Thread by @DAcemogluMIT

AI myth 3. Abundance myth. Building on AI myths 1 and 2, a third, and perhaps more pernicious one emerges: automation and human-like performance by AI will bring economic abundance, out of which all or most of society will benefit.

There are many versions of this, going back to IJ Good’s statement that “the first ultraintelligent machine is the last invention that man need ever make provided that the machine is docile enough to tell us how to keep it under control.”

Or futurist and Google technologist Ray Kurzweil’s periodic pronouncements that “singularity” is just around the corner, e.g., amazon.com/Singularity-Ne…. Or less fanciful versions where AI boosts productivity, so that most people can get by without work.

There are at least three problems with this abundance myth. First, once you abandon myths 1 and 2, it loses its appeal. If machines will not be ultraintelligent, it is less clear that they can really bring boundless abundance.

Second and more fundamentally, the automation logic underlying this view is not as compelling as it first appeared. This is in fact a more general point (building on my work with Pascual Restrepo, economics.mit.edu/sites/default/…). Let me illustrate it with a hypothetical example.

Imagine that we move towards this type of abundance based on automation. Suppose a very fast-paced automation in which 4% of all human tasks are automated every year (so that all human tasks are eliminated in 25 years). And suppose technology does nothing else than automation.

How much growth will this generate? Suppose AI is 50% more productive than humans (a whopping amount!). Then the aggregate productivity growth will be less than: 0.04 times 0.50 = 0.02. This is slower than US economic growth before the digital revolution; amazon.com/Rise-Fall-Amer…

Just automation will not get a huge productivity boost to the economy. If you want to increase productivity, it’s much easier and more effective to make workers more productive — hence we are back to machine usefulness and human-complementarity again.

Third, all of this discussion ignores how dystopian world without work would be. There are those who think this could be a happy world, e.g., “fully automated luxury communism” amazon.com/Fully-Automate…

Yet much work shows that people want to feel they are making a contribution to society. See, e.g., Reshmaan Hussam and co-authors: hbs.edu/ris/Publicatio…, showing people’s willingness to sacrifice one third of their earnings in order to do work rather than receive transfers.

The reality is probably even worse, because social networks and democratic participation would be badly affected by the world in which people become just consumers, not contributors. (I tried to write on this before, e.g., here: project-syndicate.org/commentary/soc…)

There are deep issues here. Michael Sandel, in his wonderful book, Tyranny of Merit, amazon.com/Tyranny-Merit-… suggests a notion of social justice based on contributions rather than just what people consume. The best of my knowledge, these ideas have not been fully developed yet.

I also recently wrote on this as part of a collection of essays, here unlocked.microsoft.com/ai-anthology/, that have just been released. My essay is here for those who are interested: unlocked.microsoft.com/ai-anthology/d…

Bottom line: the idea that AI will create abundance is on shaky grounds, and that we could build a well-functioning society on the back of AI-based production and little work for humans is even on shakier grounds. For more on this, see #PowerAndProgress, amazon.com/Power-Progress…