Cash transfers are built on a basic premise: instead of deciding in advance what people in need should receive, you give them cash and let them decide how to use it. Part of its advantage is simplicity. But even though cash is more straightforward than deciding whether to distribute food, fertilizer, school supplies, or vouchers, it doesn’t eliminate a core problem: figuring out who should receive it, and getting it to them in time.

In normal circumstances, governments and aid groups rely on tools like proxy-means tests, social registries, or community nomination to identify low-income households. Household surveys are accurate, but slow and expensive to run. Social registries are cheaper, but often out of date. This trade-off between speed and accuracy is one of the perennial problems in grant-making — you can get money out the door much faster if you are willing to sacrifice a bit of certainty. GiveDirectly is dealing with this using a new, AI-based method that’s quite interesting.

Imagine an increasingly common scenario: a flood hits, or a hurricane makes landfall, or a pandemic shuts down local economies. You want to get cash to the people who need it most — and quickly, before savings run out, food prices spike, or families start selling assets. But in many of the places most exposed to these shocks, there’s no reliable, up-to-date list of who is most in need.

Subscribe now

To get around this bottleneck, GiveDirectly has started experimenting with a different approach in some crisis settings: using anonymized mobile phone data and machine learning to identify people likely to be in need. The idea is not that phones “measure poverty” directly, but that patterns of phone use correlate with economic status. Intuitively, call frequency and duration, SMS use, mobile money and airtime recharges, the diversity of contacts, and how much a phone moves between cell towers are all related to income and vulnerability. People with fewer social connections, less frequent recharges, and more restricted movement are, on average, more likely to be poor.

They call this approach MobileAid. The validation is relatively straightforward: train machine-learning models on anonymized phone data and then validate them against the in-person household surveys that are considered gold standard. That allows for benchmarking against what we know works. In practice, phone data is used to generate a shortlist of households likely to be in need, after which people can self-enroll via text message or a call center, be remotely verified, and receive mobile money payments. This means cash can reach those who need it sometimes within days or even hours.

So how well does this actually work, relative to the tools governments already use?

The first real-world test came in Togo during COVID-19, when lockdowns made door-to-door surveys infeasible. Working with the Togolese government, researchers used phone metadata and machine learning to help identify people eligible for emergency cash transfers when lockdowns made in-person surveys impractical.

The program reached over 138,000 people, and when researchers later compared different targeting approaches, the machine-learning method reduced exclusion errors by 8-14% relative to geographic targeting. Extrapolating a bit from the sample, this means roughly 4,000-8,000 additional people who should have received aid actually did.

The authors emphasized that phone-based targeting worked best as a rapid, supplemental tool in crisis settings, rather than a wholesale replacement for traditional surveys — this seems right, and appropriately cautious, given the accuracy-coverage tradeoffs.

Subscribe now

GiveDirectly has since launched studies testing MobileAid in Bangladesh, Malawi, Kenya, and the Democratic Republic of Congo. In Bangladesh, for example, phone metadata was consistently more accurate than community targeting, and though it was less accurate than full proxy-means tests, it was far faster and cheaper to deploy at scale. When researchers simulated large programs operating under tight budgets — screening many households with limited funds — phone-based targeting showed the greatest welfare impact per dollar spent, exactly because it could move quickly when surveys could not.

Overall, it seems that phone-based ML-assisted approaches aren’t the most precise ones available, but they outperform most rapid alternatives and dramatically reduce the cost and time required to act. For emergency scenarios, in which who is in need changes quickly and getting help sooner can be dramatically better, it’s easy to see why this is an attractive tool.

One obvious concern with algorithmic targeting is bias: do these models systematically exclude already-vulnerable groups? In the Togo study, researchers explicitly compared error rates across gender, age, religion, household type, and ethnicity, and found no evidence that the machine-learning approach was more likely than other targeting methods to wrongly exclude any demographic groups. I was initially skeptical that these models wouldn’t have demographic bias given differential phone ownership and use, so this was a notable and important aspect of the study for me.

That said, the biggest risk seems to be not model bias so much as coverage. Phone-based targeting can only reach people who have access to a phone. While mobile penetration is now very high in many low- and middle-income countries, around 97% of households in Bangladesh and over 90% in Kenya have at least one phone, ownership is still lowest among the very poorest households. This increases the risk of excluding precisely those most in need.

MobileAid systems are also typically more expensive to set up than traditional in-person surveys. They require regulatory approval, telecom partnerships, and technical integration. But once those systems are in place, they become much cheaper in terms of marginal cost — that is, per additional household screened. In Bangladesh, researchers estimated that while in-person proxy-means testing remained the most accurate method, phone-based targeting was the most cost-effective option for large programs operating under tight budgets, where governments might spend only $10-50 to screen each household and need to reach many people quickly.

GiveDirectly’s response to these tradeoffs strikes me as pragmatic: hybrid designs that use phone data where it’s strong but maintain in-person outreach to avoid excluding the poorest households. In Malawi, for example, about 35% of recipients were identified using phone metadata, with the remaining 65% enrolled through in-person outreach. The combined approach cut the cost of identifying and enrolling each additional recipient roughly in half compared to a fully field-based program, while still reaching households without phone access. Similar hybrid models are now being tested in Kenya, where households in flood- and drought-prone areas are enrolled both digitally and in person so cash can be sent quickly when early-warning systems predict extreme weather.

During humanitarian crises, phone-based targeting offers a way to reach vulnerable households more quickly and more cheaply than existing rapid-response methods. In the wake of an unfolding crisis, any delays in aid can significantly impact the course of events: they change behavior, force bad tradeoffs, and lock in losses that can’t be undone later. Selling livestock, skipping meals, pulling children out of school are not as easy to solve with help after the fact.

This morning, I published a piece in MIT Technology Review on AI pain assessment tools, including how they work, who’s using them, and the clinical results so far. This is a companion piece focused on the questions these tools raise: what happens when we quantify something we don’t fully understand, and what clinical judgment might we lose in the process?

We don’t actually understand pain all that well.

We have a model of the basic process, of course. We know that when you stub your toe, for example, microscopic alarm bells called nociceptors send electrical impulses toward your spinal cord, delivering the first stab of pain, while a slower convoy follows with the duller throb that lingers. At the spinal cord, the signal passes through a kind of neural checkpoint — the “gate.” Depending on what else your nerves are doing and what your brain instructs, that gate can amplify or dampen the pain before it ever reaches your brain.

But there’s a lot we don’t know, especially for something that one in five U.S. adults report experiencing every day or most days. For example, scientists can’t predict what causes someone to slip from a normal injury into years of hypersensitivity. Some research suggests that genetics, past injuries, and a person’s environment, stress levels, and even their thoughts and feelings might all contribute to whether they develop chronic pain. Americans have also been experiencing more of it over time, and we’re not entirely sure why.

Phantom-limb pain remains equally opaque. About two-thirds of amputees feel agony in a part of their body that no longer exists, and yet the competing theories we currently have don’t fully explain why the other third feel nothing.

There is some identified link between having red hair and lower pain tolerance and red heads need up to 20% more anesthesia. Congenital insensitivity to pain, a rare genetic disorder, makes some unable to feel pain at all. Outlandishly, a Scottish woman named Jo Cameron is purportedly incapable of suffering.

Pain is also stubbornly subjective. Feedback from the brain in the form of your reaction can change how you physically feel, meaning that expectation and emotion can change how much the same injury hurts. And pain can also be affected by a slew of external factors. In one study, experimenters applied the same calibrated stimulus to volunteers from Italy, Sweden, and Saudi Arabia, and the ratings varied dramatically. Italian women recorded the highest pain scores, while Swedish and Saudi participants judged the identical burn several points lower, implying that culture can amplify or dampen the felt intensity of the same experience.

How much pain you expect to feel can also directly alter your experience: in one trial, volunteers who believed they had received a pain relief cream reported a stimulus as 22% less painful than those who knew the cream was inactive — and a functional magnetic resonance image (fMRI) of their brains showed that the drop corresponded with less activity in the parts of the brain that report pain, meaning they really did feel less hurt.

All of this to say, our understanding of pain is far from complete and the phenomenon is far from straightforward.

Subscribe now

How much does it hurt?

We usually measure pain on a self-reported 1-10 scale, where 1 is very mild and 10 is the worst pain imaginable.

This is the standard of care for a reason. Pain scales are relatively quick and easy to use, they provide us with some basic information about how much pain someone feels they are in at the moment, and they can track relative pain for the same person over time reasonably well.

On the other hand, some people can rank their pain at an 8 out of 10 and still go to work, while others can’t get out of bed with a score of 4. How people interpret the question, their personal pain tolerance, and fluctuations throughout the day all influence the answer in ways that make it hard to use clinically.

It’s also hard to make the scale concrete, because people often have had different experiences to reference. For example, we couldn’t say ‘breaking a bone is a 5’ because many people haven’t broken a bone, and those who have might have had wildly different experiences depending on which bone, under what circumstances, and so on.

Some start-ups are trying to change this, making pain a little more legible. Using AI, they provide direct inferences on how much pain a person is in on that same numerical scale through readings of things like facial expressions, pulse, heart rate, or body temperature. Clinicians can then use the output to inform medical treatment plans.

I wrote about this in an article out today in MIT Technology review:

PainChek, one of these behavioral models, is already being used in health care settings. It acts like a camera‑based thermometer, but for pain: A care worker opens an app and holds a phone 30 centimeters from a person’s face. For three seconds, a neural network looks for nine microscopic movements—upper‑lip raise, brow pinch, cheek tension, and so on—that research has linked most strongly to pain. Then the screen flashes a score of  0 to 42. “There’s a catalog of ‘action‑unit codes’—about 52 facial expressions common to all humans. Nine of those are associated with pain,” explains co‑inventor and senior research scientist Kreshnik Hoti. This is built directly on the foundation of FACS. After the scan, the app walks the user through a yes‑or‑no checklist of other signs, like groaning, guarding, and sleep disruption, and stores the result on a cloud dashboard that can show trends.

These companies are initially targeting patient populations that can’t communicate — like those in dementia care, infants, and surgical patients under anesthetics.

I talk about the types of technologies at play and how they’re being used in more depth in the article. But beyond their technical promise, I’m interested in what they might change about how we understand pain itself.

Subscribe now

Automating empathy

There’s some reason for unease: what happens when we start delegating the measurement of pain, which we can’t even define precisely, to algorithms?

These systems inherit the limits of the data that train them. Across domains, from self-driving cars to voice recognition, machine-learning models consistently perform worst on the kinds of cases they’ve seen least. This can mean consistently worse performance on groups as widely ranging as people with darker skin tones, children, or stroke survivors. In the context of pain measurement, that might mean underestimating pain in patients whose facial expressions or vital signs differ systematically from those represented in training data.

On the other hand, people’s self-reported pain scores and clinicians reactions to them are notably biased. Like I mentioned before, things like culture, experience, or expectations can all alter what someone reports their pain level to be. And bias inside the clinic can drive different responses even to the same pain score. A 2024 analysis of discharge notes found that women’s pain scores were recorded 10% less often than men’s. At a large pediatric emergency department, Black children presenting with limb fractures were roughly 39% less likely to receive an opioid analgesic than their white non-Hispanic peers, controlling for pain score and other clinical factors. These algorithmic outputs might be systematically biased, but so are humans.

Relatedly, as I noted in my previous article on radiology, having a model output to refer to can change clinician behavior. Once a pain score appears on a screen, clinicians may defer to it more than they should, even when their judgement would steer them in a different direction. This “automation bias” has been documented across fields: people trust machine output more than it might warrant. One review found that when a system gave incorrect guidance, clinicians were 26 percent more likely to make a wrong decision than unaided peers. With a specific numerical output, there’s an added risk of false precision. Getting a 7.5 out of 10 might feel like much more information than the number actually contains.

Over time, reliance on automated assessment could also dull clinicians’ own interpretive skills, replacing judgment with trust in the tool. There is some early evidence for this deskilling effect: a recent study in the Lancet found that doctors who spent three months using an A.I. tool to help spot precancerous growths during colonoscopies were significantly worse at finding the growths on their own after the tool was removed.

And beneath all of this sits something harder to quantify. Diagnosis has always been, at least in part, a relational act: it requires attention, context, and the willingness to sit with someone else’s discomfort. When we offload that act to a machine, it risks becoming overly procedural. I’m not making a purely humanist claim here; when pain is recorded automatically, clinicians may ask fewer questions or miss patient cues. The risk is that medical professionals start to see less of the full picture of any given individual patient.

I don’t mean for this article to be pessimistic: these tools have real promise, especially for patients who can’t communicate their pain, and my original article focuses more on the positive, exciting aspects of the frontier of algorithmic pain measurement. But the more we quantitatively measure complex signals like pain, the more tempted we may be to mistake the number for the thing itself. How we choose to use these tools will decide whether AI deepens our understanding of a phenomenon, or causes us to simplify it and stop looking too soon.

In my last few posts, I wrote about why radiologists have not been displaced by AI, in spite of rapid advancement in models that superficially appear to be as capable as their human counterparts. I suggested that understanding what happened (or didn’t) to radiologists would shed light on what we can expect for some other professions in the next few years.

But is radiology just a unique case? Let’s look at one argument I made in a recent Works in Progress article a little more closely: elasticity of demand. Even when models are capable of taking over some of a radiologist’s workload, I suggested in the piece, the ability to read scans cheaper and more quickly might actually cause an increase in demand for those scans. That increase, if large enough, could offset the additional assistance from AI and instead cause greater need for radiologists.

This isn’t without precedent. Elasticity of demand showed up when ATMs spread across the U.S. A study by James Bessen showed that teller employment rose during the 2000s, growing 2% annually, and he argued that this was not despite ATMs, but in part because of them.

Bessen used detailed U.S. occupational data from 1980 to 2013 to analyze how computer use affected job growth, skill demands, and wage distribution across over 300 occupations. His estimates controlled for factors like offshoring and industry-level demand shifts, allowing him to isolate the specific effects of computer automation.

He suggested that, by reducing the cost of operating bank branches, ATMs made it economically viable for banks to open more locations — expanding, rather than contracting, teller employment. He supports the claim not through a formal causal estimate, but rather through historical correlation and economic reasoning: employment rose during the period of ATM adoption, and theory suggests that automation can increase demand for complementary labor when it lowers costs.

The assumption that productivity gains would translate to fewer workers didn’t hold because the demand for the service wasn’t fixed. Some of that demand may have been previously suppressed by supply constraints — for example, if limited branch hours and long lines had kept some customers from doing as many transactions as they wanted — and more of it might have been stimulated by new capabilities, expectations, and incentives. Rather than remaining constant, demand expanded as the effective supply increased, a dynamic similar to a rightward shift in the supply curve, where prices fall and the equilibrium quantity rises.

ATMs automated routine cash-handling but left tellers doing everything from fraud detection to retirement account setup. Far from replacing tellers, automation reallocated their time; they focused more on customer service and cross-selling. The teller example illustrates a broader pattern in Bessen’s research: occupations that adopted computers often grew faster than those that didn’t, partly by substituting for other jobs in the same industry. At the same time, computerization tends to increase wage inequality within occupations, as only some workers acquire the new, valuable skills these technologies demand.

Subscribe now

This pattern is called Jevons paradox. When technological advancement makes something cheaper or more efficient to use, this can conversely result in more use if demand increases as the cost declines, and this effect dominates. It’s often used in the context of goods — for example, the increased efficiency of using coal leading to the consumption of coal in more industries and a net increase in coal consumption.

Since then, however, it’s been applied to professions where increased efficiency means that one would need fewer people to do the same amount of work — but that increase in efficiency means that work can get done faster and more cheaply, resulting in more people demanding it. That the advent of spreadsheets resulted in more demand for accountants, rather than less, is another commonly-cited example.

In legal services, there may be a similar effect. Predictive-coding tools can reduce the burden of document review in large-scale litigation. But instead of cutting legal headcount, they enable lawyers to take on larger or more complex cases. Firms still staff teams with associates, but now those associates may integrate the AI’s outputs into strategic decision-making, respond to its limitations, and navigate risk in higher-stakes environments. The work shifts upward: demand has grown, but especially for specialists. A 2017 paper by Dana Remus and Frank Levy estimated that, even under widespread adoption of e-discovery AI, the net effect on overall legal employment would be modest. The tasks that remain require context, interpretation, and discretion, and these are skills that machines still struggle to emulate.

But this dynamic has its limits. Economists distinguish between the scale effect, where cheaper and faster services expand output and increase labor demand, and the substitution effect — where machines directly replace human inputs. When automation reduces costs modestly, the scale effect can dominate, like with ATMs and bank tellers. But if productivity gains come very quickly and are extreme, demand growth can’t keep pace indefinitely.

In the case of bank tellers, online and mobile banking was a stronger substitute for their skills, and the potential for additional demand didn’t compensate for that. As a result, employment in the role has declined by ~30% since 2010. This applies across roles: if the technological advancement either completely substitutes for all tasks needed or introduces efficiency gains so large they outstrip any potential increase in demand, then employment in the role will fall rather than rise.

In radiology, for instance, patients will likely never need a thousand times more scans than they do today, no matter how cheap they become. Once demand saturates, the substitution effect takes over, and employment declines. We’ve already seen this dynamic in industries like textiles and agriculture, where mechanization enabled massive output growth but ultimately displaced large numbers of workers once consumption hit a natural ceiling.

Elasticity of demand complicates the story of automation. Cheaper, faster tools don’t always shrink the need for human labor; they can expand it by unlocking suppressed demand or creating new tasks that complement machines. That pattern, visible in radiology, banking, and law, suggests that automation’s most durable legacy may not be mass displacement, but reallocation — reshaping what workers do and where expertise is needed.

However, this requires several things to be true:

• Technological change results in increased efficiency, but does not entirely substitute for the role

• The efficiency boost lowers the cost of use — this can be non-monetary, for example, in the case of faster turn-around of scan results

• That lower cost results in more demand for the service

• The magnitude of the demand increase outstrips the increase in efficiency in impact

The balance between scale and substitution is not fixed. In my next piece, I’ll turn to the other side of the ledger: when substitution overwhelms scale, and what characterizes professions where automation does quickly reduce the need for human workers.

Last week, I wrote about why radiology seems like the perfect candidate for automation — the canary in the coal mine that should tell us when we’re all about to be out of a job. AI models that outperform clinicians on radiology benchmarks have existed for many years, and many experts have repeatedly predicted their demise over a decade ago. Yet, radiologists are thriving.

I wrote a piece about this in Works in Progress.

Subscribe now

One historical case study I found especially interesting: at one point, Medicare reimbursed radiologists $7 more per mammogram if they used an early form of AI that performed very well on benchmarks to help them do it. But by 2018, they rolled back that policy.

Computer-aided diagnosis turned out to be a disappointment. Between 1998 and 2002 researchers analyzed 430,000 screening mammograms from 200,000 women at 43 community clinics in Colorado, New Hampshire, and Washington. Among the seven clinics that turned to computer-aided detection software, the machines flagged more images, leading to clinicians conducting 20 percent more biopsies, but uncovering no more cancer than before. Several other large clinical studies had similar findings.

Radiology is, in many ways, about a decade ahead of other white collar fields in automation. I find technical, legal and regulatory, and economic reasons algorithms have not taken their jobs (but rather, in some cases, made more of them!). In the article, I investigate what it can tell us to expect in the first years after AI models are competitive with us at our core tasks, according to benchmarks. In some cases, better AI may mean more working humans, not fewer.

Depending on who you ask, AI is either decimating entry-level jobs or going through an ineffectual hype cycle. A New York Times column profiled Mechanize, a San Francisco start-up whose founders openly say their goal is to “fully automate work” — from coding and research to law and medicine — “as fast as possible”. Axios reported on Anthropic CEO Dario Amodei’s stark warning that AI could wipe out half of all entry-level white-collar jobs within just five years, calling it a potential “white-collar bloodbath”.

Economists are starting to find evidence that this disruption may already be underway. A new paper by Erik Brynjolfsson and co-authors at Stanford, using millions of payroll records, found that employment for young workers aged 22–25 in highly AI-exposed jobs, like software developers and customer service reps, fell by 13% since the launch of ChatGPT, even as older workers in the same roles kept steady or grew. Derek Thompson summarized this research as the strongest evidence yet that AI is already destroying entry-level jobs for young people.

But not everyone sees the same picture. The Economic Innovation Group, analyzing government survey data, argued there is “little evidence of AI’s impact on unemployment” overall, noting that most firms still report no net change in hiring due to AI. Other commentators, like Noah Smith and John Burn-Murdoch, have highlighted that much-discussed drops in graduate hiring don’t present a clear picture, or appear to be reversing. Economist and newly-appointed Dean of Harvard College reassured his students that “this moment, which feels risky, is actually a tremendous opportunity for you” — pulled from his post entitled ‘Shorting the AI jobs apocalypse’.

