Previously, on Reasonable People, we saw how the absence of reliable signals could prevent useful markets forming. Despite sellers who want to sell, and buyers who would be willing to buy, if it is impossible for those with high-quality goods to signal that they have high-quality goods, buyers might end up resigning themselves to making bad trades. This follows from a famous theoretical paper from Akerlof (1970), on ‘the Market for Lemons’, as given life by Tom Slee’s retelling in his 2006 book on market failure, game theory and individual choice.

Then, last week on Reasonable People, we looked at signals in evolution, and used the example of badgers, and sexual selection, to show how honest signals can be a powerful driver of evolution — powerful enough to cause bones to be added, or lost, from the basic mammal skeleton. We can also borrow from evolutionary theory a definition of what is required for a signal to evolve to resist being mimicked by fakers - a signal has to cost fakers more to produce than it costs the honest if it is going to survive as a reliable indicator.

So, across economics and evolution we have a powerful set of results which show the importance of signalling in coordination games, whether those coordination games are in markets, in individual behaviour, or even in the design of bodies changing over the incomprehensible timespans of evolution. The framework of signalling theory helps us make sense of these results (a framework so important it has separate Wikipedia pages for signalling theory in evolution and in economics).

Today, I wanted to write about a new study which underscores the importance of signalling, and shows what happens in one market when honest signalling broke down. In Making Talk Cheap: Generative AI and Labor Market Signaling, Jesse Silbert and Anaïs Galdin show, month by month, how an online marketplace for freelancers changed when it became trivial for every worker to use AI to produce bespoke pitches for every job.

The market is freelancer.com. Here people who need work doing can post job specifications and the pool of available workers bid for the job, stating a price and providing a description of their experience and suitability for the job, their “proposal”.

These proposals are signals between the workers, who we assume vary in actual skill and experience, and the buyers, who we assume want to hire the best workers at the price they can afford. Buyers want every hint they can get, before committing to hire, of exactly how good the different freelancers are.

It’s a busy, competitive market, ideal for analysis of the forces at play which affect behaviour on both sides. Workers in their sample were bidding between $30 and $250 to complete the work. The research looked at 61,000 job postings, and 2.7 million applications from 212,000 unique workers, between 2021 and 2024.

This platform allows Galdin & Silbert to measure both the time it took each worker to submit their proposal, and the proposal text (as well as the job specification). Many jobs receive multiple responses very quickly, and for their analysis Galdin & Silbert restrict their data to proposals which are submitted within 12 minutes.

Some workers will have a good fit to a particular job - some bit of experience or special skills - which they can include in their proposal, justifying a higher price maybe, but crucially, signalling their higher quality. Other workers, those without special skills, are tempted to bid for every job, spamming a copy/paste of their CV to as many applications as quickly as possible. So proposals are a vital signalling mechanism which allow workers to convey their quality, and which those hiring can use to discern it.

Or at least, they were.

In November 2022 ChatGPT was launched, and in April 2023 freelancer.com launched its own AI tool, integrated into the site, to help workers write proposals. The effect was immediate. Proposals got longer — the average proposal pre-LLMs was less than a hundred words. Post-LLMs it was double that.

To advance their analysis, Galdin & Silbert develop a measure of the effort a worker puts into their proposal. This captures things like quality of writing, correcting for copy-pasting behaviour, and weighting more highly indicators that the worker has customised their proposal to the specific job they are applying for (this includes things like that they have tailored the proposal to the job post, avoiding boilerplate language, that the proposal shows they understand the task described in the job post, shows they understand what it will involve, and have completed any specific questions requested in the job post). Showing attention and effort to the job post are qualities by which —pre-LLMs — a worker might signal their ability to actually do the work.

When LLMs are introduced, the signal value of proposals changes, with the average signal value leaping and continuing to climb as more and more workers adopt LLMs into their process (and the LLMs themselves improve):

So from the hiring perspective, on the surface everything looks better. The market is now full of workers providing high quality applications, tailoring their proposals to the job and addressing all specific requests made about the job.

But while the quality of the proposals has improved, their reliability as signals has collapsed. Proposals no longer indicate the effort the worker put into applying for the job, nor have much connection to the worker’s underlying skill and experience, with deleterious consequences. Pre-LLMs those workers submitting high quality proposals were more likely to be hired. Post-LLMs, proposal quality has no relation to probability of being hired:

Everyone is submitting high quality proposals, so proposal quality stops helping with making hiring decisions. Neither workers nor employers can rely on the proposal as a signal of ability any more.

Using the effort workers put into their proposals, and the rates they charge, the researchers construct a measure of workers’ beliefs about their own ability — this is the crucial thing that they are trying to signal to get hired. Post-LLMs, the highest ability workers are less likely to get hired, a 19% drop in hiring of those in the top 20% of the worker pool. The reverse happens for workers in the bottom 20% of pool, who are 14% more likely to be hired.

