This post is an attempt to categorize AI as an economic actor alongside more recognizable market forces. My goal is to contextualize what is otherwise a highly confusing and dynamic shift, without underrating any of the true novelty of what we're experiencing. Some of my thinking on this has developed more rapidly since the popularization of LLMs, mostly due to the fact that there are far more examples and use cases to consider now. But most of it I've been developing for a few years.
AI as Deflation
Inflation is when the supply of money increases relative to the amount of valuable things ("goods" is a sufficient term) in a market. The most common example of this is when governments print money. But the opposite can happen as well: the goods side of the equation can increase while the amount of money stays the same.
This is deflation, and it happens whenever technology advances. When the cotton gin was invented, it suddenly became cheaper than ever before to buy cotton-based materials. Everyone had the same amount of money, but that amount went further (in the cotton market at least) than ever before. TVs are a modern example of this. It used to cost thousands of dollars to purchase a flat screen TV, but now anyone can have one for a few hundred dollars. Imagine for a moment if everything decreased in cost as fast as TVs have over the last 15 years. Economically speaking, this would present as a few thousand percent currency deflation rather than the 2% inflation we actually saw.
Where does AI fit in? I believe AI could end up being the most rapid deflationary agent in history. In my own industry, software development, I estimate that it's now 30% cheaper to get a given software product created than it was 2 years ago. There are currently models being trained to create realistic CGI for movies, which could reduce the cost of making a movie by 90%. According to the National Private Truck Counsel, driver-related costs make up 45% of trucking costs. Without human drivers, or more likely with a remote human driver monitoring many trucks at once with the help of self-driving AI, the majority of that cost could disappear.
To get an intuitive model of why these falling costs are equivalent to deflation, imagine having $1,000 in a "rainy day" piggy bank sitting on your desk over the course of a couple of years. At the beginning of the period, your thousand could purchase a certain amount of goods and services from the market. 50 boxes could be shipped to you across the country, or you could purchase a certain number of newly built pages on your website. Now imagine that you push off spending your piggy bank for a couple of years, and by the time you get around to it, the AI improvements have set in. What can that same $1,000 do now? You can now have 10x more boxes shipped to you, or add 20 pages to your site instead of 2.
Here's another way to think about it on the macro scale. Let's say you had to sell everything in the world tomorrow in the world's largest garage sale. At the end of the sale, you will hold all of the world's money (naturally, since you hold all of the world's goods). Knowing this, you try to price every good and service in the world. You have to assign a relative value to each thing, and then proportionally spread the global amount of dollars across each good and service.
Now, imagine the same garage sale happening in 10 years following a period of AI development, in which more goods are created at a lower cost. Assuming the same number of dollars, what will have happened to the value of the average item? It must necessarily decrease relative to each dollar, since there are more items to spread all the dollars across.
So AI is deflationary from an economic perspective. The interesting part is what deflation really means for all of us.
Income Earners and Equity Holders
It's interesting to think about who technological deflation benefits and who it harms. Based on our previous example, you might be tempted to say that deflation benefits those who are already wealthy, since their wealth is suddenly more meaningful. And this is true about those who hold their wealth in cash. But there is an interesting side effect to tech deflation that adversely effects the holders of equities.
The first of these problems is something that I'm going to call "margin collapse," because I didn't go to business school and I don't know the real term for it. I would characterize margin collapse as the opposite of "margin compression." Margin compression commonly occurs in an inflationary, rising cost environment; it happens when the cost of delivering your services rises faster than you are able to raise your prices. By contrast, margin collapse, by my definition, occurs in falling cost environments, where the acceptable price for your product falls faster than you are able to reduce the costs of delivering it.
Technological innovation causes broad-based margin collapse for existing industries as competitors arise using new methods which have a lower cost basis. This is a lose-lose for equity holders in those industries. Even if you manage to beat back the competition by adopting new technology and lowering your prices, you must spend capital to modify your product or processes, and that expenditure is simply a loss. Many companies will fail to adopt the new tech at all, and slowly experience margin collapse until they can no longer do business.
This understood, let's go back to the question above: does technological deflation benefit the wealthy? In general, probably not. Some already-wealthy people may be smart enough to invest in companies which use the new process and experience gains in market share (especially if they ride the wave early), but for the most part, existing companies will lose market share and have to spend capital to keep up.
There is also the possibility that cash, gold, or equities in uncorrelated industries may benefit from AI deflation. But in the era of bad money, large holdings of gold or cash are quite rare. And I'm not sure that there is going to be such an industry as one uncorrelated with the development of AI.
In contrast to equity holders, people with smaller amounts of cash savings, cash income, or workers on the forefront of new technologies have the tailwind of deflation at their back. Their salaries will go farther, their savings will be able to buy more, and their early stage stock in the next big thing becomes more valuable as market share is captured.
(As an aside, everything said above can be reversed for inflation. During inflationary times, your salary and savings mean less every year, whereas large public companies are able to use practically free capital to expand indefinitely.)