This back-and-forth captures why the moment feels confusing: AI is clearly advancing in capability, but whether it is actually eliminating jobs at scale remains contested. To cut through the noise, it can help to look at a single case study.

Subscribe now

Built for automation

And, if any profession should have given us early clarity, it’s radiology. On paper, it looks almost designed for automation:

• The work is digital. Radiologists deal in images, already captured and stored electronically. Unlike engineers inspecting physical samples or judges hearing live testimony, radiologists begin with clean digital inputs, ready for software to process.

• There is a strong pattern recognition component. When radiologists look for certain indications, they are thinking about the many other images they have seen throughout their training and recognizing patterns across them, a type of task where AI is relatively strong.

• The tasks are repetitive. The job involves scanning for the same set of anomalies over and over — lung nodules, hip fractures, breast tumors. There is variation in each case, of course, but compared with the open-ended problem solving of strategy consulting or the creative drafting of advertising, the work looks very similar day to day.

• It’s easy to grade. Did the system flag the pneumonia? Did it miss the hemorrhage? Success can be measured against relatively clear ground truth, something much harder to define in fields like management or education where outcomes are diffuse and long-term.

• There’s plenty of data. Hospitals produce millions of scans every year, far more than any individual radiologist could ever look at. Where many professions struggle with data scarcity, radiology is drowning in examples.

• The human element is lighter. Radiologists do interact with patients, but less than many other physicians. A core part of the work happens in dark rooms, behind monitors, focused on images rather than on face-to-face conversations. That makes it easier to imagine machines taking over, compared with professions that involve more human interaction.

• There’s pressure to go faster and cheaper. Waiting for a radiologist’s report can delay treatment across entire hospital systems. If algorithms could read scans more quickly, hospitals could cut costs, shorten wait times, and treat more patients.

This has fueled widespread speculation that radiologists would be the first automated white collar profession. In 2016, Geoffrey Hinton, a pioneer of deep learning, warned radiologists were “like the coyote already over the cliff but not yet looking down.” Vinod Khosla, in 2017, said “the role of the radiologist will be obsolete in five years.”

Predictions meet reality

And AI radiology models have come as promised. Today, about 1,200 medical AI devices are FDA approved, and over 900 of them are targeted at radiology — more than every other specialty combined. They perform competitively with humans according to benchmarks: in 2019, a neural network performed better than 2 of 4 tested radiologists at identifying abnormalities on head CT scans, and a labeler performed better than three tested radiologists at detecting the presence of three of five pathologies on chest radiographs. A 2022 paper found that “a self-supervised model trained on chest X-ray images that lack explicit annotations performs pathology-classification tasks with accuracies comparable to those of radiologists.”

Yet, nearly a decade after those predictions, radiologists are not just still working; they are busier and better paid than ever. In 2025, American diagnostic radiology programs offered a record 1,451 positions across all radiology specialties, a 5.2 percent increase from 2024, and the field’s vacancy rates are at all-time highs. In 2025 radiology was the second-highest-paid medical specialty in the country, with an average income of $520,000, over 48 percent higher than the average salary in 2015.

So, why has the obsolescence so many predicted failed to manifest? This is an interesting question in and of itself, but it is also a useful case study. Radiology is ahead of other professions in terms of AI automation, and so it offers a preview from which we might extract ideas about how AI automation may play out for the rest of white-collar work, at least in the first ten or so years of capable models.

I dig into the reasons why radiologists still exist, and what we might take from that about the prospects of other industries, in a forthcoming Works in Progress Article — out next week.

At Mila, a leading AI research lab, an artist named Violeta Ayala created an AI-driven jaguar that tells stories of fire and kinship in the Bolivian Amazon. Her piece, Huk, greets exhibition visitors, selects individuals from the crowd, and delivers tailored, voice-generated reflections and stories. The experience is part myth, part machine.

These residencies and the art that they enable are ending up in places like the Museum of Modern Art in New York and Centre Pompidou in Paris. They form one of our early interactions with AI in day-to-day media. Those interactions help shape what we see as the natural purpose of AI: what uses are legitimate, exciting, or extractive. That public perception can, in turn, influence how new technologies are governed.

I wanted to, maybe a bit self-indulgently, write this short post to reflect on the broader context in which the article sits. Some of the questions that came up in my reporting echo long-running themes in academic work on emerging technologies and governance. And they can help clarify the way cultural institutions like these may influence AI governance in the future.

Cultural governance happens first

There’s a well-established idea in science and technology studies that policy and regulation aren’t the first or only ways that new technologies are governed. Instead, we often see ‘cultural governance’ emerge earlier, through things like stories and media representations. These shape how people understand a technology: what it’s for, what it threatens, and who it belongs to.

Sheila Jasanoff and Sang-Hyun Kim’s concept of ‘sociotechnical imaginaries’ captures this well through the idea that communities build collective visions of desirable futures made possible by science and technology. These imaginaries are shaped in film, in fiction, and in galleries and public spaces.

That’s what drew me to this story. Artist residencies might seem peripheral to the AI debate. They’re not developing models or filing lawsuits, but they matter because they shape what feels legitimate and play a role in defining what AI means to people, not just what it does.

Subscribe now

Framing as a form of power

In talking with curators, ethicists, and artists for the piece, I kept coming back to the idea of framing: a concept that spans media studies, political theory, and moral psychology. The way a technology is framed early on influences how it’s later debated in courts, in policy, and in public imagination.

AI that appears in a Discord meme might feel unserious, but AI framed as a tool of careful artistic inquiry, displayed at the Museum of Modern Art, might elicit different reactions entirely. That shift can eventually normalize new uses of technology.

One ethicist I spoke to called this “a kind of soft governance.” Another described the artists and institutions hosting these residencies as “norm entrepreneurs”, a term from international relations theory used to describe actors who shift what is culturally acceptable before formal rules exist.

History rhymes

The pattern isn’t new. In 1908, the US Supreme Court ruled that piano rolls, an early music-reproduction technology, weren’t subject to copyright because they weren’t readable by the human eye. The cultural response from musicians and publishers was swift, strong, and negative. A year later, Congress passed the Copyright Act of 1909, establishing new rules for mechanical reproduction.

In this particular case, laws followed sentiment, and sentiment followed framing. Now, artist lawsuits against Stability AI, Midjourney, and others are winding through US courts. In the meantime, institutions like Villa Albertine, SETI, and Max Planck are funding artists to use AI tools, and encouraging them to reflect on authorship, ownership, and labor.

These programs don’t erase many artists’ concerns about generative models: the use of scraped data, the loss of attribution, and the risk of eroding creative livelihoods. But they do play a role in determining how AI art is perceived by audiences.

Of course, these dynamics don’t always follow a clean arc. Cultural perception doesn’t guarantee policy change, and symbolic framing can falter, especially when legal, economic, or political incentives push hard in the opposite direction. This piece isn’t a prediction or an answer, but a snapshot of one ecosystem engaging with a small part of a much bigger question.

What I found most interesting in reporting this piece is that no one I spoke to within these residencies framed them as advocacy. They didn’t claim to be pro- or anti-AI:

“We’re not choosing sides so much as opening space for inquiry,” says Mohamed Bouabdallah, Villa Albertine’s director. “Some residents may critique AI or explore its risks.”

Instead, they emphasized giving artists the space and time to figure out what this technology is to them and to the audiences with whom they’re engaging.

In the years ahead, judges and policymakers will weigh in, but by the time those decisions arrive, many of the symbolic battles may already be won or lost.

In Denmark, people say a year is worth about $24,000. In Japan, maybe $67,000. Not infinite. Not priceless. Just... a number.

It sounds cold to put a price on life, but we do it every day. When an insurance company decides whether to approve a drug, when a hospital allocates ICU beds, or when a development agency funds one intervention over another, they are implicitly answering the same question: how much are extra years of life worth?

This post is part 1 in a series on how people and policymakers actually value those trade-offs. Here, I look at what people say when asked directly how much a healthy year of life is worth to them. The answers are oddly consistent—and consistently strange. In part 2, I’ll explore what people actually do: the risks they take, the jobs they accept, and the trade-offs they make with real money and real danger.

If you ask people directly how much a year of life is worth, they will usually try to answer. They’ll pause, reflect, and perhaps wince — but they’ll tell you. Economists call this the stated preference approach, and for decades they’ve used it to tease out the monetary value people assign to health and longevity. The idea is simple: present people with hypothetical scenarios involving risks to life or health, and ask what they'd be willing to pay to reduce those risks. Aggregate the answers, and you get a kind of market price for life, at least in theory.

In Denmark, one economist asked a cross-section of the population how much they’d personally pay, out of pocket, for a treatment that would give them one extra quality-adjusted life year (QALY) – defined as a year of perfect health. The average answer was about DKK 88,000. This was $14,660 in 2003 – or ~$24,000 in today’s dollars. That’s not particularly high for what is an almost stereotypically wealthy and health-conscious country. In the Netherlands, a similar study found that people would pay a bit more: between €12,900 and €24,500. Converted into today’s dollars, that’s $24,000-45,000. Instead of inferring preferences from choices between health states, the Dutch researchers just asked people to name their price. What’s one year of full health worth to you? Turns out that for many people, it’s somewhere between a used car and a new one.

But when you add some risk, something strange happens. In a follow-up study, the Dutch team asked people how much they’d pay for a treatment that had a chance, say 10%, 40%, 70%, or 90%, of working. Just like in real life, where everything from malaria nets to surgeries to vaccines reduce your risk of illness but don’t guarantee safety. Under standard economic theory, your willingness to pay should scale with the probability: if you'd pay €90,000 for a 90% chance of a drug working, you’d pay about €9,000 for a 10% chance of the same drug working. But that’s not what they found.

Instead, people’s willingness to pay dropped far less than expected as the probability declined. For example, someone might pay €60,000 for a 90% chance, and still offer €25,000 for just a 10% chance — nearly three times more than expected. If you extrapolate under the assumption that people’s willingness to pay should scale linearly with probability, their answers implied a QALY valuation of over €250,000 ($400,000 today); over three times what the previous study suggested. This was the same researchers, same methodology, and same country.