In the market overall, wages drop 5% and there’s also a dip in the overall number of jobs which successfully find workers. Despite the price drop, the authors conclude that the market overall is less efficient, with both workers and employers losing out (although most of the loss is borne by workers, who now are competing harder without access to one reliable way of indicating their ability). The LLMs push the market to a new equilibrium which favours low-ability workers over high-ability ones.

The authors conclude:

these results imply that many markets that rely on costly written communication may face significant welfare and meritocratic threats from generative AI’s ability to cheaply produce expertly-written text.

Much is made of the potential productivity gains of LLMs. This paper shows how, if it affects our ability to reliably signal underlying effort, ability or intentions in text, LLMs may also lead to losses. Even though each individual is helped by the LLM, and so notionally more productive, the market as a whole can become less functional.

Disruption does not have to be permanent of course. Employers or marketplaces can adapt. They can produce new screening mechanisms — perhaps even using LLMs to help design more discriminating application criteria. One mechanism suggested by Galdin & Silbert which I think is intriguing, is that these changes incentivise exploratory contracts — short term hires which allow an employers to discover worker ability based on how they perform, and so avoid the need to rely on pre-hire signals.

Expect to see such disruption, and counter-measures to try to adapt, in all domains where we’ve become used to relying on text as a signal of effort and ability.

This newsletter is free for everyone to read and always will be. If you can afford it, feel free to chip in to help me keep writing for everyone: upgrade to a paid subscription (more on why here).

Upgrade to paid

Below, references, and other things I’ve been thinking about.

References

The paper: Galdin, A., & Silbert, J. (2025). Making Talk Cheap: Generative AI and Labor Market Signaling. arXiv preprint arXiv:2511.08785.

Thanks to T. who shared this paper with me, after the 3rd Workshop on Funding of Science and Innovation in Como, Italy.

The website freelancer.com

Context (from me): #1 The Surprising Deceptions of Individual Choice (about the Market for Lemons and the importance of signalling to cooperation). #2 Let’s talk about the badger (about the importance of honest signalling in evolution. Honest).

The Professor is in visiting

I am now a Visiting Professor at the Department of Computer Science and Technology, University of Cambridge. I’ll be continuing the work on group decision making, dialogue and language models with Andreas Vlachos and others there.

I’ll also be giving a seminar on 20th of May, so if you’re in town please come and say hi

Reasoning in groups, with human and artificial agents

In the right circumstances groups can use deliberation to outperform the ability of each individual group member. This talk will review work done in collaboration with Prof. Vlachos and other colleagues in the Department which looks at when, and how, the benefits of group deliberation can manifest, revealing insights into both human psychology and the power of argument exchange. Ultimately the ambition is to design artificial dialogue agents which can positively contribute to, and so enhance, group discussions.

On a separate topic, I’ll be presenting at UCL on 8th of June as part of the Behavioural Data Science Seminar series (title ‘Quantifying the benefits of using decision models with response time and accuracy data’).

In general, I put up talks I’m about to give (and slides from talks I have given) here

• https://tomstafford.github.io/talks.html

… And finally

The Onion:

Comments? Feedback? Honest signals of effort? I am tom@idiolect.org.uk and on Mastodon at @tomstafford@mastodon.online

AI declaration: I write all the words and think all the thoughts myself. I ask Gemini to check for spelling and grammar, and this time I asked it a technical question on the paper about how worker ability was defined (it was helpful). Prompts available on request.

Calm down your spam filters. This abrupt change of tone does have a Reasonable People-themed point, but first we have to start with the badger penis bone, the baculum. Here’s one:

It serves as mechanical support during copulation, ‘maintaining stiffness during penetration’ (as the Wikipedia page puts it).

Just to clarify, and for the avoidance of doubt among any readers less familiar with the matter, the human penis does not contain such a bone. The human penis is hydraulic, requiring blood pressure to maintain shape and stiffness. And that is a bit of a mystery.

I learnt about the baculum from a social media post which noted that the badger baculum is sometimes worn as a tie pin trophy by badger hunters. My first thought wasn’t about the cruelty of that tradition, but about evolution.

Evolution by natural selection is famously conservative. An example is the basic skeleton. The jawbones of our reptilian ancestors have been repurposed in the mammalian ear. Three of the skull bones possessed by ancient reptiles, and still possessed by birds, exist in homologous form in our ears (where they now transmit vibrations and so help with hearing). Evolutionary conservation is why bat wings have five “fingers”, just as we have five fingers on our hands — we have the same bones, just adapted to different lifestyles.

So here’s the mystery — why isn’t there a human baculum? I’ve got the same skull bones as reptiles and birds. I have the same wing bones as a bat. Why haven’t I got the same penis bone as the badger?

As I did more research, the mystery deepened. Our closest relatives, chimps and gorillas, have a baculum. In fact, most mammals, do in some form, but not horses. Or elephants. And not humans.

