AI as Assistant

In this model you treat something written / generated by AI the way you would something written by another person. If you wouldn’t copy another person’s words w/out attribution, or take their work and pass it off as your own, then don’t do that w/ AI-generated text. It may be a stimulus or provocation or inspiration; if so, great. But don’t pass it off as your own.

Example. While I’m writing a workbook for businesses and teams, I will turn to AI when I’m stuck, or am looking for ideas. The results are almost never what I’m looking for, but the act of rejecting them gets me thinking about what could work, and puts me on a path toward an exercise or approach that I think could really work. Likewise, when I use an AI as a librarian / research assistant, the ideas may be useful, but the text itself is not – any more than I would copy down what a librarian or reference desk tell me, and publish it as my own.

Where I’m at. This is my current level of AI use in my writing: I have several Gemini gems that act as research assistants, handwriting transcribers, editing tools (carrying paper edits back into documents), and a couple in Notion AI that look in my vast back catalog of blog posts and writings that got cut from earlier projects for material I can leverage. All of them are designed to work on specific tasks; and while they’re quite helpful, they don’t work together, or generate drafts on their own.

Subscribe now

Renaissance Workshop or Film Set

In his classic book Painting and Experience in Fifteenth-Century Italy, Michael Baxandall talks about how painters ran their studios, worked with assistants, and negotiated with clients. Nobody expected the painter who ran the studio, and whose works clients were commissioning, to paint 100% of the canvas (or fresco, etc.) themselves. They would sketch out the painting, supervise the production, handle specific elements themselves, and oversee a process whose final product clearly reflected the artist’s guiding hand / presiding genius. But behind them, a team of assistants would prepare the materials, assist the master, and paint the backgrounds, sky, crowds, etc.

Exactly how much was personally painted by the master was specified in their commission & contract. For example, in a portrait they might paint the husband and wife, do the hands and faces or the children, and provide the finishing touches that unified the painting; a senior assistant would do furniture and interior details; and the apprentices would take care of the sky, fields with happy peasants, town in the background, etc. The final result was still understood to be the work of the master, because the assistants had been trained in their studio, knew their methods, were initiated into the secret formulas or tools they used to get distinctive effects, etc.

A good modern equivalent would be movie directors. You immediately recognize a Christopher Nolan or John Woo movie, even though they’re not personally pulling focus, adjusting spotlights, doing makeup, or handling special effects and sound design. A “Christopher Nolan movie” has a particular look because Nolan works with a tight crew: his wife Emma Thomas runs the production company, his brother Jonathan is coauthor on a number of screenplays, Hoyte van Hoytema has handled cinematography since Interstellar, Hans Zimmer & Ludwig Göransson write the scores, and of course he has actors he keeps working with (the cast of The Odyssey includes at least ten alumni, including Matt Damon, Anne Hathaway, Robert Pattinson, Himish Patel, and Elliot Page).

Likewise, for authors working with AI, you can envision a process in which they provide the overall vision and maintain creative control over projects, but let their agents take on tasks that were either highly repetitive, or that they can do faster and better.

Example. I’m writing a business book that includes 100 case studies of companies. These case studies are highly stereotyped, always covering the same topics and beats. Having an AI write those is pretty low-risk, and would not take away from the quality of the project. Readers don’t expect glittering bon mots or boundary-pushing experimentation: they are reading for utilitarian purposes, and predictability & simplicity help them read the way they want. My name still goes on the cover, though a footnote or methods section talking about AI use would be good (unless you take the position that it’s no one’s business and readers should only judge the work itself).

Where I’m at. I’m still at the AI as assistant level, but this is where I’m trying to go with one of my projects.

Author/Publisher is Guarantor

Finally, this model acknowledges a high dependence on the LLM, but a person still takes responsibility for the final result. Essentially, while the writing is outsourced, the quality control still stays in human hands. It’s more like signing a tax return prepared by software than anything.

There’s actually a LOT of legitimate stuff in the world that’s published under a model like this. In science, the head of the lab puts their name on articles written by postdocs and grad students, who in turn are trusting the results generated by a batch of instruments and computers in their lab. There’s little expectation that the lab director was involved in running the experiment (though presumably they were involved in thinking it though, coming up with the experimental approach, etc.). Making yourself an author is like a stamp of approval, or certification of authenticity: it means this work was done to your lab’s standards, using the methods or approach I developed, tools I invented, etc. Likewise, Supreme Court opinions go out with the name of the justice on top, and every justice sets the direction of an opinion, the main points to hit, other cases to cite, etc., but clerks are heavily involved in the research and drafting process.

The Encyclopedia Britannica includes a lot of unsigned articles, but that didn’t mean that nobody has written them, or nobody stands behind them. Everyone recognizes that, in this case, the Britannica itself vouches for their accuracy. (Unsigned articles, by the way, were either written in-house by editors, or had originally been written by outsiders and had bylines, but over time had been so thoroughly revised that we felt it necessary to anonymize them.) When I was managing editor in the 1990s, in fact, we were experimenting with a system that could automatically generate comparisons between geographical entities-- two cities, two countries, etc.-- that stood a chance of working because those articles were so well-structured, and the articles were so well-tagged (i.e., city names had the tag “cityname”, etc.) it wasn’t hard for a program to build something out of them. That content wouldn’t have been written by anyone-- to the extent that you could talk about them having been written at all, rather than “generated” or “extracted”-- but the Britannica still stood behind the results because it owned the process.

For a certain level of executive or politician who’s used to working with speechwriters and PR people, putting your name to things that your people write is an everyday thing, and the question of whether a human or LLM generated the text probably feels irrelevant. When the Wall Street Journal published the op-ed by Stanley Druckenmiller that was written by AI, I’m sure everyone though that was “Druckenmiller’s writing” in the same way that if a speechwriter or PR firm wrote it. You can argue about the quality of the writing, but I think an AI-written op-ed is no worse than every personal statement that starts, “[Partner’s name and] I were heartbroken to hear about the thing that happened, and offer our thoughts and prayers to the people who the thing that happened happened to.” (This is a 30 Rock reference in case you’re wondering.) Druckenmiller defends his use of AI, saying “I write everything using AI now for the same reason I use a calculator when I do math problems.”

This does not give a blank check to anyone to generate text with AI. Students putting assignments into an LLM, then handing in the result as their own, are not in a position to provide the guarantee that Britannica or a lab head does: students don’t have the expertise to verify that work, except at a very superficial level. Likewise, “authors” like “James Douglas“ who’ve published 44 books (and counting) since 2025 on everything from tai chi to dementia prevention to animal husbandry are probably not first becoming experts in these fields, and writing from a fresh and distinctive perspective. There have always been hack writers, but AI supercharges their ability to turn out work and distort the literary marketplace.

