“Senator, we run ads”
In the mid-1860s, as the Union Pacific Railroad pushed its first miles west out of Omaha, the line's chief promoter, Thomas Durant, had his crews lay track in wide, needless curves across the Nebraska plains. There was no engineering reason for it. The government paid the railroad a subsidy in bonds and land for every mile built: a straight line paid less than a bent one. The railroad's own shareholders bore the cost of the extra rails; Durant and his circle, who controlled the construction company awarded the contract, pocketed the proceeds. Within a few years the scandal that surrounded this kind of self-dealing had implicated a swathe of congressmen and a sitting Vice President, and the Union Pacific Railroad Company itself had limped into receivership.
Last month I tried to map the AI economy. This month I want to ask who, in the end, will benefit and who will control the outcome? History offers us many examples of new technology bringing change. I want to focus on two in particular.
The first is the 19th Century development of the railroads across the continental United States. Here was a genuinely transformative technology that, in aggregate, treated its investors badly. The lines were overbuilt, ruinous rate wars broke out between them, and waves of bankruptcy washed through the sector in 1873, 1893 and after. Yet America was remade. Farmers reached markets they could never have served before; towns sprang up around the depots; the wider economy captured an enormous benefit that never showed up in the railroads' own accounts. This was a social good bought with private losses; the value leaked out to everyone except the people who funded it.
The second is social media, and it appears to run in the opposite direction. The technology made fortunes. Those who founded the platforms, backed them early, or simply held the large technology stocks through their initial public offerings to today have enjoyed huge growth in their wealth. And yet the social ledger looks, to many, far less flattering: shortened attention, coarsened politics, a generation's mental health fretted over in a thousand op-eds. Private gain, arguable social cost; the mirror image of the railroads.
It makes for a satisfying contrast. Railroads lost money but did us good; social media made money but may have done us harm. The obvious exercise is then to understand which of these two historical examples best applies to the AI boom. The trouble is that the contrast is too neat, and the neatness is brought about by telling each story in its simplest form.
What the two episodes share is more instructive than where they differ. In both, vast sums were committed to a technology which wrought huge social, cultural and economic impact. But the private returns proved wildly uneven and impossible to predict in advance. The building of the railroads did make a lot of money for the construction companies that built them even as the outside shareholders lost out, and it was all paid for by generous subsidies from the public purse. The triumph of the tech giants has enriched a broad swathe of investors globally, not just a Silicon Valley cabal. In both instances, the gains spread in ways the early enthusiasts did not foresee. And in both, the rest of us worked out what had really happened only slowly, and largely after the fact.
One of the things up for grabs on the ledger of winners and losers in the AI boom are our jobs. The most important historical parallel, in my view, is how long it took for the world to develop an understanding of what was happening to the internet in the 2000s. The paradigmatic example is US Senator Orrin Hatch's questioning of Mark Zuckerberg in a 2018 hearing. Hatch plainly had only a misty grasp of Facebook's business model and asked how the company made money when people could use the site for free. Zuckerberg's faintly smirking reply, 'Senator, we run ads,' became a meme. The deeper point is less comfortable: the technical details of the AI boom are complex enough that most of us are currently not able to understand their meaning fully. While we develop that understanding, potentially over years, a technical elite inevitably build their own perspectives and incentives into the things they make. The technically complex and commercially sensitive nature of how these systems work retards our ability to engage with them politically. I worry that drastic, undesirable and preventable social changes could happen with almost no scrutiny. They will simply be smuggled in under the guise of technological and economic progress or hidden deep in model weights and data agreements that most of us are not equipped to take a view on. In other words, the historical precedent suggests we tend to grasp what is really happening only slowly and in retrospect, which leaves important political questions in the hands of a few large technology companies for too long.
For me, the more interesting question isn't whether Anthropic's IPO will be a blockbuster or whether Nvidia can maintain its market dominance. It is whether those of us not directly involved in developing and implementing these technologies are capable of learning quickly enough to ensure our interests are best served. I fear that those who are not paying attention will be the ones most disadvantaged. If you want some help keeping up, you can get in touch here.