The Industrial Revolution analogy doesn't work for AI

The Hugging Face incident, recent warnings by prominent figures in the tech industry (such as Anthropic’s Dario Amodei), and the prospect of massive job obsolescence have many of us understandably worried about the potential costs of AI. But AI boosters insist that these costs are a small price to pay for the immense benefits the technology will bring. They typically use the industrial revolution as an analogy: yes, industrialization caused dislocations and problems, but ultimately lifted millions out of poverty and led to a doubling of human life expectancy. Why should we let our current fears of technology deprive future generations of similar blessings? The problem is that even if AI delivers on its promises of accelerated economic growth and medical breakthroughs, these benefits would have only marginal effects on American well-being. Abundant research (including the Harvard Study of Adult Development) has shown that once the people in a given society have enough resources to meet their basic needs, additional wealth has almost no measurable effect on their happiness (this is often called “The Easterlin Paradox”). Even if AI doubles or triples our wealth in the next few years, our lives will be no better in the ways that matters most. This suggests that the industrial revolution doesn’t work as a historical analogy to the AI revolution because it was a one-off. You can only go from absolute poverty to affluence once and there’s no amount of economic growth that can replicate that initial boost in well-being. Doubling the amount of food available to 1830’s Americans made a big quality-of-life difference, but it would likely have no effect on quality of life for Americans in 2030. The economic principle of “declining marginal utility” is real and, for most Americans, the marginal utility of another dollar is zero. AI-fueled economic growth in the 21st century wouldn’t have anything like the positive impact of industrial growth in the 19th century. Sadly, the same is likely true of medicine. In the past 200 years, the life expectancy of the average American has roughly doubled, and yet that was a one-time benefit achieved primarily by all-but-solving the problem child mortality. Now that over 98% of children live to adulthood, any further AI-driven medical advances could only have a small effect on life expectancy, and almost all of that extension would happen when the quality of life is lowest (end of life) rather than highest (beginning of life). Mother nature has set a hard limit to how long we live and, unless AI can conquer mortality itself, no amount of medical advance will lead to anything like the improvement in life expectancy seen over the past two centuries. In a recent viral TED Talk, Washington Post columnist Megan McArdle argued that AI would probably put her (and presumably the rest of us) out of work, but that this is a small price to pay for the overall economic and medical benefits that AI will generate. What if, she says, the Luddites had succeeded in stopping the Industrial Revolution? But putting a pause on AI wouldn’t be like smashing the machines at the dawn of the industrial revolution and today’s Luddites are on much sturdier ground than their industrial age predecessors. There is always the possibility that AI will bring unanticipated benefits, but those we do anticipate—faster economic growth and medical advance—really won’t do much in terms of making life better for Americans.