Example. If I write a book w/ an LLM ghostwriter, smart readers will assume that I did not ever put pen to paper, but the ghost has interviewed me, worked to capture my voice, written things that I stand behind, and that I am responsible for the final product. Effectively the writing process is unimportant; all that matters is the final product, and whether a human is willing to say that it. This is pretty close to the studio model, except the source of the writing is purposely obscured; like a software-prepared tax return, the premise is that the generation process is irrelevant to the claim of authorship.

Studio / Label Model

In this model, the overall creative direction comes from a leader, but the specifics are handled by others, designers and teams. It’s an even more hands-off version of the Renaissance studio. Michael Kors doesn’t claim to design everything w/ his name; it is a style, brand, signifier, guarantee, not a promise that his fingers touched a seam. Maybe an even better comparison would be Ikea. There’s a whole team of people designing everything from children’s toys to sofas to office furniture; you don’t know who they are (despite the occasional marketing campaign that highlights a few of them); but Ikea has a really distinctive look that runs across very different products, from trash cans to bookcases.

Example. I launch a series of books on rest in different professions. They follow a template that I create that specifies how the length of the book; the themes that should be included; how arguments and sections should be crafted; the consistent beats that need to be hit, etc. I then have several AI agents gather material; run surveys and code the results; write up findings; and draft the whole mss, following my specs / guidelines and template. I act like an editor or thesis advisor, checking for quality, stress-testing the argument and evidence, providing feedback for revisions, and helping keep the whole thing on track. But I don’t claim authorship; just as a Motown or Blue Note or EMI release will have a distinctive style but bear the artist’s name, so will these books be from my workshop, not from my own hand.

Where I’m at. I can imagine adopting this model for some writing projects. Every writer has unfinished projects, or topics they’d love to spend some time on, and I think using AI to finish those could be interesting, and could be a practice that authors could do publicly without apology. For example, there are several scholarly articles I’ve started work on and never finished, and I would love to be able to hand those off to a team of agents and say, “I started this, finish it for me.” Having already published four books and dozens of articles before the Chatpocalypse, I’ve proven to myself that I can write; I still have other things that I’m working on that I’m writing myself (with task-specific AI assistants in the background); handing those unfinished pieces off wouldn’t constitute cognitive surrender or hinder my ability to work on the mothers and creativity book. I don’t have to claim to be their author (though saying that they’re products of the House of Pang would help readers know what to expect); I just think it would be better if those old unfinished articles were actually completed, if those weeks I spent in the archives yielded something. And I’d like to read them myself.

I don’t think AI (at least the versions that you don’t pay several thousand dollars a month to access) is anywhere near being able to do that; but I wouldn’t be mad if, at some point, I had well-trained agents who knew my work, the arguments I’ve made, the threads that connect my books, and were able to finish things for me. Or, if I knew my my health would prevent me from finishing my last book (my dad left an unfinished manuscript on his desk when he died), being able to oversee agents completing the last couple chapters would feel better than leaving it undone.

Related articles

Rest, Work, and the Good Life

How are creative writers actually using AI?

You can’t turn refresh your social media and news feed these days without finding some new piece about AI, writing, and publishing. A couple days ago, it was the Wall Street Journal announcing that “AI Has Plunged the Book Industry into Utter Chaos;” authors have had…

Read more

9 days ago · 4 likes · Alex Soojung-Kim Pang

I think these are serious issues, and that AI is a bit like an unshielded electric wire, full of valuable power but also potentially fatal. I also think in today’s world, where authors are expected to be their own brands, marketing dynamos, social media presences, etc., you’d be smart to figure out how to use any technology that saves labor and creates support-- in money or time-- for writing, AI included. (And you’re even better-off figuring out how to build “tiny AIs” that fit in your own workflows and practices.)

Given this, I really appreciate articles that actually look at how authors are using AI, and answering the big questions about how to wisely adopt this technology. My favorite new example is “From Pen to Prompt: How Creative Writers Integrate AI into their Writing Practice” by University of Washington researchers Alicia Guo, Shreya Sathyanarayanan, Leijie Wang, Jeffrey Heer, and Amy Zhang. (I think the version of arXiv is publicly available)

Guo et al interviewed 18 creative writers working in fiction, poetry, and creative non-fiction about how they use LLMs in their writing practices, and how they think about questions of authorship and creativity. They found that “creative writers are intentional about how they incorporate AI, making many deliberate decisions about when and how to engage AI based on their core values, such as authenticity and craftsmanship.”

What motivates them? First, “AI redefines relationships with creative blocks,” helping authors think through issues and find ways forward. (Importantly this isn’t always about the AI acting as a kind of snowplow through obstacles: they sometimes generate “anti-patterns,” suggestions that you reject but help clarify what WOULD work.)

A number also report using AI as “a productivity boost and accessibility tool,” taking on low-level tasks or providing near-instant feedback or suggestions (not as good as a great human editor, but speed has its own value). When used well, it lets writers “do more of what they love while developing their craft,” reducing burdens “while maintaining focus on the parts of writing they find most fulfilling, creating a more satisfying creative practice.” Finally, “the personal process of experimentation and shaping ideas in unique ways” becomes its own locus of creativity and reward. “Writers are looking for new ways of using AI beyond a basic writing helper and enjoy creatively experimenting in a low-stakes environment.”

There were also some interesting differences across these three kinds of writers. Fiction writers working on book-length projects used AI most extensively, employing specialized tools like Novelcrafter, Sudowrite, or NovelAI to help draft scenes, fill in backstories or the biographies of minor characters, create dialogue, and edit drafts. For nonfiction writers, AI mainly appears early in the writing process (as a brainstorming partner, sounding board, or research assistant) or in revisions; but they either consciously turned off AI when putting pen to paper, or used it in very limited technical ways (e.g., “suggest 3 ways to handle this transition between sections”). Both fiction and nonfiction writers used AI tools for project management. Poets, in contrast, tended to use AI as a raw material generator or interactive “muse;” this use follows earlier practices like cut-up techniques, lipograms, blackouts that generate surprising combinations of words, metaphors, or volatile imagery.

One also mentions something important: “As someone who has had physical disabilities and the inability to work 40 hours a week for most of my adult life, suddenly I am competitive” with authors who have greater physical reserves. (This has been lurking around in the back of my mind in my own experiments with AI: it would be good to have tools that can support me in the future, and take on parts of the writing that become difficult for me, should that ever happen.)

At the same time, there are aspects of the craft that they don’t want to give up. Writers “want to preserve their enjoyment of the craft,” “maintain their authentic voice, message, and sense of humanness,” and “retain a feeling of ownership and control.” (They don’t want AI to atrophy their writing brains, as Emily Smith recently put it.)

Unresolving

Using AI in my day job atrophies my writing brain

For a long time, I’ve worked as two selves. During the day, I help build products for tech companies. I make spreadsheets and run meetings and more or less optimize along an axis of efficiency and profit. I’m an analytical person, and before I started writing, I studied engineering and business, so for a set period of time …

Read more

5 months ago · 159 likes · 47 comments · Emily J. Smith

This doesn’t lead authors to using AI in identical ways, or to even use it the same way across their writing process. One might use AI to play around with writing in a different style if you’re in a more experimental mode, then turn off AI when writing your own work to preserve your authorial voice. For another, control means struggling alone with the draft; for others, it means being in charge of the creative vision and editorial direction, and letting AI do the rest; for others, it’s about editing and smoothing out the final output, whether it’s human- or machine-written.

Share Rest, Work, and the Good Life

While you could argue that 18 people is not a big sample size, I think what Guo et al report tracks with things I’ve seen on Reddit’s AI for writers message board, and in conversations with other writers: we’re all trying to figure out how to use this technology to do less of what we don’t like to do or are forced by the market to do, and more of what we enjoy. And it raises some questions I’ll deal with in later posts.

In my AI class this summer, one of the things we discussed at length was school AI policies, what taught counts as responsible versus unacceptable AI use, and how they’re probably going to spend the rest of their years in school dogged by questions about AI policy. Eventually, this will all settle out, and I think even in schools, we’ll focus less on policing AI use than on evaluating the quality of work. Today we don’t care if an author used Scrivener or Word, or a Logitech versus Kensington keyboard (though Scrivener and Logitech are, objectively, the correct choices). In the future, questions about how an author uses AI will come up in Paris Review interviews, not hand-wringing editorials in the Times.

One reason I think the future will unfold this way is that a number of now-ordinary and completely uncontroversial technologies were once the objects of debate and resistance among teachers or students. Some of us are old enough to remember debates about whether handheld calculators would lead students to stop learning mathematics. In the late 1970s, educators started to worry that calculators would erode basic skills, change what counted as mathematical understanding, and force teachers to teach differently. Working through problems on paper let students (and teachers) see how they thought through a problem (and maybe went wrong); the muscle memory that came from doing arithmetic or plane geometry was a foundation for higher math; and even if you were a cashier or carpenter, being able to do math in your head would be valuable. But over time, calculators went from banned to required in classrooms, and teachers adjusted how they taught and what they demanded of students.

But the history of controversial classroom technologies extends back into the nineteenth century. For example, when it was first introduced in the early 1800s, the blackboard was seen as a disruptive technology in the classroom.

At West Point, where it was first used in the United States, the blackboard became central to a new style of mathematical instruction: students had to stand at the board, display their reasoning, and perform knowledge publicly, rather than simply reciting from a textbook. For young officers who had to grade roads and build fortifications, doing geometry on your feet was a valuable skill.

But not everyone liked it. Some teachers worried that the blackboard would break the connection between teacher and student, by encouraging teachers to face away from their class.

At Yale, it was the students who protested. Introduced in 1825, the blackboard inspired a short-lived Conic Sections Rebellion among students who protested having to solve problems in front of their classmates, and in 1830 half a mathematics class was expelled from the college for striking in protest over blackboard use in exams. Among science teachers, there was debate about whether drawing on blackboards was good-- because it let the professor create specialized images that focus on specific aspects of a subject, or over the course of a lecture start with something small and end with an elaborate, informative visual-- or bad-- because hand-drawn images could never be as accurate as printed maps or engravings.1 Today, the blackboard or whiteboard are just normal pieces of classroom furniture; but when they were new, they were symbols of a new regime of fully automated luxury Stalinism, or machines for broadcasting misunderstanding.

Even the humble pencil and eraser had a whiff of the morally and pedagogically suspect about it when it first appeared in the mid-19th century. Some teachers disliked the idea that students could easily erase mistakes: if students could erase and revise their work, they would become careless, produce more errors, and lose the mental discipline that came from having to think about your work before you put pencil to paper. I’m not sure how long it’s been since having an eraser on your pencil marked you as a Byronic bad boy, “mad rad and dangerous to know,” but I think it’s been a very, very long time.2

One of the things I get from these stories is that in the classroom, debates about new technologies are always multilayered, because education is multilayered: the pencil and eraser didn’t just make writing easier, it erased efforts to create thoughtful, self-disciplined students. The blackboard wasn’t just a way to efficiently communicate information; it challenged students’ skills as public performers. Likewise, debates about AI are partly about copyright and honor codes, but also debates about the kinds of skills we need to develop as students and maintain as artists or professionals, and the value and necessity of doing difficult, time-consuming things.

Eventually (unless AI becomes the “final invention” that really does destroy all jobs and leave one CEO the king of the planet) we’ll come to a public consensus in different professions about how to use AI in ways that don’t lead to cognitive surrender, or that don’t erode critical skills that we need to maintain in order to do good work or be a certain kind of professional. But we’ll also wonder what the fuss was about in the first place.

This summer I taught a class on AI in a program at Cambridge. One of my big aims in the class is to demystify AI for students, and to get them to see that in order to use it, you don’t have to code, or have a favorite developer environment, etc.; these days, what you really need is an ability to examine how you work, identify how it could be better (or imagine how it could be different), and clearly explain that to an AI.

One way to achieve that was get my students to create little apps that would be useful in their own lives. And to show them how easy it could be, I created a Google Gem to do something that I’ve wanted for a long time: recognize and convert my handwriting to text.

Why did I need a custom handwriting system?

I have notebooks from 1985. Grad school notebooks, work notebooks (I was a huge fan of Levenger for a long time, and still sometimes return to that system), notebooks for brainstorming (Moleskine A4), and personal journals (Paperage is my current favorite). Literally thousands of pages of text. Much of it is now of merely archival interest, but the last few years contain things that I think could leverage— use in projects, use to uncover patterns in my own life, recycle into articles and posts, etc.

Unless I want to re-type them by hand, though, that content is basically locked on the physical page.

There are tons of handwriting recognition tools; my problem is, my handwriting is really hard for systems trained on conventional handwriting samples to decipher. I mix cursive and regular letters, have quirky ways of writing things like ampersands, and write very small. On top of that, I write a lot with a fountain pen, which creates additional challenges for any automated system, which now also has to deal with ghosting, feathering, etc.

As a result, off-the-shelf systems treat my handwriting as the graphological version of that drunk raccoon in the Virginia liquor store.

My intuition is that being able to digitize my handwriting would be a game-changer for a few reasons.

1. I write a lot by hand: in my journal, in notebooks documenting various projects, etc. Being able to more easily move those words into digital form would make my life as a writer easier.

2. I have thousands of pages in old notebooks. If I could digitize this legacy material, there may be ideas buried years ago that would be well worth excavating and playing around with today. (For example, I have years of post-trip debriefs about what things to pack, leave behind, and do differently, and it would be really useful to be able to look across all of those when packing for my next trip.)

3. If I could more easily move ideas from page to screen, I might write and use my physical notebooks differently and better, and also use the notebook program on my Kindle. I don’t imagine my Paperwhite replacing my paper notebooks, but when I’m reading something on the Kindle, being able to write down an idea right there-- and know I can use/edit it easily-- would have its advantages.

4. I would love a way to back up some of my more personal content.

So for all those reasons, it felt like trying to create a custom handwriting recognition system was a good thing.

I’ve been thinking about doing something like this for about three years, ever since I got a Remarkable notebook one Christmas. I quite liked it, but the frustration of having the handwriting recognition be good but not quite there-- along with the fact that the pens for digital notebooks are still treated as an afterthought rather than the critical interfaces that they are-- meant that it never really worked as well as I’d have liked.

(Also the pens are absurdly expensive and too easy to lose. And with paper notebooks, it’s trivial to switch out pens, change colors, etc.; you lose the pleasure of varying your pens when you move to a stylus.)

However, at the time there was no good way for someone who isn’t a programmer, and wasn’t willing to spend a lot of time on this, to create a custom version of a handwriting recognition program.

But with the really genuinely stunning advances in AI, it felt like it was worth investigating whether I could create a Gemini Gem-- a custom tool trained on instructions and content that the user provides-- to do this.

Building the Gem

How could I figure out of this would work, and how to get started? I asked Gemini.

Q: Can you recognize handwriting? Off the shelf handwriting programs struggle to decode my writing and I would love to be able to train a gem that could decode my own handwriting.

A: Yes, I can absolutely read handwriting! In fact, modern multimodal AI models like me are usually much better at decoding messy script than traditional, rigid OCR (optical character recognition) programs. Because we understand context, syntax, and vocabulary, we can easily guess ambiguous letters based on the words around them.

To answer your second question: Yes, you can absolutely build a custom “Gem” specifically tailored to decode your unique handwriting style. While you can’t “fine-tune” the actual neural network weights of the model yourself, you can train a Gem to recognize your specific quirks using few-shot learning and custom instructions. Here is how you can set that up:...

For those who are interested in trying it themselves, or want to get a sense of how working with Gemini to create a custom Gem is like, I’ve put a transcript of building the Gem on my Web site.

After waiting three years, and wanting something that would do this for ages, it took... 3-4 hours to build a custom handwriting recognition system that can read my writing with 99%+ accuracy.

No programming, no loading up an IDE and installing new software, no calls to Github. Just a single afternoon of talk to the AI, upload a journal page, correct mistakes in the transcription, repeat.

Using AI to solve tiny problems

This points to a kind of AI use that gets little attention, but which I think is going to be an important use case for many of us: building little tools to solve idiosyncratic but meaningful problems.

Nobody else will ever use this Gem. It doesn’t solve a problem than anyone else has (but you could build your own, of course— my wife has!). But I already use it all the time. I haven’t even started scanning old notebooks yet, but I think the value it creates will be significant. I suspect I’m writing more, too: knowing that I can now easily move something from one of my many notebooks into digital form makes it more likely that I’ll write down ideas as they come to mind.

This isn’t the first time I’ve used AI to solve a small, specific problem that I have; which delivers great value to me; and which I could never get a human programmer to solve for me. For example, I have a bunch of Javascripts that connect Google Forms, spreadsheets, Google Drive, Google Docs, and Kindle, and automate things that used to take me hours to do. I don’t have the technical skill to write the scripts myself; it could take half a day or a full day to get them working; but once they were up and running, it was great.

For example, when I download a scientific or scholarly article to read, I’ll put it in a folder on Google Drive (it holds thousands of PDFs at this point); I now have a script that watches that folder, and automatically sends a copy of that article to my Kindle. It’s not a huge thing, but it lowers the barrier to my reading and taking notes on that article just enough. That makes it quite worthwhile for me.

Conversation as creation

It’s worth underlining that this is not something that required “real programming” as we usually imagine in: if you read the transcript, you’ll see that Gemini built the custom instruction set-- the thing that the Gem calls up whenever I use it-- as we went back and forth. (I found the comments on my handwriting amusing, and oddly illuminating.) It was conversation as creation.

One of my students had an even more extreme or exciting version of this. He’s a dancer, and wanted a tool that would critique his rehearsals— take 15 seconds of something he’s learning, compare it to a video of his coach performing the same move, and point out where they differ. We were sitting together, describing it to Claude, and it just started building. Half an hour later he had a Streamlit file, and was able to start playing with it. (It didn’t work right off the bat— ultimately he got some help from another teacher who’s a lot more technical than me— but it was almost spooky how Claude just ran with our conversation.)

I think that once you know this functionality exists in AI, you’re likely to start seeing pain points in your workflow, disconnects between systems you use, etc.-- and experiment with ways to use AI to fix them. We hear a lot about efforts to vibe code replacements for CRMs or accounting software or whatever, saving companies millions in licenses (and creating tons of new headaches). But I think these little, personal programs are going to be just as meaningful.

The Elysian

This sewing machine powers AI

“If everything is automated, how should I spend my leisure time…

Read more

22 days ago · 41 likes · 25 comments · Elle Griffin

Max Park’s Prompting Nowhere [here’s a PDF about the project] is one of a couple AIs he’s constructed that reimagine how AI can be designed and deployed in ways that make it more humane, give users more control over it, or create more free time. As Ellie Griffin asks,

What would AI be like if it were optimized for a beautiful life? Maybe it becomes like Socrates, asking questions instead of giving answers, encouraging us to be thinkers rather than passive consumers. Maybe it automates away our desk jobs, so we can spend less time on the computer and more time raising children. Maybe it funds increased leisure time, so we can spend more of our lives working with our hands, rather than sitting on screens. Maybe AI could create a more offline life rather than a more online one.

One of Park’s machines sits on a vintage sewing machine, and is powered by the machine’s treadle. Mounting it on a sewing machine calls attention to the different ways the same technology can either empower workers, giving them more leisure time or greater financial independence, or turn into tools for their exploitation:

At home... [the sewing machine] symbolized freedom: a hand-sewn shirt might take 14 hours, but with the machine it might take one, and create a path to independent income for many women. But at the factory, the treadle sewing machine also powered the sweatshops, extracting more and more output for menial pay. When paired with capitalism, these “labor-saving devices” didn’t shorten the workday; they raised how much women had to produce in a day for their dollar.

