A One Percent Solution Towards Uncertainty

Sal Khan had an op-ed in the NY Times the other day, positing that the solution to the impending jobpocolypse caused by AI should be for companies to dedicate 1% of their profits to job retraining programs.

I'm skeptical. Not just because Khan is invested in that outcome (he runs a non-profit organization that would conduct such retraining), but because many assumptions are made without strong evidence¹.

The first assumption is that AI will take away many jobs. I think this is a terrifying outcome. The creative destruction of technology has typically produced new job categories. The worry is that this will not happen with AI. Also, the speed at which this shift might occur is frightening, as society will not have time to adjust (after all, we had a good few centuries to adjust from agrarian to industrial... and it's not like that change was a walk in the park).

Of course, this assumption is also based on AI companies' promised return on valuation (a valuation that is justified by replacing the expenses of workers). Hype makes might (in the markets) and right (in the eyes of investors).

I personally think we will have an insane amount of societal unrest if, as Dario Amodei believes, 50% of all white-collar jobs disappear in 5 years. That makes the future far more opaque.

Regardless of whether AI will displace many jobs (a subject of future writing), the second assumption is that job retraining works. Here, the evidence is rather nuanced (and sometimes not clear). For example, the US's Trade Adjustment Assistant Program (which supports job retraining) isn't very effective at job replacement - particularly at former wage levels. Meanwhile, community college programs appear to a moderate to mild affect on wage earnings and job placements. The strongest positive correlations lie with programs that feature internships and apprenticeships. Also, job retraining really doesn't work if there are no jobs in the community in the first place (a consistent historical reason for migration).

Where might Khan Academy fit into the efficacy of job retraining programs?

I'm unaware of any research specifically on Khan for this use case, but there have been studies on Massive Open Online Courses (MOOCs were the rage some 10 to 15 years ago). Those studies (here's a sample) don't paint a rosy picture (they may feature modest impacts).

In short, is a 1% dedication of profits on retraining programs (that probably don't work) really a good way to address the (currently hypothetical and predicted) jobpocolypse? Probably not.

But look, maybe I'm wrong

I'm armchair opining here and would really like more evidence for any of the above assumptions. It's not that retraining workers for growing industries isn't a worthy goal! You can't get better at policy ideas without trying them.

That said, this feels like a bit of a sop to big AI firms. "Just give 1% of your profits to help workers, and you too can wash your conscience clean of the high unemployment you caused." We could just tax firms or models at a rate that expands the social safety net for those workers. Given the massive economic upheaval AI is likely to bring, I don't think citizens will be happy with such a small olive branch.

I sometimes think that while job retraining is a worthy goal (when it works), the higher priority might be in teaching citizenship and history. Enlightenment for the workers to help shape the society they want.

Footnote (1)

In truth, there are many assumptions at play with this framing, such as the belief that jobs should be tied to money (something I increasingly question) and that work is necessary and good (a core belief). This post does not dive into those concepts 😀.