10 Comments
Jun 28·edited Jun 28

This isn't contra Acemoglu at all, because you haven't claimed anything specific. Make your own estimates and provide warrants for them, and then we can see who is right in 2034.

We can look at things from the demand side. American household budgets are taken up with housing, transport, healthcare, living costs (utilities, childcare, food, clothing, household goods), education, government services (via taxes), and entertainment and leisure.

In which of these sectors can AI make a big difference, freeing up budget for new things, inside the next ten years? Possibly entertainment and leisure.

Acemoglu's estimate looks more plausible than any claims of earth-shaking change.

Expand full comment

Your claim about the sectors holds only if you assume that the progress stops right now. Which goes against every single empirical observation since the Scaling Laws paper came out in 2020. What warrants such a claim?

Expand full comment
Jul 2·edited Jul 2

I don't think it does depend on that.

We can extrapolate by observing wealthy households now. Since all of the things I listed are essentially free for wealthy households, what do they spend their money on?

The answer is status displays. Collecting things whose value consist in scarceness: high-end property, high-end artworks, other scarce things. Elaborate rituals that involve the labour of thousands of people, and ideally a great deal of scarcity like getting Taylor Swift to perform at your daughter's birthday party. "Experiences" like health retreats and sitting at the feet of gurus. Weird "medical" treatments.

AI can help with none of that, because it's all intrinsically zero-sum, and it's intrinsically about commanding the labor of other humans.

Providing more of those things via AI just makes their value collapse.

What else could the median household spend freed-up money on beside aspiring to the lifestyles of the rich and famous?

Expand full comment

The optimistic case for AI is that everyone gets a therapist, coach, teacher, mentor, and assistant (basically a mom); the likely case is everyone gets an AI boyfriend or girlfriend and the population goes to zero in a generation.

Expand full comment
Jun 27·edited Jun 27

Great post.

Doesn't Acemoglu have a point that generative AI doesn't deepen existing automation? Over the long term, AI will become less dependent on data and more robust. But in the near term, the only applications of deep learning that work are those that have an abundance of data, and don't need to be reliable.

Expand full comment

I don't see how one can do any macro assessment about an unknown future

Expand full comment

Isn't that like... the whole point of marco?

Expand full comment

Yes it is. But given the wide variety of predictions coming from economists suggests that it hasn't been very successful.

Expand full comment

That is a very weak claim in the circumstances, I am confident it will not hold up in 5 years, and I would not be surprised if it breaks in under 3.

Expand full comment

Is there a simple macroeconomic impact of work from home?

Expand full comment