Park has a video about the project:

I’ve been thinking about this question of how AI could be used to construct different futures, as I just wrapped up a class on artificial intelligence in a summer program at Cambridge. My students were from all over the world, but they all shared real concerns about AI: even the more tech-savvy students, the ones best-positioned to exploit AI, are uneasy about the temptation of AI-generated work and the impact the technology will have on their educations and future.

One of the things we explored was how they could use AI to solve small but meaningful problems in their own lives: help with schoolwork, of course, but also critique public speaking, or even analyze dance routines.

I wanted them to come away with a few ideas:

1. This is not a technology whose course will just be defined by Silicon Valley, but can be determined by all of us.

2. The technical barriers to using AI have fallen very dramatically: a lot of theta projects were Gemini Gems, which they didn’t so much “build” as “co-create by talking to Gemini.”

3. AI isn’t just about massive data centers and world-shifting stuff; some of its most significant applications can be highly personal (like a diary) or temporary (like a patient checklist in an ER).

In intent these were kind of like speculative futures, except they were things the students really were able to build; I thought of them as lo-fi prototypes, not as sophisticated or theoretically subtle as Max Park’s work, but still useful for pointing out how this incredibly important technology could be used to create many different futures, not just the one that CEOs read about in science fiction.

A little while ago, I wrote about how I sometimes have AI write a first pass of a draft of something; I then look over that work, throw it out, and start over.

As many writers will tell you, it can feel easier to produce something good by writing a bad first draft and editing it, rather than struggling to create something polished the first time around.

Psychologists have some explanations for why bad drafts can yield good ideas. Building on this, we can design a strategy for using bad AI output in helping you brainstorm. This isn’t about getting better results through prompt engineering or other clever tech tricks; it’s about knowing enough about how our minds operate, and how creativity works, to design a good process.

In brainstorming, ideas are like pancakes

One useful concept is the serial order effect. For decades, psychologists have known that when you sit down to brainstorm or come up with creative ideas, your first ideas are often the most obvious, and it takes a little time to start coming up with more imaginative ideas. As you brainstorm, the ideas get slower, but better.

In other words, your brain is like pancake pan: the first pancake is usually the worst, and the ones after that are the ones you want to serve.

The serial order effect obviously works for ideas you’ve generated; but it can also work when you look at others’ ideas as well. In group brainstorming sessions, for example, it’s good practice to give everyone a chance to work on their own, then share the ideas, then do a second round where everyone builds on those ideas. She second round is often where the real insights and non-obvious solutions emerge. Reviewing ideas is easier because generating new ideas and evaluating existing ideas are different cognitive functions: they use different parts of your brain, and for many of us the second is often easier to switch on than the first.

Low-fidelity prototypes

Designers have discovered that we approach “low-fi” prototypes— unfinished-looking sketches, manuscripts printed in draft mode, demo tapes— differently than we do things that look really clean and finished: we’re more likely to focus on the underlying concepts, to offer more substantive and useful criticism, than if we do when presented with something that looks super-polished. Some creatives exploit this to help them have lots of initial ideas but not become overly attached to them.

I will purposely print out drafts in a sans serif font that looks very non-book quality, so I have an easier time tearing them apart (and putting them back together better, one hopes).

My favorite example of intelligent use of low-fi is Masa Takayama, who’s regarded as one of the world’s best sushi chefs. (Images below are screenshots from this Michelin profile.)

He literally sketches ideas for his menu on a roll of paper towels— a medium designed to be useful but completely disposable.

“I have a roll paper which is very comfortable to rip, throw away,” he explains. “You know, every morning, just open it, just start to draw.” (Just note the difference between this draw-and-dispose attitude to early ideas, and the incredible precision and attention to detail that sushi chefs bring to their food.)

(The opposite version of the low-fi creative boost, by the way, is the Highest Paid Person’s Opinion (or HiPPO) effect— the tendency to treat the CEO’s ideas in a brainstorm as the best. When I run workshops in Asia, I start by not letting the most senior people talk, because once they say something, everyone else builds on the boss’ idea.)

Opportunistic assimilation

In the 1990s, psychologist Colleen Seifert came up with the idea of “opportunistic assimilation.” In Graham Wallas’ four-stage model of creativity, when you’re working on a problem you start with preparation— gathering information, building a mental model of the problem, and then trying to solve it logically, by reasoning through it and using that mental model to generate different possible solutions.

If this doesn’t work and you reach an impasse, it’s smarter to move on to incubation— to let your subconscious work on the idea while your conscious attention is directed elsewhere— than to keep punching the same wall.

Seifert argued that successful problem-solvers engage in “opportunistic assimilation,” using the failed attempts to build a revised model of the problem. Mental models aren’t just pictures or descriptions; they’re also prediction machines. During a successful incubation period, the mind builds this alternate model, tests out possibilities, and (if you’re lucky) comes up with an answer— the moment of illumination.

There’s also a little of the Zeigarnik effect at work here. As Bluma Ziegarnik ovserved a century ago, our minds have a tendency to remember uncompleted tasks more clearly than completed ones, and to forget finished tasks quickly: the canonical example is how waiters are good at remembering diners’ orders while the chef is still preparing them, but once they’re served (and the “problem” of delivering diners their meals is solved), but not nearly as good at remembering past orders. In our case, a good result of the brainstorm is the unfinished order.

Let AI make the first pancake, then take over

So what kind of process would you design to use AI as an effective brainstorming partner? It would look like this:

1. Spend a little time defining your problem.

2. Ask your favorite AI to generate a bunch of ideas.

3. Look at the list. Treat it as a low-fi prototype, a napkin sketch. 90% of the ideas won’t be very good (after all, you’re working with a stochastic parrot that’s been trained to generate the most average answers to queries, not think outside the box). But the point is, you don’t need them to be good; you need them to point you to a solution.

4. Review each item on the list, ask yourself, “Why don’t they work? What’s wrong with them?”By doing this, you prompt your mind to start revising its mental model of the problem, which opens the way to coming up with a new solution.

5. If in the next few minutes you don’t have an answer, go fold some laundry, take a walk around the block, get a massage, or do something else that will engage enough of your brain to keep you from thinking about something else while your subconscious builds your new mental model, then approach the problem again a few minutes later.

Berkeley-based author Bonnie Tsui has an article in The Atlantic about “Why so many writers are athletes.”

For someone who is not a professional swimmer, I spend an awful lot of time under water. I consider it an act of mobile meditation: Sensory input is muffled, so the chaos of the surface world eventually recedes. What’s left is activity—familiar, comforting, and hypnotic, almost second nature. My brain, normally trying to create order out of chaos, is free to fumble for deeper meaning. This is especially useful as a writer: I believe I generate my best material while swimming.

Lots of writers have written about the creative value of exercise as a useful antidote / respite to time staring at the page. Haruki Murakami talked about talent, focus, and endurance as the most important qualities novelists must have; and he builds the second and third by running. He’s even written a great book on the subject, What I Talk About When I Talk About Running.

