Decline, productivity and why an AI-fueled world is going nowhere too fast

I saw a diagram in the FT by John Burn-Murdoch that really distills what I’m trying to convey. AI creates more output (which fits many people’s informal perception of productivity), but for many companies (according to studies by MIT, McKinsey, Bain, etc.) it doesn’t deliver much in the way of RoI, and doesn’t change GDP.

A specific map from the FT is about mobile apps (h/t Zen Ju) but you can make similar graphs in many domains: nominal productivity, but not much real-world impact.

For example, you can say something about books. GenAI led to many more books, but it’s unclear whether they were better books or (sorry I don’t have the data immediately) sold more books. Here is the number of books available over time The Washington Post:

Bookselling on the other hand have decreased slightly during the same period. I don’t hear the argument that books are available The best.

The post essay I sketched above had graphs of the increase in uploaded music tracks, auto-represented cases, submitted scientific papers, and web content.

Again there is no reason to think it added any GDP or quality to science or music. Most of it is slanted (and in some cases Job decline):

The FT and The Washington Post’s graphs show the same thing over and over again; We’re inundated with AI-generated content, from apps to music to scientific papers, but it doesn’t do anything better. Most (and to be fair, not all) of it isn’t.

And then there you have it The collapse that sinks Wikipedia, Libraries And so on.

The same is likely to happen in mathematics. A group of mathematicians, apprehensive about the many aspects of AI’s impact on mathematical research, have written an open letter.Leiden Declaration”, Reported by NYT(among other concerns) expresses their fears

Current automated techniques can produce plausible but implausible (or false) arguments that are difficult to distinguish from valid mathematical proofs. This applies not only to informal arguments but also to formalization, where there is difficulty in translating between computer-encoded and human presentations of concepts. These fast-moving developments are putting our current peer review system under increasing pressure, affecting our ability to enforce traditional standards for correctness, transparency, and independent verification of evidence.

In other words, mathematical collapse.

A long time ago, when I was thinking about official productivity measures and whether they would change with AI, an investor friend wrote to remind me how narrow the technical definition of productivity is relative to GDP, “In the technical world of national accounting, paying people to dig and fill holes increases GDP.”

Using GenAI increases tokens costs and fills the world with content. But often that content — generated by a giant but unreliable word prediction engine that’s more revisionist than innovative — has little lasting value.

May be The biggest exception is coding, but even there, it’s not yet clear how many systems built with the help of agent coding will endure.

§

And, especially in agent code, there are Massive Losses, up and down the food chain. Providers (Open AI, Anthropic, Cursor, etc) are losing tons of money, scrambling to raise prices, and customers grumbling at new usage models. (In a new study, Gerben Wierda argues that Anthropic and OpenAI.For every $100 you spend you can earn $1000“. Data center intermediaries like CoreWeave are also losing money. It’s hard to tell how chip companies are doing, and when you factor in all the round-robin funding and shrinking cash flow, even a potential major productivity story (code) is burning more cash.

Once providers (like Open AI, Anthropic, Cursor etc.) try to collect enough money to cover their losses, AI costs more than humans.

Talk about digging a hole without creating real value.

Share it

B.S. See more Paul Krugman And Paul Kedroski From different angles, to new pieces that achieve somewhat similar results.

The PBS apropos tweet may have been somewhat tongue in cheek from yesterday: