihood of (pre-specified) bad outcomes (e.g. significant consumer harm, spectacular blow-up).

But, there are two things that might prevent this: first, do we want to give such power to AI? You don’t need to believe in skynet to realise that outsourcing so much of the decision-making, or even the analysis, to AI is dangerous: humans will never be able to verify the AI’s calculations, and so this approach takes us down a path where we give up human agency. It may start well, but a few years down the road, we may be in a world where the AI is optimising for outcomes that are incomprehensible to us.

Second (and a less good reason for avoiding regulatory reform), given that as per the previous section AI is already likely to take power away from corporate behemoths, I doubt corporate giants will be eager to do away with regulation. Contrary to popular belief, most regulation (not all) protects incumbents. The ban on tobacco ads means new entrants would struggle to ever win market share; GDPR is easier to navigate when you have armies of lawyers at your disposal — armies that scrappy start-ups cannot afford; the myriads of licenses and permits you need to build new factories in some parts of the world make it harder for a wannabe Ford to get started. And so, it’s likely corporate giants will oppose a radical re-writing of the rules, if such reform makes it easy for challengers to enter the incumbents’ markets.