AI and Labor
Recall the 40% quality and related gains for BCG consultants using GPT-4? Let’s look at the issue of AI’s impact on labor through the lenses of two MIT economists—David Autor and Daron Acemoglu.
Thanks to Steven Kahn (Hoge Fenton in Silicon Valley) for sharing this on the AI impacts to labor: NPR did a good interview with David Autor, discussing the results of more studies on thousands of call center workers, writers, and coders. The results were similar and Autor adds interesting commentary, e.g., “It’s been a great four decades for elite workers [… but] AI lowers the cost of elite expertise, makes it more available, and increases the value of the middle skilled workers of the future.”
In contrast to Autor's optimist argument, Daron Acemoglu describes in related book and paper a series of major historical inventions that, contrary to intuition, did nothing or sometimes even worsened the lives of most people. For example, since 1990, each additional robot has reduced employment by approximately six humans while also lowering wages. Acemoglu believes that labor-friendly adjustments have always been necessary to force elites to permit wage growth and shared prosperity from technological gains. In the U.S. context, he urges multiple policies to ensure that AI contributes to shared prosperity.
Here are a few of Acemoglu’s proposals and a few notes:
Tax Software Over Wages
This straightforwardly aims to make AI less appealing as a replacement for human labor on the margins.
Worker Advocacy Organizations
This idea can be viewed as an extension of trade unions, adapted for the AI era.
Repeal of Section 230 of the Communications Decency Act
This would make internet companies, not just AI companies, accountable for the content they disseminate. (Congress seems prepared to deny 230 protection to AI companies while maintaining it for other internet uses. Indeed, this distinction is what AI companies have been asking for.)
Federal Subsidies for Human-Complementary Tech
Promote technologies that enhance human labor rather than replace it. (To consider this seriously with respect to AI, we will need to see more differentiation in the AI product space. Even then, what technologies should qualify for such subsidies and how to measure their impact on labor? Naively, would AutoGPT, Dall-E, and Google's Gemini be too autonomous, while basic GPT-4 and AlphaFold would deserve subsidization?)
Break Up Big Tech
This aims to fostering competition and reducing monopolistic power, as in past eras. (This seems fraught with political, logistical, legal, and tactical complexities. Smaller entities might not have the resources to conduct groundbreaking research in AI. Would the government step in to ensure progress? To what extent would research labs be permitted to share research on AI safety?)