To figure out how much people were overweighting, they treated responses to the 90% chance as a reasonable anchor, assuming people process near-certainty more rationally, and compared it to how they behaved at lower probabilities. Based on those comparisons, they estimated that people were treating a 10% chance more like 30-40%, inflating the likelihood. Once they corrected for that distortion, the implied value people placed on a QALY fell from €250,000 down to a consistent range of €80,000 to €110,000 ($130,000 to $190,000 in today’s terms).

It’s been shown across many settings that people overweight small probabilities, which could cause them to overestimate the benefit of the treatments. Maybe when the question got more complicated, the answer also felt harder to pin down – and respondents defaulted to playing it safer with their health.

Subscribe now

In Japan, researchers explored a different angle: what if the health you’re buying is relief from something truly awful? Participants were asked how much they'd pay for a treatment for a specified illness. Across scenarios, the health gain was fixed — either 0.2 or 0.4 QALYs — but the initial health states ranged from mild to severe, based on EQ-5D utility scores. The results showed that people were willing to pay more per QALY when the starting condition was worse. For mild states, the average valuation was around JPY 2 million ($25,000 today) per QALY; for severe ones, closer to JPY 8 million ($100,000). The overall average was JPY 5 million ($67,000), aligning with other Japanese estimates.

This finding challenges a central assumption of cost-effectiveness analysis: that a QALY is a QALY, regardless of context. In theory, a one-QALY gain should be valued the same whether it’s achieved by relieving mild pain and extending someone’s life or by rescuing someone from severe illness. But, in practice, people placed a higher monetary value on QALYs gained from more severe states. That suggests health improvements aren't perceived as interchangeable units — people care not just about how much health is gained, but from what baseline state. For policymakers, this raises tough questions: should interventions that relieve extreme suffering be funded at a higher cost per QALY? Should thresholds for cost-effectiveness vary based on severity? Studies like this suggest many people would say yes.

Across the ocean, a team in the U.S. posed a similar question, but grounded it in a familiar pain: shingles. They asked people, some who’d had shingles and some who hadn’t, what they’d pay to avoid contracting the disease and suffering a bout of temporary pain and disability. The results spanned the map. Median values clustered around $7,000 to $11,000 per QALY ($10,000-$15,000 today), but the means were much higher, around $26,000 to $45,000 ($36,000-$63,000), thanks to a few people who were willing to pay eye-watering amounts. Those who’d actually experienced shingles – well, unsurprisingly, they were willing to pay more. In this case and the Japan study, people seem to be willing to spend more to avoid pain they remember or can clearly imagine, as in the case of specific, painful diseases.

Subscribe now

And in Spain, a group of researchers added another twist: what if you’re paying with your own money versus paying through taxes? When asked what they’d pay out of pocket for a year of perfect health, people gave an average answer of €10,119 ($17,000 today). But when the question was reframed as a public cost, in terms of what society should pay through taxes, that number nearly tripled to €28,187 ($48,000 today). People think it’s worth spending more collective money than their own. Maybe this is because public spending feels more abstract – a rise in taxes can seem less concrete than spending out of pocket, even if by the same amount – or because we see health as a natural shared social responsibility, or because it feels indulgent to spend for another year of perfect health for oneself.

Taken together, these studies form a patchwork portrait of how people say they value life. The numbers vary, but not wildly: when converted into today’s U.S. dollars, most estimates fall between $25,000 and $100,000 per healthy life-year, somewhere between the cost of a Rolex and a year of college. A 2022 systematic review of 39 studies and 511 observations globally found a mean value of $52,619 per QALY ($67,000 today), with some higher estimates in specific regions (for example, about $98,450 in Quebec). That’s a provocative finding. It suggests that, at least in the abstract, people treat an extra year of life less like a priceless treasure and more like a durable good: valuable, yes, but bounded.

These numbers aren’t uniform, and one of the clearest patterns is that, unsurprisingly, wealth matters. People in higher-income countries tend to give higher valuations, and, within countries, richer respondents are willing to pay more. In the Netherlands study mentioned before, for example, willingness to pay per QALY varied from €5000 in the lowest to €75,400 in the highest income group — ~$10,000-140,000 in today’s dollars. Education sometimes raises valuations too, though inconsistently, while older respondents tend to report lower willingness to pay than younger ones.

People are also more generous when costs are distributed across society as a public good than when they’re asked to pay out of pocket. When the stakes feel more emotionally charged, like a disease you’ve personally experienced, the numbers tend to climb. In health valuation, as in so many other domains, we do not coldly optimize. We react to stories, to fairness, to mental images of who might suffer and how. The same life, the same year of health, can be deemed two or three times more valuable depending on whether the question feels personal or collective, distant or near.

In spite of the many survey studies that have been conducted, there are still important open questions about this work. Why exactly are people more generous when the money comes from taxes rather than their own wallets? What are they imagining when asked to value a “year of full health”? These factors reveal how much context, framing, and personal experience shape people’s responses.

That matters for policymakers. If valuations shift considerably depending on how a question is asked or how vivid the scenario feels, it could mean we are undervaluing diseases that are abstract, unfamiliar, or hard to picture, or underestimating willingness to support public health spending by relying on estimates based on private trade-offs. These gaps deserve more scrutiny, especially if we're using these numbers to decide when to spend on health.

But whatever their quirks, these survey results give us something concrete: a set of numbers that anchor abstract talk of health and safety to the world of money and trade-offs. They are imperfect, but they are honest in their own way. They tell us that when asked to put a price on life, people don’t reach for the infinite; they reach for a figure they can imagine paying, or taxing, or justifying.

I intend for this to be part 1 in a series of posts about valuing trade-offs between health and income. In the next one, I’ll look at what happens when people don’t just answer surveys, but instead make real trade-offs between money and risk. If you’re interested in the ways we reveal our values through action, stay tuned. And if there’s a specific question you’re interested in related to how we value health, let me know.

But this piece isn’t about those.

It’s about weird ideas—edge cases and overlooked risks where the research is far from settled, but might be doing way more harm (or good) than we think. Each of the examples that follows is grounded in data, but not enough to be conclusive. They're strange, they're striking, and they’re still uncertain.

What they have in common is this:

1. Surprising upside: The impact, if real, could be much bigger than anyone expects.

2. Not easily dismissed: The evidence is credible enough to take seriously.

3. Still an open question: We don’t know yet whether it’s true—but it’s worth asking.

These are the kinds of ideas I think we should poke at. If they turn out to be real, it could reshape how we think about improving health.

Subscribe now

Biodiversity conservation

Diclofenac is a common painkiller, used in humans and animals alike. After its patent expired in 1993, cheap generics flooded the market and became widely used in Indian cattle.

That turned out to be a disaster for vultures.

When vultures fed on livestock treated with diclofenac, they died—usually within two weeks. India’s vulture population crashed. One species, the Indian white-rumped vulture, declined by over 97%.

This matters more than it might seem to. Vultures perform a public service by eating the bodies of animals. When they vanish, carcasses pile up, polluting water and attracting rats and wild dogs. The pollution leads to diseases like gastroenteritis, cholera, typhoid fever, hepatitis A and E, leptospirosis, and cryptosporidiosis; the rats and stray dogs spread diseases like rabies. One study, Frank et. al in the American Economic Review, estimated that in parts of India where vultures once thrived, the die-off led to a 4% rise in human mortality. The authors attribute 500,000 excess deaths over five years to the crash.

If true, given most of those deaths would have been from rabies or enteric infections, it would mean around 3.8 million disability-adjusted life years (DALYs) lost per year over those five years. That’s about half the global burden of cervical cancer during that time period.

Luckily, this is a problem with some obvious fixes. Diclofenac has been banned for veterinary use in India since 2006, but enforcement is patchy. Cracking down on illegal sales, promoting meloxicam (a vulture-safe, cheaper alternative), and supporting captive breeding and release programs could help reverse the decline. In the meantime, stopgaps like carcass incinerators can reduce the risk of disease spillover while populations recover.

Frank also published another striking finding—this time in the U.S., and this time about bats. White-nose syndrome, a fast-spreading fungal disease, has decimated North American bat populations. Some bat species eat up to half their body weight in insects each night. When they disappeared, farmers responded by spraying more pesticides—about 31% more, on average, in affected counties.

Pesticides come with health risks, especially for children. Frank’s study finds that infant mortality from internal causes rose by about 8% after the bat die-off. Another unbelievably large figure.

Both studies suggest that protecting certain animal populations might have large, direct effects on human health that we’re overlooking. But there are good reasons to be cautious. These are outlier results; there isn’t much else in the way of evidence for estimates of this magnitude for the impact of biodiversity loss on human mortality. There’s also the possibility of publication bias. In particular, since both papers come from the same author, this may be driven by a file drawer effect, where a researcher looks at many potential similar cases but the null findings are less likely to see the light of day.

Still, if these effects are real, they could change how we think about conservation. Saving vultures or bats wouldn’t just be about biodiversity—it could also be a form of public health policy.

Replacing dirt flooring

Dirt floors are breeding grounds for disease. Diarrhea, parasitic infections, and respiratory illness are all more common in homes without finished flooring. And downstream of those health problems come others: poor cognitive development, higher mortality, and lower income over time. Roughly a billion people still live in homes with dirt floors today.

In 2000, a program in Mexico tried something simple: it gave households up to 430 square feet of concrete—about $150 worth—to replace their dirt floors. A follow-up World Bank evaluation found that doing so led to a 78% drop in parasitic infestations, 49% less diarrhea, 81% less anemia, and up to 96% improvement across early childhood cognitive tests. Adults reported feeling less depressed and more satisfied with life, even 15-18 months later.

There’s an important catch: the study wasn’t randomized. Families weren’t selected by lottery or assigned randomly, but rather chosen based on their poverty status and household characteristics, so it’s possible the households that received cement floors were already different in ways that mattered. To adjust for this, researchers used a matched cohort design, comparing each treatment household to one that looked similar on observable traits. But unobserved differences—like motivation, or local infrastructure—could still explain part of the effect.

A replication team led by Basurto et al. reanalyzed the study. They dug into the sampling strategy, cleaned up some coding errors, and tested alternative assumptions and analytical methods. Despite all that, their results closely matched the original and were robust to different analytical decisions: flooring really did seem to improve health and cognition.

Still, the effect sizes are so big that I feel like this story is likely missing important context. The authors found a: 49% reduction in diarrhea, 81% in anemia. But from what baseline? If anemia was 10% prevalent, an 81% decrease is meaningful but modest in absolute terms; about eight fewer cases per hundred people. If it was 70%, that’s a very different story. Without clear baseline rates, durations, or age-specific disability weights, it’s hard to translate these results into a comparable metric.

