The Econ 101 of Avoiding AI RIsk

 Many academic economists are experimenting with AI such as Chat GPT, Gemini, Claude and Grok to do all sorts of things for us.

I have written a free paper on my optimistic thoughts on how AI will Improve Undergraduate Economics Education. You can read it here.

While economists are genuinely optimistic about AI’s role in improving our quality of life, there is at least one prominent economist studying its potential to cause existential risk.  

In this Recent Substack, I discuss his work and I discuss my new paper that responds to his paper.

In late 2024, Stanford’s Chad Jones published an important paper.

Intuitively, Chad explores a key tradeoff. The more we use AI, the more economic Growth is unleashed but at the same time —- the risk of existential loss increases with the scale of the deployment of this technology.

To make analytic progress, Chad makes a series of algebra assumptions that allows him to make progress here in quantifying the tradeoff.

After listening to him speak last summer at Stanford, I’ve been thinking about how my work on the Microeconomics of climate change adaptation relates to his work.

Here is my new working paper.

Some “deep” thoughts

• I have mixed feelings about economists tangling with “existential risk”. All of us will die. There is just the question of when this occurs and what we anticipate will happen to our children and their children. If it is common knowledge that life will end then by backwards induction, we will front load more pleasure and make the best of our remaining time here. Prince had that song; “Party like its 1999”.

• We can adapt to any challenge that we anticipate. We always have substitution possibilities. Yes, if killer robots spawned by AI chase you through the streets —- you have a problem! But, why didn’t you head for the hills and join the other survivoralists?

• All kidding aside, if you don’t trust the Closed AI labs then substitute to open source models.

• If you overly trust the AI Labs such as Anthropic to protect you, does this induce a lulling effect that you don’t take proper precautions to protect yourself? I explore this Peltzman/Viscusi effect.

• If each individual engages in self protection against private AI risk (such as the AI stealing your private data and Ideas), what is the macro aggregate increase in our overall safety? I discuss the Samuelson public goods production function and the aggregation exercise.

• A weakness in the Chad Jones paper is that he doesn’t engage with data and benchmarking of AI adaptation progress. I use Ideas from climate change adaptation to explore the role of real estate prices, insurance prices and other data to test for adaptation progress. I discuss why Barro’s Ideas regarding rare disasters as the equity premium cannot really be used here. This is a negative result. I report ways to use insurance claims and loss ratios to measure whether the private costs of AI are rising or whether firms and households are successfully adapting to facing Lower private costs over time.

• I also explore how market dynamics and the demand for less AI risk encourages individuals and firms to continue to seek out safety Solutions. Milton Friedman has that great video where he discusses where the Pencil comes from. Avoiding AI risk arises from the same messy Hayek process that Jones and other macroeconomists don’t bother to (and can’t) model.

There is a fundamental analogy between the work of Chad Jones and the Nobel Laureate William Nordhaus.

Nordhaus writes out a deterministic and stationary equation that maps how tons of greenhouse gas emissions raise global AVERAGE temperature levels and then how this increase in global temperature maps into a deterministic and stationary aggregate consumption loss.

Jones writes out a deterministic and stationary equation that maps aggregate AI use into societal existential AI risk.

Both model a scale effect of resource consumption and the negative externality associated with the scale of consumption (fossil fuels in the case of Nordhaus) and (AI use in the case of Jones). But, neither of these great scholars model the Endogenous composition and technique effects.

In their mechanical models, there is no intelligent Design. Agents in their models are passive victims who do not have a growing menu of adaptation strategies. I reject this pessimistic view .

Here are some links to my work on the economics of adapting to climate change.