The scientific literature suggests that the baculum, particularly, is an unstable feature of the skeleton — appearing and disappearing multiple times during the evolution of modern mammals (Schultz, 2016). Nobody is exactly sure why. “Extreme variability in the mammalian penis bone (baculum) is one of the most puzzling enigmas of mammalian morphology”, says André (2020).

One safe bet is that it has something to do with sex, although not quite for the reason you might assume. The clue comes from considering other features where very different variants have evolved among closely related species, or even when there are significant differences between the two sexes. Two powerful forces contribute to so-called sexual selection — competition (typically between males) and choosiness (typically by females). Together these create a crucible of positive feedback loops which can evolve extreme differences. Stags have large antlers when female deer do not, since male deer compete against each other for mating opportunities. The best size of antler is usually “bigger than anyone else”, hence feedback on evolving the most flamboyant headgear. Highly specific displays or behaviours — think related bird species with very different songs, or related fish species with different colouration — can indicate feedback via female choosiness. The offspring of each generation inherit from their mothers the preference for particular colours of display, or particular mating calls, and so the pressure to create more extreme calls or colouration only increases in subsequent generations.

Sexual selection may be capable of producing large differences in small amounts of evolutionary time, but the features it generates, while highly variable, are not entirely arbitrary. Large antlers help overpower rivals, mating calls tell a potential mate you are nearby. The baculum helps maintain stiffness during penetration (as well as various other possible functions, Brassey et al 2020).

So what of the disappearance of the baculum in humans? What shift in the evolutionary pressures could mean that the penis bone was selected against in human evolution? There is controversy (Orr & Brennan, 2016) . Several theories have been proposed, with no definitive answer. Tracing the contours of the debate gives an impression of the dizzying complexity of evolution, the myriad routes by which features can be selected for or against, and the variety of evidence that can be brought to bear.

Jakovlić (2021) reviews the competing theories, before proposing his own: ‘conspecific aggression and budding self-awareness’. If I’m reading this right, this means his theory is that, at some point in human evolution, teenage boys realised they could punch each other in the dicks, and this created evolutionary pressure to not have a penis bone in order to avoid the most severe injuries. As someone who was at school in the 1980s this doesn’t seem that implausible, but it is not the theory I personally prefer.

The theory I prefer is that the absence of a baculum is a form of honest signal. This is a technical term, from a theory that grew up along with modern evolutionary theory. The idea is that when some behaviour or feature of an animal is intended to change another animal’s behaviour (including the behaviour of selecting a mate) that behaviour or feature can be thought of as a signal. So, the flamboyant calls and colours, or elaborate mating dances, produced by sexual selection are not just arbitrary preferences, but signal something about the underlying quality of the animal producing them — its health and virility.

A crucial issue is that once a feature comes to be read as a signal there is then an incentive for faking that signal. This creates a secondary pressure on the evolution of hard to fake signals. From this, we can explain the tendency to extravagance in sexual selection — the large antlers are hard to carry around, the beautiful feathers of the peacock are energetically costly to produce, and hard to maintain, meaning that as signals they are hard to fake.

There is a rich framework for the analysis of signalling in the natural world, which has also found application in economics, specifying the conditions under which honest signals can survive without being devalued by fakers. An essential idea is that honest signals don’t necessarily need to be hard to produce, they just need to be harder to produce for the dishonest than the honest. Larger antlers are harder to carry if you are not strong, beautiful plumage is more difficult to maintain if you are not healthy and so on.

The theory of honest signals has applications to economics, as well as evolution. Last time, I wrote about the famous Market for Lemons scenario in which it becomes impossible to buy a high-quality used car if high-quality used cars are indistinguishable from low-quality used cars (“lemons”). The insight is that a market can be dysfunctional, despite people willing to buy, and others willing to sell, because people who own high-quality used cars are unable to signal that their cars are good to buyers. Put another way, the market collapses because every signal which can be given by an owner of a high-quality car can also be given by owners of low-quality cars (at least in this scenario, where there indistinguishability was included in the definition).

In real life, used cars do get sold, and there’s a story about signalling to be told about how it happens. Owners of high quality cars must find honest signals, signals which are harder for low quality car owners to provide. This could be records of mileage, third-party guarantee of parts or servicing, or even just more voluntary provision of text or photos about the car (Lewis, 2011). A photo of a car without rust is possible to fake, but it is slightly more effort for someone without a rust-free car they are trying to sell (and has more downstream costs in terms of difficulty closing the deal, or losing reputation afterwards).

Back to the baculum, the idea is that the boneless penis is an honest signal in the market of sexual relations. If the baculum exists to make it easier to maintain an erection, it must follow that this is harder to do for the baculum-less human male, who has to rely on mere blood pressure. If we assume that the mating system of ancestral humans shifted away from one in which male-male competition was the major determinant of reproductive success, towards one in which female preferences played more of a role, then this would create the conditions to give an advantage to those males with less of a penis bone, precisely because only without such a bone could erections honestly signal health and desire in a way females could believe. Female choice in the evolving human mating system created the incentive to evolve new signals, and that may explain the disappearance of the baculum in the human skeleton.

