Garrek Stemo

The Vision for AI Is Disappointing So Far

Friday, August 01, 2025  ⚓︎

The automobile transformed the American urban and rural landscape from 1910 to 1930. Gone was the manure from city streets and the sight, stench, and sounds of horses. Cities expanded as people moved out from urban centers. Distant recreational activities were suddenly possible, increasing the value of leisure time. Rural isolation ended, along with single-room schoolhouses in favor of consolidated schools. Farmers were freed from the tyranny of their local merchant and could drive to towns. Today’s tech CEOs believe that AI will grow powerful enough to surpass many or all previous human inventions, but the short term promises are considerably underwhelming.

A recent interview on The Verge’s Decoder podcast illustrates the point. AI startup, Caption, uses generative AI to produce short video clips. Caption’s CEO, Gaurav Misra, says the biggest use case for AI video is marketing. Small businesses and individuals can do without hiring videographers or learning to shoot and edit video. Misra promises to democratize advertising. Perhaps a boon for some, though hardly revolutionary.

Visions from other elsewhere are similarly narrow. Agentic AI promises to reduce the burden of completing online tasks like booking flights and compiling slides. Google’s AI Applications page mentions things like “data analysis” and “increased efficiency”. On Internet forums you will find those waxing about AI being a game changer for increasing blog post output.

Some, like Anthropic CEO Dario Amodei, predict AI could increase GDP in developed countries to a sustained 5-10% per year. This would mean a doubling roughly every ten years, a rate of economic change humans have never seen before. But are we going to get there by writing more mediocre blog posts? Are we going to see massive GDP growth because AI manage our online branding by outputting a deluge of content that maximizes engagement and purchasing? Will it come from small efficiency gains in existing business operations, like summarizing meetings?

The answer is almost certainly “no”. Real change will come from real innovation — completely new products that will change how we live, like the automobile. We have not yet imagined how AI will plug in to society to effect such change. The biggest bottleneck may not be technical, but human ingenuity for finding applications.

Generative AI may yet change the world. But for now, it mostly wants to sell you things efficiently.

Robert Gordon, 2016. The Rise and Fall of American Growth. Princeton University Press, Chapter 5.

We are not ready for better deepfakes

Artificial intelligentsia: an interview with the boss of Anthropic