Butterfat, Bots, and Bureaucrats: The Limits of Optimization

Across various fields, we constantly attempt to refine, control, and optimize complex systems. But whether it’s chatbots reshaping beliefs, cows engineered for more nutritious milk, drugs that suppress desire, or regulators chasing vanishingly small risks, each effort reveals the same tension: control involves tradeoffs. The issue isn’t whether optimization works—it’s what it costs and whether we notice it in time.

Much has been said, and rightly so, about generative AI’s tendency to hallucinate—its ability to confidently mislead. However, that same skill to produce tailored, persuasive language may also make it particularly suited to address persistent human issues. A study in Science indicates that AI, when used interactively, can accomplish something traditional interventions have struggled with: truly changing minds.

“Conspiracy theory beliefs are notoriously persistent. Yet previous failures in correcting conspiracy beliefs may be due to counterevidence being insufficiently compelling and tailored. To evaluate this possibility, we leveraged developments in generative artificial intelligence and engaged 2190 conspiracy believers. The intervention reduced conspiracy belief by ~20%. The effect remained 2 months later.”

From Science, Durably reducing conspiracy beliefs through dialogues with AI

We’ve been writing, and I’ve been reading quite a bit about milk over the past few weeks. We often think of food as natural, but much of what we consume is the result of ongoing optimization. Even something as familiar as milk has been quietly engineered into a more consistent and improved product. As The Atlantic reports, modern milk shows how technological advancements can transform even the simplest staples.

“In recent years, American milk has undergone a quiet transformation. The milk produced by our dairy cows has become creamier and more luscious as breakthroughs in cow genetics and nutrition have pushed the fat component of milk—also known as butterfat—to all-time highs. In 2000, the average dairy cow made 670 pounds of fat in her milk a year; today, she’s making 1,025 pounds. No single trait in dairy cows has improved as rapidly with genomic selection as fat.”

Pasteurized milk is both heated and its fat content is reduced to provide a standardized version.

From The Atlantic, American Milk Has Changed

When confronting complex public health issues like obesity, there are competing approaches: redesign the system or directly change human behavior. The rise of GLP-1 drugs highlights a shift toward pharmacologically managing desire itself, rather than trying to improve the food environment. As The New Atlantis notes, this raises a deeper question about whether controlling inputs (MAHA moms) or managing appetite (GLP-1s) is the more realistic solution.

“If Kennedy is correct that stripping the food system of hidden technology – the hundreds of compounds used by food scientists to enhance the color and flavor of everything from bread to burritos – will be a significant boon to public health, then the surprisingly tame policy response he has pursued as Health and Human Services Secretary makes sense. But if his prior, more ambitious critique is correct – if the problem is the more ubiquitous hedonic technology of the modern food system itself – then his fixation on dyes in breakfast cereal is not just inadequate but a big distraction. The only solutions that would make a real difference would be either drugs capable of dampening desire as a whole or radical systemic change.”

From the New Atlantis, MAHA Cedes the Obesity War to Ozempic

My colleague Susan Goldhaber has discussed the overreach of some of our environmental regulations, especially those related to water. Attempts to improve safety can sometimes lead to new costs and tradeoffs. In environmental policy, decades of increasingly strict regulation have aimed for ever-smaller risk reductions, often at significant expense. As Works in Progress states, water policy, like nuclear regulation, shows how focusing solely on minimizing risk can overshadow a more balanced evaluation of benefits.

“The roots of America’s water problems lie in the 1970s. Since the Clean Water Act of 1972 and the Safe Drinking Water Act of 1974, federal environmental rules have ratcheted up. Although state and local water systems had spent the previous century implementing massive improvements in water quality, national regulators treated these systems, supervised by local voters, as delinquent and needing prodding to reform. Water policy parallels nuclear power, similarly, forcing operators to reduce risks, no matter how trivial, and increasing costs to the point where nuclear is expensive and seldom used. This is another example of ‘safetyism’, the tacit ideology that the government’s main goal should be to minimize risks rather than balance them against benefits.”

From Works in Progress, The gold plating of American water