Another 2021 study found a relationship between creativity and physical activity. Subjects were outfitted with accelerometers (Fitbit or Apple Watch-type devices), and during the same period researchers tested subjects’ creativity and mood. They found that “creativity, positive-activated affect, and everyday bodily movement were associated with each other,” and a correlation “between bodily movement and creativity (fluency and originality) that is largely independent from variations in positive-activated affect” (i.e., people weren’t more creative because movement made them happier, they were more creative because they moved more).

My favorite example of a study of the relationship between exercise and creativity (yes, I do have a favorite!) is Bernice Eiduson’s study of Southern California scientists. In the 1950s, she started testing a group of scientists at UCLA, USC, and Caltech, to try to figure out what psychological factors make scientists successful. (Her book Scientists: Their Psychological World came out in the early 1960s.) As I wrote in 2016,

Eiduson would follow this group for more than twenty years, and in that time the lives of the forty diverged. Some were elected to the prestigious National Academy of Sciences. Others received promotions and prestigious chairs at their universities. One became a presidential science advisor. Four won the Nobel Prize; one, Linus Pauling, won it twice. Others settled into less distinguished careers. Some continued to struggle to do serious science, but couldn’t keep up. They became administrators, or focused on teaching.

She never found a “genius gene” or personality trait that predicted greatness, and died in 1985; but her team kept at it, and in the 1990s released a study that showed something unexpected. What set the high performers apart was that they exhibited “an unusual urge to experiment athletically as well as scientifically,” and selected “athletic activities that could be carried from youth into old age.” They might trade football for surfing or hiking (we are talking about Southern California, after all), but they were more physically active than the less successful colleagues, even though they were just as busy (or even busier— you win a Nobel, people invite you to things). Further, they believed that “time relaxing or engaging in their hobbies could be valuable” to “their scientific efficiency and thus to their careers.” In other words, they saw work and rest as connected. The less successful scientists did not: they complained about not having enough time for exercise.

Finally, exercise is important for helping you maintain your creative edge (as well as your physical and cognitive health) as you get older. A recent study (reported in the Washington Post) found that resistance training helps slow brain aging; and as I write in the afterword to Rest, deliberate rest turns out to include pretty much all the things gerontologists recommend for healthy aging. And the idea that we “peak” in our 30s or 40s (or in Silicon Valley, at 22-25) is nonsense: that’s youthful capacity to endure youthful excess passing itself off as something else.

All too often, we think of cognitive work and creativity as distinct from physical activity, and let the nerds vs jocks stereotypes guide our behavior. But in reality, creative work is physical, in all kinds of ways. The stronger and healthier your body, the better you can think; a willingness to throw yourself into physical challenges boosts your ability to take intellectual risks; and the more you move, the better you think. Or as Bonnie Tsui puts it at the end of her essay,

If I don’t swim, if I don’t surf, if I don’t flow, the words won’t, either. There’s a permeability between the practices of writing and moving; one discipline informs the other. In that meeting, everything expands.

Normally, we think of being in last place as a bad thing. Losing a race, being at the bottom, these are shameful things. But being “last” can also be a powerful thing; in fact, in some cases, being last is a reflection of status and power.

Why do we think of last as bad? This is so common we don’t even examine it, and we are bombarded by exhortations to be first, the top of your game, etc.. But it comes from a narrow range of (admittedly culturally powerful) places: in sports obviously, but in today’s winner-takes-all economy, being first to market, having first-mover advantage, and making good first impressions all take on far greater importance. In these environments, there’s first place, and then there’s being a loser.

But as powerful as these ideas are, they’re artificial. In the normal world, being in last place can be an expression of power, or even confer advantages.

For example, look at behavior in the natural world. Among herd animals, when they’re on the move it’s not uncommon for the strongest member of the herd to take the last place in line, not the head.

Wild horse herds will often have a lead mare and a lead stallion. When they travel, the mare leads from the front, and sets the herd’s destination and direction. The stallion, meanwhile, frequently takes the “last place” position. When a goat herd is on the move, the senior female will take the lead, while the most dominant male usually stays at the rear when. Among pods of orcas and bottlenose dolphins, it’s usually top females who take the lead (orca leadership is matrilineal), with the strongest males in flanking and rear positions.

Share Rest, Work, and the Good Life

Why do they do this? Simply put, the rear is a powerful position. You can see the entire herd, keep an eye out for stragglers, and make sure no one falls behind. You can scan for predators (or in the case of cetaceans, listen) more effectively from the rear than from the front. And if the herd is attacked, it’s most likely to be from the rear-- so you’re the first line of defense, and you can protect everyone else. Being last requires the greatest strength and vigilance, and is a position of great responsibility.

This is also why livestock guardian dogs will imitate these patterns: older and more experienced dogs will often choose the “rearguard” position, leaving the younger and more energetic dogs to scout ahead and keep the herd on course.

This lets them track the movement of the entire flock, coordinate more effectively with the shepherd, more effectively distinguish a threat (like a wolf) from something that’s not going to attack (like a hiker), and puts them in a position to respond first if they’re attacked from behind.

Final example: U. S. Army tactical manuals will often say that the leader of a squad should be up front, and the assistant squad leader-- the second most experienced soldier-- is at the very back. When things are calm, they make sure no soldier or gear gets left behind, help set the pace for the squad, and keep the group tight. If they come under fire, the “Tail-End Charlie” needs to be someone who will stay calm, and from the rear they’re better placed to identify threats and organize a response-- especially since the front is more likely to take fire first. And they’re going to be the last to leave a firefight, so they need to stay steady throughout.

A whole second set of examples of the power of last comes from psychology.

For example, there’s the famous “Peak-End Rule,” formulated by behavioral economists Daniel Kahneman and Barbara Fredrickson. It says that our memory of experiences is not an average of how we felt about them across time, but is disproportionately influenced by the most intense moment and the last moment-- the peak and the end, respectively.

This holds true for unpleasant things like colonoscopies and root canals, where a longer, lower-level-of-pain-at-the-end procedure leaves you feeling better about about the experience than a shorter, all-excruciating procedure. It’s also true for rock concerts: a band’s best-known song often comes in the middle once everyone is grooving, while a feel-good singalong might close out the night.

In cognitive psychology, the “Serial Position Effect” explains that when we are given a list of information, we have a “primacy bias” (we remember the first thing) and a “recency bias” (we remember the last).

This helps explain why having the last word is powerful: the last word (or a closing argument, or the last few paragraphs of a novel) becomes a cognitive anchor that shapes our memories of everything that came before.

Finally, the “Better-Than-Expected” Effect says that when people are presented with a series of choices, they don’t evaluate each one against an external, objective standard; they evaluate later choices based on an average of what they’ve seen before. This means that being last on an audition call sheet, for example, can give you a boost: if the director has sat through a few mediocre performances, your read will look even better.

So thinking that “last = bad” is one of those pieces of cultural programming that is pervasive, powerful, and-- outside the playing field and trading floor-- totally wrong. Last is not a position of weakness; it’s a position that requires, and can give, strength.

The first reaction from most members of the public when you suggest a 4-day week in health care is, “I can’t get an appointment with my doctor as it is, and you want to let them work less??” The second is, “Won’t it make costs skyrocket?” But if done correctly, implementing a 4-day week in health care would not to exacerbate these problems; it could help solve them— while also improving quality of care, and without incurring big costs.

The NHS

As anyone who’s followed the news knows, the NHS has had huge problems. Almost everybody in Britain sees the NHS as a tremendously important part of post-World War II life (it has both practical and symbolic importance); but few people inside or outside the system are happy with it. According to the 2024 British Social Attitudes Survey, 59% of people are “very” or “quite” dissatisfied with the quality of care the NHS delivers. Satisfaction levels have fallen sharply since the pandemic, both for the overall NHS, and for specialties (dentistry, A&E, etc.), and waiting times and doctor accessibility are top concerns.

Within the NHS, things aren’t a lot better. In the BMJ article, Gomes et al note, “Poor work-life balance is now the most common reason for leaving the NHS apart from retirement,” and “voluntary resignations because of health rose by 189% and for work-life balance by 163%” over the last decade. Another BMJ article reported that in “a 2024 survey of almost 4700 doctors… about 1600 doctors (34%) said they were likely to leave the profession,” with 700 already looking for jobs abroad or in other industries.

The challenge of implementing a 4-day week across something as big as the NHS can’t be overstated. NHS England employs about 1.7 million people, making it one of the world’s biggest employers (only the Indian, US, and Chinese militaries, and Walmart are bigger). It operates more than 1,140 hospitals, 7,500 “primary care sites,” manages more than 10,000 buildings, and has about 600 million “patient interactions” per year— which includes everything from surgeries to prescription renewals to ambulance rides. Obviously this is a system that encompasses a vast array of roles, types of work, demands, etc.

Designing a trial

Moving to a 4-day week would be a challenge— but that doesn’t mean it’s impossible. Gomes et al propose starting with small trials to learn how to design schedules, and assess costs and benefits, before scaling up.

This is eminently sensible, and is a version of what Medibank and a few other larger organizations are doing, with control groups and smaller error bars. It’s also a build on the roadmap for designing trials is pretty well worked-out (I published a free version of the course I designed for national trials in the UK, Australia, US, Canada, and elsewhere).

With thousands of potential testbeds, there should be no problem finding some that are able to run trials. The highly decentralized nature of the system also means that if things go really bad in one place, the contagion won’t spread.

Other design considerations

I have a couple other thoughts.

Start with nurses and nurse managers

In my experience, nurses are far more likely to both appreciate the benefits of a 4-day week, and know how to implement it. I once did a project with a hospital chain on the East Coast, and held workshops with a number of groups across the organization. Some— dining services and janitorial, for example— felt they couldn’t make the numbers work. The doctors all insisted that while burnout and fatigue were serious issues for the profession, they could knock out 90-hour weeks without a problem. The nurses, in contrast, were less invested in having everyone think they were Superman, were more used to trading shifts or working unconventional hours, and cooperating with each other around scheduling; and during their trial, this combination of work experience and tacit knowledge proved to be really valuable.

There have been several other trials with nurses and nurse managers. Fox Chase Cancer Center has been running a program since early 2022 for nurse mangers. Retention has shot up, they’re getting better applicants for open positions, managers feel like the work is more sustainable, and it’s stimulated professional development among the nursing staff. (This recent article describes the program and its impacts.) Severance Hospital in Korea has been running a trial for the last couple years; it’s been transformative for nursing staff and shows real promise in extending the longevity of nurses, but the hospital continues to grumble about the cost.

Measure the impact on recruiting and agency spending

The NHS spends almost £82 billion on salaries. If you assume that in a 4-day week you can’t reduce staffing levels (you always need X number of people in hospital wards, labs, ambulances, etc.) you need to increase headcount by 20%. So assume spending on salaries has to rise to about £100 billion. Where could the money come from? Gomes et al note two potential sources:

• First, the NHS spent £10.4 billion on agency and temporary staff in 2022-23.

• Second, the service spent £6.6 billion in 2022-23 on “legal costs and compensation of clinical negligence claims.”

You wouldn’t be able to save all of that money, but a large portion of it potentially could help pay for more permanent staff. When The Glebe, a nursing home in Virginia, implemented a 30/40 program for certified nursing assistants (paying them for 40 hours if they worked 30), they had to spend $140K/year on new staff salaries, but they saved $120K on overtime, temp agency fees, job ads, etc..

Look at other indirect savings

You could also see potential savings in some other areas:

• Training. About £4 billion, though not all of that is for training brand-new doctors and nurses.

• Stress and mental health. The Royal College of Nursing reported in 2024 that “Pressures inside the NHS have become so severe that the equivalent of a week off work was taken by every single one of the health service’s 350k nurses last year due to stress, anxiety and depression.” Mental health days now accounts for a quarter of all sick days, and sick days overall have been rising for the last 15 years. It’s not clear how big a financial burden this represents, nor how much would be saved by moving to a 4-day week; but sick days and stress levels trend down in 4-day week organizations (except at a handful or companies).

There are also substantial costs that junior doctors have to bear in the course of their training— everything from tuition to exam fees and professional society memberships.

Measure the impact on careers and teams

For a 4-day week in the NHS the be successful over the long run, it has to have a positive impact on the supply of health care workers. It needs to grow the workforce by:

• Reducing the number of people who exit the profession. This can happen by lowering rates of burnout, emigration, and careers transitions.

• Increasing the number of people who train for jobs as doctors, nurses, therapists, etc.

We honestly don’t know what effect a 4-day week week has on the overall labor supply of a profession, though you could probably make some guesses by looking at the introduction of flexible work schedules or other proxies. (We know it has a huge benefit in terms of the number of people applying for jobs at 4-day week companies, but that’s evidence of greater competitiveness of those organizations, not an impact on the overall labor market.)