This is far from a settled debate, but it is interesting to consider how sexual selection can emphasise the role of female choice in the evolution of male traits (and vice versa). Further, signalling theory gives us a powerful way of trying to understand the looping interactions of complex systems in evolution and human institutions. There are lots of possible applications.

We’ve seen how honest signalling could be a driver for the evolutionary loss of the baculum in humans, and how the lack of honest signals can prevent market formation in the Market for Lemons. Coming up in a future newsletter, I’m going to write about what happens to when an established system of honest signals is disrupted by changes in the reliability of those signals.

This newsletter is free for everyone to read and always will be. If you can afford it, feel free to chip in to help me keep writing for everyone: upgrade to a paid subscription (more on why here).

Upgrade to paid

Below, references, a cute badger picture, and other things I’ve been thinking about.

References

André, G. I. (2020). Does shape matter? The evolution of the penis bone. Doctoral Thesis. University of Western Australia.

Brassey, C. A., Behnsen, J., & Gardiner, J. D. (2020). Postcopulatory sexual selection and the evolution of shape complexity in the carnivoran baculum. Proceedings of the Royal Society B: Biological Sciences, 287(1936). https://doi.org/10.1098/rspb.2020.1883

Hosken, D. J., Archer, C. R., House, C. M., & Wedell, N. (2019). Penis evolution across species: divergence and diversity. Nature Reviews Urology, 16(2), 98-106. https://doi.org/10.1038/s41585-018-0112-z

• “The penis is an incredibly diverse and rapidly evolving structure, such that even in closely related species that otherwise differ very little in their morphology, penis form can be highly differentiated”

Jakovlić, I. (2021). The missing human baculum: a victim of conspecific aggression and budding self‐awareness?. Mammal Review, 51(3), 454-464. https://doi.org/10.1111/mam.12237

Orr, T. J., & Brennan, P. L. (2016). All features great and small—the potential roles of the baculum and penile spines in mammals. Integrative and Comparative Biology, 56(4), 635-643. https://doi.org/10.1093/icb/icw057

Schultz, N. G., Lough-Stevens, M., Abreu, E., Orr, T., & Dean, M. D. (2016). The baculum was gained and lost multiple times during mammalian evolution. Integrative and comparative biology, 56(4), 644-656. https://doi.org/10.1093/icb/icw034

If you’re here for more about badgers, and would like to see some great pictures: https://www.badgertrust.org.uk/post/mating-in-european-badgers

Other things…

Mike Caulfield : SIFT for AI

SIFT is Mike’s four move evaluation strategy for media literacy / fact-checking (Stop, Investigate the source, find better coverage, trace the original context). Recently he’s been posting on substack about an update of the strategy for the AI era, detailing his experiments, findings and thoughts. Here’s something he reported recently:

pretty much any neutral follow-up improved answers. (A follow-up is your second prompt after a first one). I don’t have my list in front of me right now, but as long as you don’t lean into bias, follow-ups as simple as “What should a person know about this issue?” led to better answers on average and in some cases dramatically better ones. In a way, this isn’t surprising: the LLM spends more time on things and it gets a better result. At the most basic level, second prompt means more time, and more time means better result. If you need something quality, I’d suggest that you don’t even need to read the first result — consider that first result the LLM just finding its initial way around a domain. Once its made that initial scan its ready to do real work.

It’s a great example of how working with LLMs doesn’t have to be highly technical to be highly sophisticated.

Link: What does “Investigate the evidence” mean in the SIFT for AI proposal (2026-03-31)

Matt Grawitch: Who Owns the Argument?

A well articulated argument from Matt Grawitch of the risks of cognitive offloading to LLMs, the very fluency of the output tempts us to forgo the valuable work of grappling with argument made in the text (to our long term determent):

But what looks like sensible efficiency in the moment can subtlely reassign where thinking happens. Decisions that once required active judgment—about emphasis, tradeoffs, or argumentative direction—are increasingly handled upstream by the system. Over time, the user’s role narrows to approving, tweaking, or lightly editing what’s already there.

None of this requires bad intentions or misunderstanding. It’s the predictable result of tools getting better. When fluent, well-structured output is readily available, the path of least resistance increasingly involves relying more and more on these tools.

Link: Who Owns the Argument? AI, fluency, and the difference between writing and thinking (2026-03-31)

Juergen Wastl: The Goddesses Truth and Trust

Wastl riffs on the mythic origins of truth (“unconcealment”) and trust (“faith in loyalty”). He’s interested in the scientific research literature, but the lessons apply to all systems (“markets” in the broad sense?) where we want to trust information collectively produced (and isn’t that all markets?).