AI and Centralization
One of the common fears I hear about the onset of AI in all facets of the economy is the complaint of centralization. The assumption is that since only the largest companies seem to be able to build and operate large AI models, the propagation of AI into every space will mean centralizing power for the owner of the models, or the political actor willing to seize them.
There is merit in considering how to prevent such massively centralized outcomes. But I think that the economics will do a better job of preventing this than anyone. Consider that all technological disruption starts out centralized and becomes increasingly decentralized over time. When Gutenberg first created the printing press, he had the only one in the world. If you wanted a book, you had to go to Johannes for it. However, this paradigm flipped on its head as the new discovery moved down the economic chain, and now there is practically an uncountable number of ways that you can send and receive written information (like this blog).
This chain of economic development holds true across all sectors. Here's how I like to formulate it:
Or, as seen from the perspective of the consumer:
I believe that today, AI is somewhere between the invention and application phase of this progression. Because of this, it is still centralized, living partially in the labs of the inventors, and partially in the hands of the few technologists who understand how to apply it.
It is clear to me that the situation will not stay this way. AI tech is going to be decentralized very quickly as it moves down the economic chain, primarily by the industrialist. We will soon see industrial-scale AIs fine-tuned to every type of use case, and finally commercialized solutions at diminishing cost which abstract all of those different options into an AGI-like "intelligence".
AI as a base service is quickly becoming cheaper as compute cost drops. OpenAI doesn't have anything incredibly unique in their underlying methodology, so it's only a matter of time until more competent models can be grabbed off the shelf from any number of vendors. For this reason, I don't think of AI as a centralizing force in the long term. Instead, it seems to be shifting power away from large entities and toward individuals. Large technology organizations are in danger from small teams of people who can use Github Co-Pilot effectively. Hollywood studios are in danger of people who are prompting entire movie scenes into existence. Google is in danger from You.com and Bing. Rapid technological development is anti-entrenching; as we saw above, it erodes the cost structures that allow a single player to go unchallenged for years.
The Only Thing in the World with Value
Every good or service that exists today is over-priced. Don't believe me? Ask any 90 year old whether they wished they had spent more time earning money and buying things, or more time with their kids. At the asymptote of technology, every possible good or service has a value approaching zero when compared with the finitude of human life.
A more humorous example: there's a story going around right now about a woman who is angry because Tesla dropped their prices 25% just after she bought a brand new model. Her anger makes for a funny story, but she is right: the Tesla was overpriced when she bought it. What she and everyone else are missing is that the Tesla is still overpriced.
When we buy a product, what are we actually buying? The communists would say that we're buying labor. We're not, but that's a post for another time. We're actually buying wisdom. If you disagree, I've got a lump of copper, a chunk of aluminum, and a handful of sand to sell you. I call it an iPhone. It's the same exact materials as the thing they'll hand you in the Apple store, but mine has a 20% discount. What, you won't pay $800 for that? iPhones are the physical form of thousands of man-years worth of wisdom about how to turn rocks and minerals into human productivity and entertainment. When we purchase an iPhone, we are purchasing wisdom. When we purchase a hammer from the hardware store instead of a stick and a lump of iron, we are purchasing the wisdom applied to those raw materials.
The fundamental reason that software has been so mind-bogglingly profitable, for both equity holders and workers, is because software allows you to rapidly embed wisdom into the physical world, and wisdom is the only thing in the world with value. AI advances this principle by an order of magnitude. Previously, hundreds of engineers were needed to embed a small piece of functional wisdom into the world. As of a few years ago, you could embed the same wisdom with only 10 engineers.
But even those 10 engineers know that a huge amount of their time is not applying the core wisdom of the business, rather it is spent following the motions of common sense. So much of company building is simply "embedding" what a company already knows. This is why so many ex-entrepreneurs turn into venture capitalists. It rewards them for being wise (their core value to the world), but allows them to skip the years of simply trying to apply the insight that they had in a few moments.
The promise of AI is to radically decrease the amount of time that is spent on simply carrying out the mechanical actions that you already have in your mind. In the same way that software engineers today simply write code and don't think about server administration, engineers of the future will seek and embed wisdom, and not think so much about loops of code. (As an aside, I don't think that this will put good software engineers out of jobs. Good engineers already know that their value comes from deciding what to code, not the actual typing.)
Conclusion
Since I was young I have always felt that ideation was important. With experience, I have been surprised to find that doing is what really counts. I believe the underlying reason for this is that it’s still incredibly hard to apply wisdom.
As technology develops, we march toward more and more direct application of the mind and wisdom to the world. Every year it becomes more a world in which the mind rules. In which the mind and soul gain power.
I am incredibly excited about the years in front of us. Remember Bill Gates' insight: we tend to overrate the 2 year impact of a breakthrough in technology, and underrate the 10 year impact of it. I'm seeking to build a philosophy of technology that makes sense of the latter outcomes. Here's to the next 10 years!
-IT