That said, let’s play it out. If these numbers are even half right, this is a wild finding. A one-time $150 slab of concrete seems to outperform daily iron pills (which reduce anemia by ~39%) and nearly match handwashing campaigns for diarrhea prevention (estimated at ~40%). And unlike those interventions, it requires no behavior change and lasts for decades.

That’s a remarkable return for a humble flooring upgrade. If this replicated elsewhere, replacing dirt floors might be one of the highest-return infrastructure investments for child health.

Which is also why it’s hard to swallow. Can concrete really beat decades of targeted public health programming? Maybe this context was uniquely favorable. Maybe the design exaggerated the effects. Or maybe there’s something deeper going on—like flooring acting as a proxy for stability, security, or perceived agency. Either way, if it’s real, even a little, it’s too big to ignore.

There are a few more reasons for caution. The method for measuring parasitic infection wasn’t clearly stated, raising the possibility that the data came from self-reports (which are more vulnerable to bias). People who got the concrete may have genuinely felt better and more optimistic, and that could shape how they remembered or reported their health. And this was early 2000s Mexico. Many regions today may have lower baseline disease risk due to deworming or sanitation improvements.

Still, the possibility that something so basic could drive these kinds of outcomes on health, cognitive development, and subjective well-being is worth a closer look. I haven’t come across any completed randomized trials that could triangulate these findings, but I’m aware of at least one large RCT underway, with results expected in a few years.

Reducing environmental noise exposure

We know that extremely loud noise—like standing next to a jackhammer—can damage hearing. But a growing body of evidence suggests that much lower levels, when experienced constantly, may also pose serious health risks. Chronic environmental noise, like from traffic, trains, or construction, has been linked to heart disease, cognitive decline, and even early death.

Noise is measured in decibels (dB). A whisper is about 30 dB, typical conversation clocks in at 60, a vacuum cleaner hits 75, and a train can reach 100. The scale is logarithmic—each 10 dB increase doubles how loud a sound feels.

At high volumes—above 85 dB—noise can damage your hearing. But for more subtle health effects, the thresholds are lower. The World Health Organization considers average noise levels above 40 dB to be harmful. At night, that drops to 30 dB, due to noise’s impact on sleep. That’s about the volume of a quiet suburban street or a humming fridge—background noise you'd barely register.

Unlike many public health risks, noise often flies under the radar. But it’s everywhere. Nearly a third of Americans live in areas where environmental noise exceeds 45 decibels, according to U.S. Department of Transportation models.

And like air pollution, it’s not distributed evenly. Lower-income communities are often hit hardest. This makes noise a surprisingly plausible candidate for a major, under-appreciated health risk. But the evidence is still emerging, and most studies are observational.

The health links are worrying:

• A study near Heathrow Airport found that people living in the noisiest areas had a 10–25% higher risk of stroke or coronary heart disease. The study adjusted for area-level deprivation and air pollution, but not for individual income or housing value. This means the observed association could partly reflect socioeconomic differences, since quieter homes near airports tend to be more expensive and house wealthier, healthier residents.

• A 15-year cohort study in Switzerland found that each 10 dB increase in road traffic noise raised the risk of heart attack mortality by 4.3%. Compared to the earlier study near Heathrow, the Swiss cohort used individual-level data, tracked address histories, and adjusted for air pollution and spatial factors. But the study didn’t control for income or housing value directly, so some of the observed risk likely still reflects underlying socioeconomic differences.

• A meta-analysis of 14 studies reported an 8% increase in heart attack risk for every 10 dB increase in road traffic noise (within the 53–77 dB range).

• Another review, covering 24 studies on hypertension, found that every 5 dB increase in noise exposure raised the odds of high blood pressure by about 3.4%.

These are modest effect sizes. But given how common cardiovascular disease is, and how many people are exposed, the total burden could be significant.

The risks don’t stop at the heart. A 2021 study of 2 million Danish residents found that long-term road noise above 65 dB was associated with a 16% higher risk of Alzheimer’s compared to levels under 45 dB. A study in China found that factory workers exposed to occupational noise scored lower on standard cognitive tests. In animals, chronic noise exposure disrupts sleep, activates stress hormones, and damages brain regions involved in memory, and prenatal exposure seems to make the effects worse.

The WHO now ranks traffic noise as the second most harmful environmental factor in Western Europe, after fine particulate air pollution.

Still, these are all correlational findings. Researchers try to control for confounders like air pollution or smoking, but it’s hard to isolate the specific effect of noise. The mechanisms are plausible—stress, sleep disruption—but still speculative. Since noise often co-occurs with poverty and pollution, there’s a real risk the effect sizes are inflated.

Beyond health, noise may also affect learning and productivity. A 2015 review found that chronic noise from traffic and aircraft impairs children’s memory and reading comprehension. A more recent study in Kenyan factories showed that persistent workplace noise reduced workers’ cognitive performance—even after adjusting for other variables. If these effects generalize, cutting noise could improve not just health but also income.

And noise isn’t random. It maps onto inequality. A global study using “hearing age gaps”—the difference between someone’s hearing age and their actual age—found that people in cities with lower average incomes often had hearing ages 15–20 years older than expected.

In New York City, exposure varies dramatically from block to block. Highways, flight paths, and industrial zones concentrate noise in places like the Bronx, Queens, and parts of Brooklyn.

This all makes noise especially hard to study, but potentially high-leverage. The concentration of exposure in certain areas means targeted reductions could have outsized impacts. But it also means that many studies are observational and vulnerable to confounding.

Still, if chronic environmental noise really does affect cardiovascular health, cognition, and learning—even modestly—it could be a bigger deal than we’ve realized. Like air pollution, it may be a slow, invisible driver of poor health outcomes.

And if that’s the case, we’re not helpless. There are already tools: subsidies for urban soundproofing or insulation, quieter road surfaces, better vehicle design, and “Buy Quiet” procurement policies (used by NASA and NIOSH) to encourage low-noise tools and machines.

Curious about your own noise exposure? If you’re in the U.S., you can search your ZIP code in the Department of Transportation’s noise map. Or download the NIOSH Sound Level Meter app to measure it directly. If you’re feeling generous, you can even contribute your readings to the Noise Across America dataset.

This wasn’t pixelated dogs or uncanny hands. These were cohesive, stylized, emotionally resonant images, created in seconds from mundane inputs. It’s a quiet leap forward in creativity-on-demand — and a reminder that the bar for what machines can do is rising fast. And that’s the point: the Ghibli trend isn’t just an aesthetic fad. It’s a playful demo of a deeper shift. The same model that can turn a tourist snapshot into a fantasy still is also getting better at coding, reasoning, persuasion, and planning. We’re watching frontier models progress — not just in benchmarks, but in everyday life.

Zoom out, and the Ghibli filter is a trivial use case. But it gestures at something bigger: a world where powerful systems increasingly shape what we see, what we choose, and what we believe. That opens up extraordinary possibilities, but also forces us to ask questions not just about the tools we use, but also about the goals we set. What does real progress look like when reality itself can be remade to match our desires? And how should we think about tradeoffs — between aesthetic delight and accuracy, short-term gratification and long-term stability, what feels good and what’s actually true?

Subscribe now

What I’ve written

In March 2024, The Matrix had its 25th anniversary — and Robert Nozick’s “experience machine” turned 50. I wrote for BBC Future about this thought experiment, which feels more relevant now than ever. Nozick asked whether we’d choose to live in a simulated world that perfectly fulfills our desires, even if we knew it wasn’t real. He argued that most people wouldn’t — that we crave something deeper than pleasure, and value reality for its own sake.

Fifty years later, we’re inching closer to that dilemma. People are falling in love with chatbots, spending years in immersive games, and curating digital lives that may feel more compelling than the real thing. That doesn’t mean reality is losing — but it does mean we need clearer internal compasses. In the piece, I explore how Nozick’s thought experiment holds up today, and why our concept of “reality” may be shifting as technology gets better at giving us exactly what we want.

Unlike either The Matrix or Nozick’s machine, this won’t be a single decision: AI will be increasingly intertwined with our lives in many ways. Every day, we’ll make choices about how and when to use it. That choice, over time, accumulates into a weighty one. Reading the full article when the AI summary is shorter, or spending time with friends over Claude prompted-into-therapist are examples of decisions that feel dull or effortful now, but might nourish something deeper over time. But this also gives us a chance to shape norms and defaults before they calcify — to build in friction, reflection, or values-aligned nudges while the paths are still forming.

Similar to the Matrix, Nozick's experience machine would be able to provide the person plugged into it with any experiences they wanted – like "writing a great novel, or making a friend, or reading an interesting book". No one who entered the machine would remember doing so, or would realise at any point that they were within it. But in Nozick's version, there were no malevolent AIs; it would be "provided by friendly and trustworthy beings from another galaxy". If you knew all that, he asked, would you enter the experience machine for the rest of your life?

Keep reading: 'Experience machines': The 1970s thought experiment that speaks to our times

In day to day life, we make similar judgments about what feels real and counts as meaningful. In a piece for Nautilus Magazine, I explored why we tend to devalue art made by AI, even when it’s visually identical to human work.

A series of experiments found that people consistently rated AI-labeled images as less valuable, less skillful, and less creative than the same images labeled human-made. This was true even when they reported that they could not distinguish between them. One reason may be what philosopher Robert Nozick called the “authenticity intuition” — the idea that we care about reality, not just appearances.

But the results also point to a deeper tension: even as AI systems become better at mimicking human output, we’re not always willing to treat those outputs as equivalent. That gap — between capability and perception — could shape how AI is adopted, trusted, or resisted in domains that depend on meaning, not just performance.

The authors of the study first became interested in how we value AI-driven art when a fellow Columbia researcher began holding exhibitions of machine-made works in 2016. “People would receive it differently if he gave more credit to the machine for the work,” says first author Blaine Horton, a former performing artist and current doctoral student at Columbia Business School. “They’d say it was cool, but not terribly interesting or creative. But when he took more of the credit for the work, it appeared people responded more positively.”

Keep reading: We’re biased against AI-made art

Subscribe now

What I’ve been reading

A look the rise of ‘buy now, pay later’ services, and how they are helping people get what they want today — at the risk of entrenching debt, weakening consumer protection, and undermining long-term financial stability.