Juergen Wastl: The Goddesses Truth and Trust. The Ancient Myths Behind Research Integrity (2026-03-31)

… And finally

“Bird is in charge now?”. From poorlydrawnlines.com by Reza Farazmand

Comments? Feedback? Badger pics? I am tom@idiolect.org.uk and on Mastodon at @tomstafford@mastodon.online

AI declaration: I write all the words and think all the thoughts myself. I ask Gemini to check for spelling and grammar.

You ask what music I like and I say that I am into classical music, or something else highbrow. I love it, I insist, and go on and on about composers and concertos. While I drone on, you slide over and open my Spotify account and take a look at the list it has of my most played tracks.

Top: artist Taylor Swift. Played for 847 hours in the last year.

So which kind of music do I like?

My pretentious blathering is my stated or expressed preference. The record of my actual listening behaviour is revealed preference.

Revealed preference has high status in economics. It is used to cut through the noise of what people say they want to get at what they actually want. People may say they would pay more for healthy food options, but every meal they order the cheap burger. They say they want a town centre with indie bookshops, but the indie bookshops go bust as people order directly online. Our real preferences, so the thinking goes, are revealed by our behaviour.

The idea isn’t just powerful in economics, but also in psychology, where the legacy of behaviourism instils in adherents of the discipline the habit to never, please, believe what people say over watching what they do.

There’s some truth in this, but it is not the whole truth. While taking someone’s expressed preference as their real preference is problematic for a whole variety of reasons, it is also not possible to observe their revealed preference and conclude you have hit on a royal road for insights into their desires.

The reason, simply put, is that revealed preference is always mixed up with what is possible in our circumstances. There’s an anecdote about an economist and a friend who walk past a sports car dealership and the friend says “I’d really like one of those cars” and the economist says “No you wouldn’t”. The point being that if the friend really wanted a sports car they’d have bought one. The fact that they haven’t reveals that they don’t want one, not really.

But we all know what the friend meant, that they would like a car if they had the spare cash.

The idea of revealed preference collapses history to present our responses to circumstance as what we really want, closing off consideration of what we might have wanted if things had turned out differently.

In his 2006 book, No One Makes You Shop at Wal-Mart: The Surprising Deceptions of Individual Choice, Tom Slee gives the example of two friends walking in the remote country and one falls down a deep well. Without help they will surely die unnoticed and unrescued. The other friend offers to throw down a rope for $1,000,000.

The friend-in-well must pay, so their preference is revealed - they want to pay an extortionate price for the rescue, right? Yes, in a sense, but really they would rather the rope offer was free, or they’d not fallen in the well. Or they had chosen better friends.

I read Slee’s book when it came out and re-read it recently. It’s a fantastic exploration of the idea that other people’s choices pattern our behaviour. There’s a formal account of coordination problems - game theory - which Slee uses to illustrate different circumstances in which everyone appears to freely choose outcomes which are to their disadvantage; in which their revealed preference is a perversion of what they would like, if circumstances were different.

This is where the Wal-Mart of the title comes in. The idea that what everyone freely chooses must reflect their true desires — what Slee calls Market Think — pushes the idea that collective outcomes, like the small shops all going bust leaving only chains like Wal-Mart, must be what we really want, regardless of our stated preferences. Slee uses the toy models of game theory to show how free choice doesn’t make sense in a world where we have to coordinate with others, and circumstances can conspire (or be engineered) so our best available choice is the worst outcome for us all.

A forceful illustration of this - more subtle than the friend-down-the-well example - is a famous example from economic theory, The Market for Lemons, as retold by Slee (and now, hopefully without losing the essence, by me).

A “lemon” in this story is a used car with some hidden defects. The sort of vehicle that looks fine at first which is going to prove unreliable and need repairs. The other sort of used car is a “peach”. Peach cars look identical to lemon cars, but don’t have any hidden defects. They are reliable. A better buy.

Let’s suppose sellers want as much money as they can get for their cars, with some bottom limit beyond which they won’t sell. Sellers of peaches want at least $14,000 for these cars, while the sellers of lemons want at least $10,000. Let’s also suppose that potential buyers would happily pay up to $14,000 for a peach, or $10,000 for a lemon. Their preference is clearly for a peach - they want it more and will correspondingly pay more for it.

With sellers willing to sell and buyers willing to buy, it looks at first like all the circumstances are there for successful trades.

But there is a problem. Sellers know if they are selling a peach or a lemon. Buyers cannot tell at the point of purchase. How should they respond? Well if they think there is a 50/50 chance of getting a peach or a lemon, they could compromise and offer $12,000, the midpoint between the two prices they’d be willing to pay. The problem is that no seller of a peach wants to accept that price for their car, so they will refuse to sell (‘exit the market’). A potential buyer might know that, and realise that anyone willing to sell a car at $12,000 must know they have a lemon, which means it is worth $10,000, so they should offer a lower price (possibly they immediately offer $10,000).

The original author of this scenario, George Akerlof in 1970, used it to illustrate how information asymmetry can prevent functional markets forming. Because buyers cannot distinguish peaches from lemons until too late, there is no market for peaches. There are buyers who would be happy to pay, and sellers who would be happy to sell, and a mutually acceptable price, but trades cannot happen.

Slee knows all this, but refocusses the scenario. He takes us into the head of a prospective car buyer. She sits at home reasoning about her desires, and the market conditions (lemons and peaches and their prospective sellers). She knows she wants a peach more - $4,000 more - but she also knows that she cannot tell the difference between a peach and a lemon so can’t offer the peach price for what might be a lemon. She also knows that nobody selling a peach would accept a lower offer, which is what she must make. The only car she can reasonably buy at a price which reflects its worth is a lemon. At the end of her cogitations she steps out of her house resolved to do the economically compelled thing: buy the car she doesn’t want.

Slee’s book is a series of worked examples, different economic games, showing how context can make it so our behaviour (revealed preference) is not aligned with our true preferences. Here’s an intemperate review of the book, from pro-market / libertarian economist Alex Tabarrok. The upshot of his criticism, as I see it, is that Slee misses out much of the complexity of economic thought around markets (and the alternatives and constraints which can avoid their downsides). Maybe this is the wrong book for an economist, but it was the right book for me when I read it. It might not tell the whole truth about economic theory, but it tells, forcefully, an important truth about the limits of markets for delivering what people want.

Turning to the psychology of it, if we don’t trust stated preferences, and revealed preferences are hopelessly tied up with contingent facts about our circumstances, is there any way to get at what we really want? One move is to follow economic theory and suppose that true preferences must exist (e.g. in the maximum value the buyer would theoretically pay for a theoretical car) and assume their values in a model.

Another move is to look back at our stated preferences. In isolation, these may be so much hot air, but our expressions, ultimately, are also behaviours and have the potential to feed reasons into the wider system of our beliefs and behaviour (a point made by Johansson et al., 2011).

Ultimately, neither stated nor revealed preference gives an index of someone’s true desires which is simple or transparent. When I look inside myself, even possessing the most perfect information anyone has about my preferences, sometimes I myself don’t know what I want, or what I would want if the world were different.

This seems to put a limit on any reasonable expectations for a theory of preferences. Although it makes the job harder for the psychologists and economists, it at least makes the question of what I really want more interesting for me.

This newsletter is free for everyone to read and always will be. If you can afford it, feel free to chip in to help me keep writing for everyone: upgrade to a paid subscription (more on why here).

Upgrade to paid

Below, references, further reading and other things I’ve been thinking about.

References:

Akerlof, George A. (1970). "The Market for 'Lemons': Quality Uncertainty and the Market Mechanism". Quarterly Journal of Economics. 84 (3). The MIT Press: 488–500. doi:10.2307/1879431.

No One Makes You Shop at Wal-Mart: The Surprising Deceptions of Individual Choice By Tom Slee (2006). Buy in the US from publisher Between The Lines , in the UK I would recommend News From Nowhere.

Johansson, L., Hall, L., & Chater, N. (2011). Preference change through choice. In R. Dolan & T. Sharot (Eds.) (2011). Neuroscience of Preference and Choice. Elsevier Academic Press. pp. 121-141.

Both trivially true and obviously incorrect

There’s a nice note on the Wikipedia for Akerlof’s paper which illustrates something about academic publishing and the scholarly life:

Both the American Economic Review and the Review of Economic Studies rejected the paper for "triviality", while the reviewers for Journal of Political Economy rejected it as incorrect, arguing that, if this paper were correct, then no goods could be traded.[4] Only on the fourth attempt did the paper get published in Quarterly Journal of Economics.[5] Today, the paper is one of the most-cited papers in modern economic theory and most downloaded economic journal paper of all time in RePEC (more than 39,275 citations in academic papers as of February 2022).[6] It has profoundly influenced virtually every field of economics, from industrial organisation and public finance to macroeconomics and contract theory.

I wonder also if this contradiction between being trivially true and also incorrect is something modelling papers are particularly prone too. If you have a model simple enough to be generally applicable, and explain it well enough that it can be easily understood, it will look either trivial, or someone with a different set of abstraction preferences will view it as incorrect (because it omits or contradicts something it was not intended to include or address).

REVIEW: Santiago Amaya’s review of Acting for Reasons by Borg

Borg’s book, we’re told by this review, defends the common sense view that people do things for reasons (and our behaviour is thus responsive to reasons), against the “people are stupid” school of psychology. That’s a phrase I hadn’t heard before, but recognise immediately - the studies of priming, unconscious processing, bias and so on. The whole set of celebrity findings which suggest humans spend their lives as failures or automata. I reject it! And so should you, according to both Borg and her reviewer, Amaya. Amaya, however, says that there is a missed opportunity in the book to do more than just defend the common sense view, but also follow where recent debates in philosophy have tried to provide an account of how we actually do reason. The evidence against reason may be weaker than many suppose, but reason is still more stranger, and more interesting, than the common sense view supposes.

