The Cure for Cancer Is Not a Pure Math Benchmark

Myles Mansfield, “Excavator Back Hoe,” 2026, watercolor. This print can be purchased from the painter’s Etsy storefront.

PERHAPS SILICON VALLEY’S BRIGHTEST have miscalculated.  I suspect those leading and investing in the AI labs have underestimated—vastly underestimated—the chance that progress in machine intelligence can be slowed, damaged, or outright strangled by political forces outside the industry’s control.

That claim sounds less controversial now than it would have seemed a few months ago. Over these months two new political forces have made themselves known. Either, on its own, might imperil the future of the industry; combined, they act as two prongs of a gigantic, tightening vise. When such a device first begins to tighten, the object set between its prongs may still move freely.  Has it eyes to see, it perceives that the vice is closing.  But once the jaws find their grip no amount of squirming or squealing will make any difference. The prongs have hold; the pressure mounts. Either the creature caught between is robust enough to bear the test, or it is mashed to bloody pieces.

The AI safety movement and the revolt against the data centers are the two prongs in question. Though the vise has not—yet!—grasped tight, the squealing and squirming have commenced. Beneath so many tweets, a flavor of panic. Panic does not breed good thinking. I find myself astonished and incredulous at the stories we see so many thoughtless men peddle, and so many desperate men embrace, to explain away the public muster. Many in the industry are fundamentally misreading events. Just as many use these misreadings to goad the industry down foolish paths of action. I do not trust their guidance. Their charts are muddled; their vision has dimmed. They are chancy pilots for any enterprise.

 I have a piece out in the New York Times that attempts to trace a sounder chart.1 This op-ed does not deal in any depth with the AI safety side of the problem—I have a great deal to say about that as well, but plan on publishing my thoughts on the matter here at the Scholar’s Stage on Sunday. My Times piece is focused squarely on the other side of the equation, the public eruption over data centers. In this editorial I draw out the following argument:

1. Two models of the data center backlash dominate Silicon Valley debates. “The first, and the simpler, interprets resistance as a species of mass psychosis. A great mass of gullible, ‘low information’ voters has been whipped into a mob… [for the second] the backlash is a foreseeable consequence of the apocalyptic register of the industry’s own executives.” Neither of these models is correct.
   
   
2. On the second model: there is no evidence that opposition to data centers bears any relation to anti-AI sentiment. Survey data and on-the-ground reporting suggest that opposition to supercompute hinges on local issues. Americans fear data centers will make the place they live more polluted, more costly, or more corrupt.  
   
   
3. Most of these specific concerns are either exaggerated or completely spurious. NDAs are standard practice in development and are not some novel and nefarious tactic invented by the hyperscalers; the average data center affects monthly electricity bills by only a few dollars; the toxic pollution emitted by a data center pales when placed next to factories Americans would welcome in their towns (such as steel mills), and so forth.
   
   
4. Does this mean it is a moral panic after all? Not quite. The precise issues cited do not truly matter—they are placeholders for a broader problem. Cue Jon Stokes: “[each complaint expresses anxiety] around inequality and fear for the common good. The narrative is about the few extracting and making proprietary the goods… that belong to the many.”  Or as I put it: “Americans understand that very far away, other Americans are becoming very rich. They fear that this is all happening at their expense.” This fear precedes any of the actual issues it globs onto. You can refute this or that faulty claim as much as you like, but the underlying fear remains. It will simply find some other argument to congeal around.
   
   
5. What reduces anxiety of this sort? Tangible benefit.
   
   
6. Here the data center buildout contrasts sharply with past infrastructure drives in American history. The railroads and electric lines also made far-off financiers very rich—and both birthed their own protest movements. Why was that blowback vanquished? Because ordinary voters could see the goods. The boons of new rail station or powerline were obvious. The benefits were not hypothetical. “Compare all that to the neighbor of the data center. What advantage does the A.I. industry deliver him?”
   
   
7. This is the crux of the problem. “the costs that the data centers impose can be seen in the present. They are tangible, if exaggerated. For Silicon Valley to win this battle, the benefits of A.I. must be just as concrete. The A.I. labs must show that they, and their products, are making America a better place today — not just in the fairy tale future.”
   
   
8. There are thus two routes for beating the backlash. Both should be taken. The first is to “donate to the schools, or the streets, or even the Trump Accounts, in the counties in which they build.” The second, and more difficult task: “working out which applications of machine intelligence can deliver visible benefit in the shortest possible time.” This question must be at the forefront of the industry. It should be debated on the panels; discussed in the group homes; it should dominate all discussion on Twitter. This question must haunt every executive, scientist, and investor in the industry.

This last idea is a small part of the Times piece, but it is by far the most critical. Allow me to expand on it here.

Among the tech brethren, it is an article of faith that to stand against the data centers is to stand against the cure for cancer. This view would have more force if the frontier labs were actually racing to provide a cure for cancer. They have done no such thing. Instead they build benchmarks in pure mathematics and coding. There are many reasons for this, some quite defendable. These are, after all, “verifiable” domains that do not require special interface with the messy world of cells or solvents for progress.  

But this is not the whole story. Consider a lesson I wish more people had learned from the Navier-Stokes saga. To recap that saga in a sentence: upon becoming aware that a pair of researchers, including one employed by Anthropic, had made significant progress in solving part of the Navier-Stokes Smoothness Problem, OpenAI threw millions of dollars of compute at the problem in hope of scooping their discovery. Their scoop was successful. OpenAI did not burn millions of dollars on this problem because it expected the solution would serve any practical benefit to mankind. It burned millions of dollars on the problem because its IPO was approaching and it wanted its agents to appear more intelligent than the other guy’s. This was a dick measuring contest—nothing more and nothing less.

The machine intelligence industry can afford to waste any more time measuring dicks. The wolves are out. It is not long before they close in.

As I conclude my piece for the Times:

To his credit, Dr. Amodei recently admitted that his company has not yet “delivered on our big promises to benefit the world” and is thus “ramping up its efforts very quickly in biology and medicine.” This suggests one possible path forward. There will be others.

Finding them, and marshaling the resources to reach them, ought to top the industry’s agenda. Either A.I. starts paying dividends that ordinary Americans can collect, or the entire enterprise will be strangled in its crib.

————————————————————-

Your support makes this blog possible. To get updates on new posts published at the Scholar’s Stage, you can join the Scholar’s Stage Substack mailing list, follow my twitter feed, or support my writing through Patreon. If you found this post worth reading, you might find some of my other essays on the history of science and technology worth your while: “Science Proceeds One Question at a Time,” “Has Technological Progress Stalled?,” “For God and Progress” and “Lessons from the 19th Century.”

—————————————————————-