What companies like Klarna once characterized as paradigm-busting behavior—young people rejecting stodgy banks in favor of more freeing forms of finance—now looks like the crest of yet another credit cycle, a familiar note in the motif of American consumption.

An investigation into a dramatic downsizing of America’s public health infrastructure under the banner of efficiency, which asks whether we’re mistaking bureaucratic function for bureaucratic bloat.

It does not take a genius to understand that pushing out 20,000 workers at our preeminent health agencies won’t make Americans healthier. It’ll just mean fewer health services for our communities, more opportunities for disease to spread, and longer waits for lifesaving treatments and cures.

An argument that real advancement means not just chasing what’s new, but taking responsibility for the problems progress creates — from toxic drugs to smog-filled cities to deadly electrical fires.

Inventing the automobile is progress, and solving the smog it creates is further progress... This isn’t a failure of progress; it is the nature of progress.

What I’ve been thinking about

The weather warmed up this week in New York, and suddenly the city flipped a switch — café tables spilling onto sidewalks, crowds laying out in Central Park, everyone shedding their coats at once. It’s the season of surface-level joy: everything brighter, lighter, easier.

But I’ve been thinking about how easy it is for everything to become surface-level. Credit apps that make you feel richer than you are. Policies that look lean and efficient until they quietly gut what matters. Advances that dazzle — and distract. And yet, that dazzle is also real. Joy and enchantment aren’t inherently shallow; they just need anchoring.

This month’s pieces are all, in their own way, about that tension: the pull of what feels good versus the value of what’s real. About how progress sometimes means going deeper, not faster — and noticing when delight comes with a cost. But also about how we might keep the delight, without losing the depth.

This month, I’ve been thinking about these often-overlooked foundations of progress. Whether in fundamental science, public health, or investments in seemingly niche technologies, breakthroughs rarely arrive in a flash of insight. More often, they emerge from a scaffolding of knowledge and resources built over years—sometimes decades—until one day, the right conditions align. In many cases, this means research that initially appears odd or impractical is actually laying the groundwork for future breakthroughs. And when those foundations are neglected, the consequences often become clear only when it’s too late.

What I’ve written

This month, I co-authored an article for Asterisk Magazine titled “A Defense of Weird Research.” Lauren Gilbert and I took a close look at how seemingly odd or impractical scientific studies—think levitating frogs or analyzing Gila monster venom—often end up delivering transformative breakthroughs that pay for themselves many times over. We wrote it in the context of looming federal funding cuts by the new Department of Government Efficiency (DOGE) — and discuss why the government is uniquely well-positioned to fund this type of research. While the current NIH and NSF review systems could certainly be improved—and Lauren and I include a few thoughts on how to do so in the piece—their funding is critical for nurturing the kind of basic research that eventually becomes tomorrow’s medical breakthroughs and market-shaping technologies.

We didn’t know that studying Gila monster venom would lead to the invention of GLP-1 agonists like Ozempic, or that horseshoe crab blood would prove crucial to vaccine development, or that studying bacteria in geysers would lead to the development of PCR, the technology which allows scientists to detect DNA in small samples, and on which much of modern molecular biology — from genetic testing and COVID diagnostics — rests.

Keep reading: A Defense of Weird Research

What I’ve been reading

This paper about how new technologies may be best-suited to the environments in the countries that develop them, and how that contributes to lower agricultural productivity in low- and middle-income countries.

An influential explanation for global productivity differences is that frontier technologies are adapted to the high-income countries that develop them and "inappropriate" elsewhere. We study this hypothesis in agriculture using data on novel plant varieties, patents, output, and the global range of crop pests and pathogens. Innovation focuses on the environmental conditions of technology leaders, and ecological mismatch with these markets reduces technology transfer and production. Combined with a model, our estimates imply that inappropriate technology explains 15-20% of cross-country agricultural productivity differences and re-shapes the potential consequences of innovation policy, the rise of new technology leaders, and environmental change.

This article about how declines in infrastructure may be contributing to widespread parasitic infections in the southern United States.

Dorsey and her family lived in Shaw, Mississippi, a town of 1,400 people about 110 miles (175km) north of Jackson. The area is plagued by sanitation problems – residents in Bolivar county filed half a dozen complaints to state officials just last year about wastewater leaks and burst pipes that have exposed them to raw sewage.

Now, researchers warn that these problems are probably contributing to widespread intestinal infections and parasites such as hookworm, roundworm and tapeworm.

This piece about why the left should embrace pronatalism, and the downsides of an aging population.

Birth rates really are falling: America’s declined 2 percent from 2022 to 2023 alone, and only six countries are expected to have birth rates above the replacement rate in 2100. The policies associated with pronatalism, moreover, naturally belong to the left, and there is a progressive case for making the country more welcoming to families in hopes of achieving a range of benefits, including a bump in the birth rate.

What I’ve been thinking about

Lately, I’ve been thinking about the tension between patience and urgency in progress. The breakthroughs we celebrate—whether in science, technology, or policy—often require decades of quiet, methodical work before they suddenly appear inevitable. Yet, the systems that sustain this kind of slow progress are increasingly at odds with a world that demands immediate results.

We see this tension everywhere. AI companies race to release new models while safety researchers warn that we don’t yet understand their full risks. Governments want economic growth but hesitate to invest in infrastructure that won’t pay off for decades. Drug discovery pipelines are squeezed by investors demanding quarterly returns, even when the next life-saving treatment could take 20 years to develop.

Foreign aid is a perfect example of this tension: it’s often the first thing cut when budgets tighten, largely because its benefits aren’t immediate or easily measured. But history has shown that well-targeted aid doesn’t just save lives—it strengthens economies, stabilizes regions, and creates long-term strategic advantages for donor countries. For example, aid efforts to combat Ebola in West Africa didn’t just prevent a humanitarian disaster; they also helped prevent a global pandemic that could have reached American shores. Beyond its obvious humanitarian benefits, in the long term, consistent foreign aid strengthens soft power and fosters more stable international relationships.

As more institutions shift toward optimizing for speed and predictability, I worry about what gets left behind: the weird research, the moonshot ideas, the unglamorous infrastructure that makes breakthroughs possible. The challenge isn’t just defending these investments—it’s finding better ways to explain why they matter before their absence makes the case for us.

But this isn’t just a story of biology. I’ve also been reading about how rigid processes in clinical trials and immigration policy create bottlenecks that limit innovation. In both cases, systems set the parameters of what’s possible. Yet, people still find ways to adapt, innovate, and challenge those constraints.

These stories seem to ask the same question: how much of the future is shaped by existing structures, and how much can we change through effort, ingenuity, and resilience? It’s a tension I’ve been grappling with in many different ways recently.

What I’ve written

This month, in Nautilus, I explored the surprising world of rove beetles—tiny insects that have evolved to mimic ants in remarkably sophisticated ways. A dozen distinct rove beetle lineages have independently converged on similar “ant-like” traits: from mimicking an ant’s narrow waist and elbowed antennae to producing the same chemical signals that allow them to slip unnoticed into ant colonies.

“The gland itself acted as this reprogrammable device… These beetles keep finding the same solution. It tells us that long-term, highly predictable phenotypic evolution is absolutely possible.”

Keep reading: The Evolution of a Mimic

The story explores how a single ancestral gland in rove beetles evolved from simple defense to a tool for infiltration, letting them intoxicate or deceive ants. Once inside ant nests, these beetles gain access to resources (and sometimes ant larvae), revealing just how inventive—yet strangely predictable—evolution can be. This repeated convergence speaks not just to the power of adaptation, but also to the crucial role of “initial conditions” in steering life along certain evolutionary paths.

Subscribe now

What I’ve been reading

This post by Jennifer Pahlka about how adherence to process and procedure can become a barrier to justice and progress.

The reality is that people are frustrated with a system in which it feels like laws — a complex, tangled, often contradictory, seemingly arbitrary web of rules that most people don’t understand — dictate outcomes at the expense of reasonable human judgement.

This report from the Institute for Progress making the case for clinical trial abundance

Promising studies never launch—not because of scientific limitations, but due to the logistics and costs of running trials. Randomized controlled trials are bottlenecks, leaving life-saving treatments backlogged in labs while patients wait.

This post from Lauren Gilbert about H1-B visas and their effect on the American labor market.

Promising ideas and innovations are being delayed or lost entirely because of restrictive H-1B visa caps—despite evidence that skilled immigration drives both job creation and economic growth.

What I’ve been thinking about

The recent U.S. foreign aid suspension has been weighing on me—not just because of the immediate risks to global health programs, but because of what it reveals about how fragile these systems can be. When policies shift, programs like HIV treatment and malaria prevention suddenly face uncertainty. It’s easy to feel like progress is entirely at the mercy of these larger forces.

But what’s been more striking to me is how quickly people and organizations have stepped in to hold the line. Researchers are rallying to quantify PEPFAR's impact, while funders are arranging bridge loans to keep NGOs afloat. These networks—often built quietly over years—spring into action, adapting to maintain stability when formal systems falter.

It’s a reminder that while structural inertia can limit what’s possible, there’s still room for agency. People find ways to push forward by creating informal safety nets and nerve centers of resilience. I’ve been thinking about how these networks become lifelines and how essential it is to build systems that don’t collapse at the first shock. In these moments, we see how much can still be changed, even within rigid systems.

At the same time, trust in science has seriously declined. The creation of the Department of Government Efficiency — a planned United States presidential advisory commission to be led by Elon Musk and Vivek Ramaswamy aiming to reduce federal spending — and related quips about cutting research funding are a prominent symptom.

It’s a reminder that progress in public health depends on more than solutions; it hinges on the systems that sustain them. Trust in science, adequate funding, and collective responsibility are as vital as the interventions themselves. The headlines this month highlight how fragile progress can be—and why it’s worth fighting to hold onto.

What I’ve written

This month, in The Economist, I explored a health issue that has quietly re-emerged in the United States: iodine deficiency. A century ago, in 1924, the U.S. rolled out iodized salt and functionally eliminated the problem. That public health victory was, it seems, temporary. Recent studies show that iodine levels have dropped by over half between the 1970s and 1990s and further still since then. Insufficiency affects a significant portion of American women, particularly those of reproductive age.

It’s fascinating—and troubling—to see how something as simple and effective as iodized salt has quietly fallen off the radar. Despite being preventable, iodine deficiency impacts 2 billion people globally and remains the leading cause of preventable intellectual disability.