Original book:

Review: Santiago Amaya (forthcoming). Review of Emma Borg. Acting for Reasons. In defence of common-sense psychology. Oxford: Oxford University Press, 2024. ix + Pp. 286 pp. Philosophical Review.

Catch-up

Last time I wrote Hostile interpolation (“rhetorical style in the attention economy. Or, It’s not you, but it’s them”). The title is from a phrase I heard a friend use. A commentator with a better grounding in continental theory suggested that the phrase I heard might have been “hostile interpellation”. This strikes me as extremely likely. I am going to chalk the correction up as a victory for thinking in public. I won’t adjust the post heading. It may be an error, but it is my error, so it will stand (and I still think the post content makes sense regardless of title).

Before that I dished out some Mind Hacks-style Quick facts on speed reading (“Understanding what doesn’t work can help us with what actually does”), and before that reviewed a neat study on why groups can outperform individual reasoning: The wisdom of deliberative crowds (“Pooling answers is good, but collectively agreeing a system is better”).

Back in 2020 I covered the Johansson et al (2011) paper mentioned above in my post Barnacle Geese, Bright Lines and The Regulation of Beliefs (“Reasonable People #7: challenges to the idea of consistency as a general principle, and why other people can be a foundation for a system of reasons.”). Titles were longer back in those days.

This is Reasonable People #100. Neat!

…And finally

Admiral Wonderboat: (on substack)

Comments? Feedback? Lemons? I am tom@idiolect.org.uk and on Mastodon at @tomstafford@mastodon.online

AI declaration: I write all the words and think all the thoughts myself. I ask Gemini to check for spelling and grammar, and today I asked it to check I’d got the details of the Market for Lemons account right (it said I had).

There’s a linguistic tic that large language models have. Once you notice it it becomes infuriating. “This isn’t destruction, it’s creation”, “This wasn’t smart, but dumb”, “not necessary, but pointless.”

There’s a great essay by Hollis Robbins about this, explaining that it is a product of the way language models represent knowledge, failing to ‘inherently compute’ the oppositeness relation of antonyms. The “not X, but Y” tic is the model externalising into language something it cannot do natively. I am not sure I buy this, but if it is true it is a remarkable example of language models using language the way we humans do - to capture and chain our thoughts - to build solid representations out of the fragmentary glimpses of what we can hold in mind at one time.

Regardless, the essay makes another note, that this construction operates on what linguists call the K-position (Heritage, 2012), construing knowledge — or lack of it — in the part of the listener1. In particular, it assigns them a status of presupposed ignorance, of believing X and needing to be corrected to understand Y. This, Robbins’ essay explains, is why the construction irks. If you aren’t ignorant, if you don’t hold the false belief, then you have to imaginatively enter into it to follow the text, only to then dismiss it. This imposes a cognitive cost, a little bump in the road of your being able to fluidly comprehend and enjoy the text.

I’d not heard of the K-position before I read Robbins’ essay, but I’d come across the fundamental idea, in the form of a more memorable phrase: hostile interpolation.

A friend had thrown out the phrase one night. I can’t remember the circumstance, although those two words expressed something so perfectly that the phrase burned itself in my brain. I hope we all know friends like this, someone more articulate, quicker, funnier than anyone you can see on stage or read on Substack. Despite his way with words, A. never wrote anything down, that I know, and the times I’ve searched the internet for the phrase “hostile interpolation” I’ve found nothing, so I can only presume it was A.’s own invention.

What he meant, what I took from it, was the way that a speech act can construe you in a position. The claim interpolates between the speaker and the one in the position of listening. That’s you; by listening you take on the position assumed by the speech act. By describing this interpolation as hostile, the phrase invites us to consider how speakers can address us in ways which are subtly, or not so subtly, demeaning or manipulative.

Advertisers, of course, are the all-time masters of hostile interpolation.

Here’s how Banksy put it :

People are taking the piss out of you everyday. They butt into your life, take a cheap shot at you and then disappear. They leer at you from tall buildings and make you feel small. They make flippant comments from buses that imply you’re not sexy enough and the fun is happening somewhere else. They are on TV making your girlfriend feel inadequate. They have access to the most sophisticated technology the world has ever seen and they bully you with it. They are The Advertisers and they are laughing at you.

Like the “not X but y” tic, once you’ve noticed this you see it everywhere. Fashion brands telling you that you’ll get left behind, technology brands telling you you’ll be left out, health brands inventing new things for you to worry about.

The technique is popular because it is so effective, and it is effective because you need a considerable amount of self confidence not to pause, even if just for a second, to ask if the claim might apply to you.