Salt iodisation was never federally mandated. This has had several knock-on effects. For one, only about half of American table salt (which makes up 11% of the salt Americans consume today) is actually iodised. Faddish alternatives, like sea salt or pink Himalayan, tend not to be. More important, the salt used in processed foods—which accounts for a dominant and ever-increasing share of American salt consumption—is also iodine-free.

Changing salt consumption is not the only dietary trend at play. Decreasing demand for meat and fish, both good natural sources of iodine, is also having an effect. According to a study published in JDS Communications in February, one cup of cow’s milk, which is often supplemented with iodine, provides about half the daily intake needed for adult women. Increasingly popular alternatives to dairy, such as oat milk and soy milk, by contrast, typically offer no such benefits.

Keep reading: As wellness trends take off, iodine deficiency makes a quiet comeback

This article has an accompanying audio version here.

I went on The Economist’s daily podcast, The Intelligence, to discuss this article here.

Subscribe now

What I’ve been reading

This piece about the rise of science skepticism in the U.S. - and how it is impacting public health policy today.

Public-health officials wonder if they have sufficient clout for the next national emergency. “Science is losing its place as a source of truth,” said Dr. Paul Offit, an infectious-disease physician at Children’s Hospital of Philadelphia. “It’s becoming just another voice in the room.”

In October 2023, 27% of Americans who responded to a Pew Research Center poll said they had little to no trust in scientists to act in the public’s best interests, up from 13% in January 2019.

This article about the large jump in measles cases in 2023 due to inadequate vaccine coverage in LMICs.

Measles cases rose 20% last year, driven by a lack of vaccine coverage in the world's poorest countries and those riddled with conflict, the World Health Organization and the U.S. Centers for Disease Control and Prevention (CDC) said on Thursday.

Nearly half of all the large and disruptive outbreaks occurred in the African region where the number of deaths increased by 37%, they said.

"At this moment, every single country in the world has access to measles vaccine, so there's no reason why any child should be infected with the disease and no child should die from measles," WHO's Natasha Crowcroft, a senior technical adviser on Measles and Rubella, told reporters.

This article about the resurgence of Victorian-era diseases in the UK due to declining vaccination rates, poor diets and cuts to public health budgets.

This weekend the British Association of Dermatologists issued an alert about an “unusually high” rate of scabies — a parasitic skin infection typically associated with Victorian workhouses. Just days earlier, doctors had warned in the BMJ of a resurgence in people developing scurvy due to not eating enough fruit and vegetables. Meanwhile, cases of the sexually transmitted disease syphilis, which was rife in the 18th and 19th centuries, are the highest they’ve been since 1948.

What I’ve been thinking about

Recently, I’ve been struck by how fragile progress can feel, especially in the context of public health. As I’ve read about the resurgence of diseases like measles and even scurvy—yes, scurvy!—it’s hard not to wonder what it says about the systems we’ve built and what happens when they’re left to erode.

This feels particularly relevant now, as conversations about trust in science seem louder and more contentious than ever. Recent data highlights the steady decline in public trust in scientists, with only 73% of Americans expressing confidence in 2023, down from 84% just a few years ago. Among some political groups, the erosion is even sharper, turning this into a cultural and political issue as much as a scientific one.

It’s hard to ignore the parallels: in the U.S., skepticism toward scientific expertise is shaping everything from vaccine uptake to health policy, while in the UK, cuts to public health budgets are allowing conditions that once defined the Victorian era to creep back into modern life.

These stories feel especially poignant as we enter a season where community health is top of mind. The holidays bring families together, but they also remind us of the shared responsibility we have to protect each other.

It makes me think about how we balance individual freedoms with collective good, how we build systems that prioritize prevention over crisis, and how we sustain trust in expertise when it feels like that trust is fraying. Maybe it’s not just about fixing what’s broken, but also about holding on tightly to the progress we’ve already made.

When we recognize these complexities, it becomes clear that there’s rarely a one-size-fits-all solution in health or development. Effective interventions require a willingness to adapt, to look beyond the immediate outcomes, and to weigh the broader impacts that may emerge over time. This month, I’ve been reflecting on the importance of this approach—of seeing interventions not as isolated acts, but as pieces of a much larger and often unpredictable system.

What I’ve written

This month, I explored the unforeseen consequences of a nutritional supplementation program in Works in Progress. In 2003, an iron and folic acid supplement randomized control trial in Zanzibar was shut down abruptly when researchers found the treatment increased mortality. Children who received an iron dose were 12% more likely to die from severe illness relative to the control.

Researchers theorize that, in some settings, iron supplementation may inadvertently increase vulnerability to malaria by providing the parasite with a nutrient it needs to thrive. This can result in the intervention having a net-negative impact on recipients’ health and wellbeing, in spite of its benefits in reducing anemia.

The group of parasites that cause malaria are dependent on iron to proliferate. As a result, some evidence suggests that being iron deficient or anemic offers protection against malaria infection. In a 2012 study, researchers followed a cohort of 785 Tanzanian children over three years and found that those who became naturally iron deficient were 60 percent less likely to die than those who did not.

Keep reading: Anemia and Malaria

Subscribe now

What I’ve been reading

This paper linking declines in bat populations to a 7.9% increase in infant mortality in impacted countries (through the mechanism of increased insecticide use by farmers).

I use the sudden emergence of a deadly wildlife disease in insect-eating bats—known as white-nose syndrome—to quantify the benefits from their provision of biological pest control. I validate previous theoretical predictions that farmers respond by substituting bats with insecticides; however, because those are toxic compounds, by design, this substitution leads to higher human infant mortality rates in the areas affected by the bat die-offs.

This paper about how economic development, and related urbanization, may increase the risk of zoonotic diseases, or infections that spread between animals and humans.

Three interrelated world trends may be exacerbating emerging zoonotic risks: income growth, urbanization, and globalization. Income growth is associated with rising animal protein consumption in developing countries, which increases the conversion of wild lands to livestock production, and hence the probability of zoonotic emergence. Urbanization implies the greater concentration and connectedness of people, which increases the speed at which new infections are spread. Globalization—the closer integration of the world economy—has facilitated pathogen spread among countries through the growth of trade and travel.

This paper about the economic — rather than educational — returns to universal pre-kindergarten, suggesting that the program may not be particularly helpful for learning outcomes, but that its benefits to parents’ earnings may mean it pays for itself.

Using a randomized lottery design, we estimate the effects of enrolling in a full-day UPK program in New Haven, Connecticut on parents' labor market outcomes as well as educational expenditures and children's academic performance… Enrollment has limited impacts on children's academic outcomes between kindergarten and 8th grade… In contrast, parents work more hours, and their earnings increase by 21.7%. Parents' earnings gains persist for at least six years after the end of pre-kindergarten. Excluding impacts on children, each dollar of net government expenditure yields $5.51 in after-tax benefits for families, almost entirely from parents' earnings gains.

What I’ve been thinking about

Like many in the United States—and elsewhere—the ongoing presidential race has been on my mind. Specifically, I’ve been intrigued by how prediction markets are shaping discourse around the upcoming election. While polls show the U.S. presidential contest as neck and neck, as of when I’m writing this, bets on Kalshi give Trump a 62 percent chance of winning, versus 38 percent for Harris. Polymarket has the race at 65 to 35, and PredictIt shows them at 61 to 43.

This shift has led to a debate about how much to trust market moves as reflecting sentiment. Critics argue that prediction markets may not always reflect true beliefs, suggesting they’re being influenced by “whales” or motivated investors. Meanwhile, others challenge skeptics to put their money where their mouth is if they disagree — pointing out the inconsistency of claiming odds in a prediction market are inaccurate without buying the other side.

What interests me most is how these markets might both reflect and shape voter perceptions. As people see odds fluctuate, it can create a feedback loop where predictions influence expectations, and those expectations then feed back into market behavior. These dynamics in prediction markets reveal just how intricate our systems for interpreting events can be. Whether in health, development, or politics, there’s rarely a simple answer—and that’s exactly what makes these shifts worth watching.

What makes this even more interesting is how it ties into recent, broader discussions. From debates over kidney donation and compensation to gene drive technologies that could alter entire ecosystems, we’re constantly faced with decisions about what should be changed and what should remain untouched. In a way, these global ethical debates reflect the same tension we see in the science of memory—how much we control, and at what cost.

As always, feel free to reach out— I'm always eager to hear your thoughts!

What I’ve written

This month, I explored the fascinating world of memory formation, learning about how our brains manage to both preserve and modify our recollections over time. New research led by neurobiologist Flavio Donato suggests that memories are stored across different populations of neurons, each playing a distinct role in how flexible or fixed those memories become. This discovery has the potential to reshape how we think about the malleability of our past experiences—and perhaps even open up new therapeutic possibilities.

“This led us to the conclusion that not only was the activation of this subpopulation of plastic neurons indeed correlated to the plasticity of the memory, but also we could extend this plasticity if we could force this neuron to be recruited into the memory trace,” says Donato. They also found that activation of the late-born neurons shortly after the event was necessary for the long-term permanence of the memory. 

Keep reading: Our Memories are Stored in Triplicate

What I’ve been reading

This article about why creating large 'evidence banks' could help governments make faster, more effective policy decisions.

Evidence syntheses are 'everything the world knows about how to solve an important problem in one place'... But most other fields lack such an extensive foundation. It can take months or years to extract meaning from a massive body of research — and funders have historically spent peanuts on synthesizing knowledge compared with the billions they spend on new research.

This piece about why compensating kidney donation could help address the severe organ shortage in the U.S. — and ethical questions of coercion, equity, and the commodification of human organs.

You go into a hospital. You do something that is physically strenuous. You take time and effort out of your life to save someone’s life, and then you get nothing for it... It drives me particularly crazy when I hear transplant surgeons talk about how it undermines the altruism of the gift to compensate it. You’re making $200,000 a year, and you’re going to lecture me about how it undermines the altruism to get paid a few tens of thousands of dollars for saving someone’s life?

This article about gene drive technology, and how it might be used to combat malaria.

What would happen if we inserted a gene drive into a mosquito with the following properties?

a) If male, do nothing.

b) If female, make sterile.

If such a gene evolved randomly, or was gene-edited in, with the usual 50 percent chance of being inherited, it would be quickly selected against and eliminated from the gene pool. But since the gene drive makes the gene transfer to offspring at a 100 percent rate, all offspring inherit the gene. This changes the outcome completely. […]

Like a meteor hitting the ocean, such a ‘suppression drive’ would spread like a wave in all directions from the origin of its release, leaving a dwindled or completely eliminated population in its wake.

Gene drives could be used to wipe out entire species.

What I’ve been thinking about

The UN General Assembly - and, with it, climate week - came to New York this month. Living in the city in September means being in the middle of the mass of global leaders, activists, and experts that flock to midtown. This year, I found myself reflecting on the unique energy that fills the city around this time of year.

New York becomes a place where global issues feel a lot closer, and the conversations happening in conference rooms spill out into everyday life. Whether it was ambitious targets for reducing emissions or the emphasis on climate justice, there was a lot of talk about making real progress. But beyond the big headlines, people were also grappling with the tough balancing act of economic growth and climate priorities—something that hits close to home for a city constantly negotiating its own environmental footprint.

Leaving Climate Week, I wasn't just thinking about policy, but about how real and immediate climate change feels. From rising sea levels threatening neighborhoods like Staten Island to the way extreme heat seems to linger in the subway, the challenges aren’t some far-off future. The question that stuck with me is: How do we bring those high-level conversations to where they actually affect people’s lives?

Last month’s post explored the shifting landscape of happiness across generations and the health and productivity costs of extreme heat. This time, we're diving into new territory; many of the things I’ve been reading and writing about relate to how physical landscapes shape human development and well-being.

The "What I've written" section takes us to Lake Kivu, at the border between Rwanda and the Democratic Republic of Congo. It's a place where geology, energy policy, and human safety collide in ways that challenge our usual thinking about development. This theme – environments shaping health and development – also resonates through much of what I’ve read this month, including new research on air pollution in Victorian and early 20th century London and an ambitious housing plan in the Bronx.

Reach out if something in this posts sparks thoughts - I'm always curious to hear what catches your attention or gets you thinking.

What I’ve written

Straddling Rwanda and the Democratic Republic of Congo, Lake Kivu’s unique geological composition traps large amounts of carbon dioxide and methane at the lake bottom, posing a potential catastrophic threat to millions living nearby should the lake overturn and the gasses explode outward. Earlier this year, I explored the Rwandan government’s efforts to defuse this risk by extracting methane from the lake - and the ongoing battle over the right balance between safety and resource utilization in the process. With climate week approaching, complex decisions about energy and resource extraction feel especially top of mind.

Lake Kivu is a geological anomaly, a multi-layered lake whose depths are saturated with trapped carbon dioxide and methane. Only two other such lakes—Lake Nyos and Lake Monoun—share these characteristics, and both have erupted in the past 50 years, spewing a lethal cloud of gas that suffocated any humans and animals in its path. When Lake Nyos erupted in 1986, it asphyxiated nearly 2,000 people and wiped out four villages in Cameroon. Folklore in the area speaks of “the bad lake” and its evil spirits that emerged to kill in an instant. Concerningly, Lake Kivu is 50 times as long as Lake Nyos and more than twice as deep. Millions live on its shoreline.

…

However, some experts believe that current efforts to remove gas from the lake may trigger an eruption—and a local extinction event. 

“It’s a compromise of safety versus commercial exploitation in the long term,” says Katsev. “If you return the water deep in the lake, you dilute your resource zone for future years. However, if you dump it higher up,” as KivuWatt is currently doing, “the water generates a plume as it sinks downward through the density layer, causing the water to mix vertically. The risk of limnic eruption is linked to this vertical movement.”

Keep reading: This African lake may literally explode—and millions are at risk

Subscribe now

What I’ve been reading

This article hypothesizing about the potential health-related causes of economic stagnation in the US beginning in the 1970s:

Americans born after 1947 and before the mid-1960s — the first of whom
were just entering their prime working years in 1971 — did not see economic
gains comparable to those of their predecessors… They had more problems as young children, and they did worse in school in the 1960s, accounting for the educational declines of that era, such as lower test scores and higher dropout rates. Birthweights also declined in the 1980s, a sign that the post-1947 cohort was
less healthy, most of all when it comes to maternal health.

This article about the ICN’s call for a moratorium on recruitment of nurses from 55 countries with pressing health workforce shortages:

The International Council of Nurses (ICN) has called on the World Health Organization (WHO) to consider a “time-limited moratorium of active recruitment of nurses” from countries on the WHO Health Workforce Support and Safeguard List. This follows a “dramatic surge” in the recruitment of nurses from low- and middle-income countries (LMICs) by wealthy countries, according to the ICN…

Tonga and Fiji reported losing 20% to 30% of their nurses, primarily to Australia and New Zealand, at the 2024 World Health Assembly (WHA)… The Filipino Department of Health has recently allocated funds to provide nurses with health insurance, housing, and other benefits in an attempt to stem the tide of nurse migration.

This paper analyzing the health impacts of exposure to air pollution in London between 1866 and 1965:

This study provides new evidence on the impact of air pollution in London over the century from 1866-1965. To identify weeks with elevated pollution levels I use new data tracking the timing of London's famous fog events, which trapped emissions in the city. These events are compared to detailed new weekly mortality data. My results show that acute pollution exposure due to fog events accounted for at least one out of every 200 deaths in London during this century. I provide evidence that the presence of infectious diseases of the respiratory system, such as measles and tuberculosis, increased the mortality effects of pollution. As a result, success in reducing the infectious diseases burden in London in the 20th century reduced the impact of pollution exposure and shifted the distribution of pollution effects across age groups

What I’ve been thinking about

The Olympics. There's something fascinating about how they manage to capture global attention every couple of years. For a few weeks, I find myself suddenly caring deeply about sports I've barely heard of, and feeling an unexpected surge of national pride.

The Olympics are an interesting mix of athletic competition, geopolitical theater, and cultural showcase. They bring out the best in human achievement, but also create a more physical playing field to battle out underlying disagreements and tensions.

This year, I especially enjoyed the New York Times’ medal count tracker - a more illustrative mid-games snapshot of which is below. The chart offers country rankings based on different relative weight for gold, silver, and bronze medals - showing even a question as simple as ‘who is winning?’ can be subjective.

Whatever your take, there's no denying the Olympics' power to make us collectively hold our breath for the perfect dive, the record-breaking sprint, the unexpected upset. For a moment, we're all just fans, cheering for excellence wherever it comes from.

Until next time,

Deena

I’ve been considering pulling something like this together for quite a while. News cycles shift rapidly, and the pressure to have a timely news peg means that evergreen stories about important problems can feel like they have have less of a place. This is particularly salient in global health and development, where challenges can recur for decades and, as a result, become buried by other headlines. With that said, here’s a bit about what I’ve written, what I’ve been reading, and what I’ve been thinking about over the past month. Reach out if any of it sparks thoughts for you!

What I’ve written

The mid-life crisis has long had its grounding in pop-culture - and in the literature on measuring wellbeing. That pattern (known as the ‘u-shaped happiness curve’) has broken down: the young now are less happy and more unhappy than the middle-aged. I interviewed Danny Blanchflower, an originator of the u-shaped happiness curve, and wrote about the implications of the rise in youth despair for Scientific American.

“We have to focus on the people at the extremes,” Blanchflower says. “Think about those who are most susceptible to commit suicide, to have deaths of despair. These are the people who say, ‘Every day of my life is a bad mental health day.’” Between 2020 and 2022, more than half of respondents reported no bad mental health days. But 7 percent acknowledged exactly 30. The proportion of those with this response nearly doubled from 1993 to 2023. That rate has grown most quickly among the young, especially women 18 to 25 years old. “This fact alone is the most striking and scary: my estimates are that 11 percent of ... young women are in despair,” Blanchflower says.

Keep reading: Young Adulthood Is No Longer One of Life’s Happiest Times

Another major heat wave is building across the United States this week, and it’s not an anomaly. Last Sunday was the hottest day ever recorded — until Monday: the global average temperature reached 17.15°C (62.87°F), topping the prior day’s 17.09°C. This month, I wrote for Works in Progress about how heat is hostile human health, productivity, and social cohesion.

The harms of extreme heat are growing. By 2100, nearly three quarters of the population may be exposed to dangerous environmental heat for at least 20 days each year, up from 30 percent. Heat-related mortality for people over 65 increased by 85 percent between 2000–2004 and 2017–2021. Countries in the global south were hardest hit - leading to worsening global inequalities.

Keep reading: Heat Waves

Share

What I’ve been reading

This article about Gavi’s upcoming replenishment cycle, a determinant of the organization’s budget from 2026 through 2030:

At the same time, raising $9 billion is never easy, and Gavi has the misfortune of asking for funding during what experts are calling the “replenishment pileup.” A huge number of international humanitarian groups — the World Bank; the World Health Organization; the Global Fund to Fight AIDS, Tuberculosis, and Malaria; the Pandemic Fund; the climate-oriented Loss and Damage Fund — are all asking for donor funds at roughly the same time. That has fueled fears that donor fatigue and tight budgets will mean some, many, or all of these groups will fall short of their goals.

This piece covering the roll-out of the R21 malaria vaccine in Côte d’Ivoire:

The vaccine costs less than $4 a dose, making it “realistic to roll out in many tens of millions of doses from now on,” and it has high efficacy levels of around 75%-80% in young children, Professor Adrian Hill, director of the Jenner Institute at Oxford University, who led the development of the vaccine, said in an interview with BBC Radio on Monday.

This paper detailing results from a randomized trial that provided 1,000 low-income adults in the United States $1,000 per month for three years:

We find no effect of the transfer across several measures of physical health as captured by multiple well-validated survey measures and biomarkers derived from blood draws… We also find that the transfer did not improve mental health after the first year… We also find precise null effects on self-reported access to health care, physical activity, sleep, and several other measures related to preventive care and health behaviors. Our results imply that more targeted interventions may be more effective at reducing health inequality between high- and low-income individuals, at least for the population and time frame that we study.

What I’ve been thinking about

How wonderful it is to have access to 24-7 public transportation. The New York City subway system has its issues - and can turn into what amounts to a $2.90 sauna in the summer months - but that it generally gets its average daily ridership of ~3.6 million where it needs to go is no small feat. The over 12,000 subway cars and buses that underpin daily commutes are likely a major reason New York is as large and dense - and desirable a place to live for many - as it is today.

The transit system is central to the basic functioning of most New Yorkers, but over time it has become more than purely practical; it’s become a quasi third place in and of itself. Walking through Grand Central and listening to the buskers, or spotting new details in the “Life Underground” art installation at 14th St is a unifying experience for New York’s commuters.

Until next time,
Deena