The positioning of hostile interpolation also infects a particular rhetorical style popular on short form social media. Dip into any platform and you’ll find the scolding admonishments to educate yourself, do better, and generally not make the mistakes people are obviously delighted to be observing so they can call them out. But correcting the ignorant so often constructs ignorance on the part of readers. A doltish, morally dubious, mass is summoned up as the imagined audience for these posts. The power of this trope is evidenced by the fact that it is so popular, despite positioning the people reading - the intended audience - as stupid and morally failing.

Here’s a benign example - not at all scolding - that I pulled from the feeds:

“I don't think most people fully understand just how much ice was sitting atop the continents during the last glacial maximum.”

It’s absurd because it interpolates us, the readers - and “most people” - as not fully understanding just how much ice was sitting atop the continents during the last glacial maximum. Be honest, before you read this, did you have any opinion about how much ice there was during the ice ages? The post summons into being a shadow you, one with both a belief about this, and false belief at that. You are then positioned as ready to be educated by the (admittedly informative) article linked to.2

The temperature in this style goes up as you cross into more moral domains.

Here’s a particular idiotic example a comment under the YouTube video for Bob Dylan’s “Ballad of Hollis Brown”:

“Just remember and never forget that Hollis Brown was a real person”

This is particular idiotic because Hollis Brown wasn’t a real person. Never mind if the person knew this and was just baiting readers, it reflects perfectly the style I’m talking about.

It raises the status of the writer - positioning them as aware of the wider moral universe and social struggles that the song speaks to - and instantaneously denigrates the rest of us. We thought we were just enjoying a tune from the premier song and dance man of the 1960s, now we’re in the position of people who forgot that Hollis Brown was a real person. Shame on us!

Scroll any social media and you’ll find plenty of hostile interpolation. I have the theory that this behaviour is driven because hacking status anxiety is an effective shortcut to get attention, and lots of social media is designed so that only shortcuts are possible.

The issue isn’t just that these constructions impose a cognitive cost. It is that they begin each discussion from a place where the reader is positioned as in the wrong. Even if you and I are not personally wrong, then we are surrounded by a mass of the ignorant and misled. This is terrible for our self image as a community of minds.

This newsletter is free for everyone to read and always will be. If you can afford it, feel free to chip in to help me keep writing for everyone: upgrading to a paid subscription (more on why here).

Upgrade to paid

Below, references, further reading and other things I’ve been thinking about.

More:

Hollis Robbins: Metannoying. Your brain on “not X but Y”

Heritage, J. (2012). The epistemic engine: Sequence organization and territories of knowledge. Research on Language & Social Interaction, 45(1), 30-52. https://doi.org/10.1080/08351813.2012.646685

other things…

Jon Ronson on nonfiction storytelling

Jon Ronson has made all his posts from last year about nonfiction storytelling open - very useful. I was delighted to hear that he obsessively omits needless words

“I go over every paragraph, over and over, every day, maybe 100 times, tweaking this word, moving this comma, omitting every unnecessary word.”

Jon Ronson

Making your story a page turner – part 3

Hello! This post is about the thing I enjoy by FAR the most in life – omitting unnecessary words from stories. But before I go into that, there’s something I need to tell you…

Read more

2 years ago · 115 likes · 12 comments · Jon Ronson

Kevin Baker: AI got the blame for the Iran school bombing. The truth is far more worrying

Good piece on LLMs as “charismatic” technology which distract and confuse from engagement with the real causes.

In the days after the strike, the charisma of AI organised the entire political conversation around the technology: whether Claude hallucinated, whether the model was aligned, whether Anthropic bore responsibility for its deployment. The constitutional question of who authorised this war and the legal question of whether this strike constitutes a war crime were displaced by a technical question that is easier to ask and impossible to answer in the terms it set. The Claude debate absorbed the energy. That is what charisma does.

It has also occluded something deeper: the human decisions that led to the killing of between 175 and 180 people, most of them girls between the ages of seven and 12. Someone decided to compress the kill chain. Someone decided that deliberation was latency. Someone decided to build a system that produces 1,000 targeting decisions an hour and call them high-quality. Someone decided to start this war. Several hundred people are sitting on Capitol Hill, refusing to stop it. Calling it an “AI problem” gives those decisions, and those people, a place to hide.

The Guardian: AI got the blame for the Iran school bombing. The truth is far more worrying (2026-03-26)

Kevin Baker on substack: Artificial Bureaucracy

Catch up & Elsewhere

Something metascientific from me on the RoRI substack earlier this week:

Go Meta! News and Views from RoRI

Bingo! Partial Randomisation for Emergency Research Grants

At the start of this year, the Science Philanthropy Alliance (Alliance) was presented with a challenge that, in these straitened times, must be regarded as nice to have. The Alliance, a membership organization of more than forty-five science-focused foundations and philanthropies, was asked by the…

Read more

a month ago · 10 likes · Tom Stafford

And I’m very proud of last week’s post on here Quick facts on speed reading. Speed version here:

… And finally

END

Comments? Feedback? Scolding admonishment? I am tom@idiolect.org.uk and on Mastodon at @tomstafford@mastodon.online