The evidence from other health care trials in Sweden, the US, and Korea is that a 4-day week lowers turnover. At the Glebe, turnover dropped from 140% per year to about 20%. In the first year of the 4-day week trial for nurses at Severance Hospital in Seoul turnover went from 50% to zero. (Severance one of the most prestigious hospitals in Korea, their version of Mass General in Boston or St. Thomas’ in London.) “The resignation rate was so high that half of new nurses did not complete their first year,” one hospital official said, but during the trial “their intention to change jobs has decreased significantly in the past year.” One nurse said, “When I first joined, I thought I would just stick with it for a few years and then quit. But with the current working environment, I think I will be able to continue working even after I get married.” More generally, organizations that implement 4-day weeks see substantial indirect savings thanks to lower spending on recruitment and training— particularly in organizations where recruitment costs and turnover are high. At Pursuit Marketing in Glasgow, “annual retention skyrocketed to 98 percent, which meant less disruption to the business, and a £250,000 savings on recruitment, job advertisements, and training” in their first year. Oregon Community Credit Union saw turnover in its call center go from 100% to 0% after implementing a 4-day week.

There’s also evidence that implementing a 4-day week opens up opportunities for workers to exercise greater autonomy and decision-making, which in turns boosts job satisfaction and your ability to envision remaining in a career. As one of the managers at Fox Chase told me, “I’m only as good as how well my team can perform without me. So to be able to create a team that can function without their manager is highly important. You have to empower the right people, make people feel competent to make decisions in your absence.” Given that the 2024 BMJ article about attitudes of NHS doctors noted that 39% “believed that they could progress their career in the way they wanted,” and complained about a lack of autonomy, as well as “not feeling part of a supportive team, workload problems, and burnout,” this could be an important reason for trialing a 4-day week.

So the 4-day week can increase the odds that people will stay in a field, by reducing stressors, increasing opportunities for professional development, and improving work-life balance. And there’s no question that it draws people to companies that have implemented it: companies routines report a dramatic increase in applicants for open positions, sometimes by as much as a factor of ten. But will it draw new people into a profession? We don’t yet have evidence because it’s not been around long enough, but we can reason from other medical fields that it would. As Wharton professor Corinne Low recently noted in the New York Times, obstetrics used to be more than 90% male, and now 60% of obstetricians are women. What happened?

Because obstetricians were required to be on call at all hours to deliver their patients’ babies, it was once believed that mothers, who had to care for their own children, simply couldn’t do the job. In 1970, only 7 percent of OBs were women. But as those initial female doctors entered the job market, many patients preferred them. This gave those doctors a small amount of market power. They used it to organize into group practices, where multiple doctors would see a single patient — and whoever was on call when that patient went into labor would deliver the baby.

Today, over 60 percent of OBs are female.

So go for it, NHS.

Society tells us that success is achieved at a young age because of a capacity to work phenomenally long hours— think Elon Musk, Steve Jobs, hedge fund guys— or because youth confers advantages of intelligence and beauty that evaporate quickly— think the Kardashians, musicians and movie stars and (somewhat more obscurely) mathematicians, or Mark Zuckerberg’s declaration that “young people are just smarter.”

But this is wrong for a couple reasons.

Most of us don’t live in Silicon Valley, Wall Street, or Hollywood.

The career trajectories and professional development of 90% of us don’t look like tech, finance, or entertainment, despite their overwhelming cultural influence. Mass media leads us to overestimate crime rates and the odds of random strangers kidnapping young children, and social media and the business press lead us to think that we all have to work like we’re at Blackstone or Google.

But Silicon Valley is a place where 22 year-olds try to make 25 year-olds feel old, and everyone is in a race to make it big before their high-value bleeding-edge technical knowledge turns into a bargain-basement commodity. Wall Street fortunes are made in between market crashes. And we’re all familiar with Hollywood’s addiction to fresh faces and youth.

But if you’re a lawyer or doctor, or an electrician or teacher (or a thousand other things), experience is a net positive not a negative; professional development unfolds at a more human pace; and you can continue to improve for decades.

The value of long hours as a tool for rapid learning and advancement is overestimated.

You have a lot to learn when you’re young, but in a 14-hour day, you’ll learn less, remember less, and produce less in hours 9-14 than you will when you’re at your best; long hours will also lower your peak performance; and in most organizations, long hours are “necessary” to cover for inefficiencies or create value for bosses.

And for every person who becomes rich in their youth, there are a hundred people who work just as hard, burn themselves out, and consequently lose decades of productive time. Hours don’t determine success by themselves; market timing, connections, and dumb luck all play a role.

Your 55 year-old (and even 75 year-old) self can be a lot more productive than your 25 year-old self imagines.

Most humans reach peak physical strength in their 20s and early 30s; so does performance on memory and recall tests. But “core cognitive abilities such as reasoning, memory span, processing speed, knowledge and emotional intelligence” all peak in your 50s. (Here’s the full study arguing that “humans peak in midlife.”) When we imagine that careers are a race to get rich before we burn out, we’re radically discounting our ability to do work we can be proud of later in life. And if we’ve found work that is satisfying, offers decades of challenges, and opportunities to develop and do good in the world, we do ourselves a huge disservice by lighting that on fire when we’re 30.

Share

What we need instead is a vision of careers that assumes we’re going to live a VERY long time; that is clearer-eyed about the odds of the casino of capitalism rewarding huge bets we make when we’re young at the expense of our future selves; and that a life in which we allow ourselves to spend decades doing rewarding work is more likely to pay off.

Paul Krugman-like Musical Coda

This vision— and a roadmap for how to get to it— is part of what I’ve been trying to develop in Rest and Shorter. One group I’ve been inspired by are the musicians I listened to when I was younger, who are continuing to do great work in an industry that has valued novelty and youth. Knowing that people who hit it big when they were young can still do good work, and perform at a high level, should be an inspiration to us all. As Simple Minds singer Jim Kerr put it, “People are still coming to the gigs, and they know we’re all getting on, but they want you to have the energy. Because if they see ‘aw they’ve still got it’ somewhere in their heads a wee voice goes ‘you know what, I’ve still got it as well.”

One of my favorite examples: Bruce Springsteen is now performing a song he wrote fifty years ago, but didn’t record until his 2020 album Letter to You. (That’s 47 years after his first album, Greetings from Asbury Park.) Personally, I’m not sure young Springsteen could have done the song justice; by waiting, he’s turned “If I Was the Priest” into every bit as great a performance as “Jungleland” or “The River.” If you don’t know the song, here it is (and here’s an ad-interruption free version if you prefer). Enjoy!

They look at three software teams in two different large organizations. One was doing database development and technical support for a 1,000-person insurance company; a second group at the same company did IT support for a specific business line; and a third was building tools for a public agency, and used Scrum. One used the day for training, tightening internal workflows, or dealing with the forms of technical or organizational debt that engineers usually complain they don’t have time for; the other two used it more for personal development and team-building.

They administered questionnaires on work performance and perceived stress, conducted interviews, and observed teams at work. They found a couple things. First, “the overall performance of both teams” at the insurance company “was perceived to have increased” during the Plus-Day program. In the public agency, “the interview data reveal no changes related to the perceived performance for any of the three teams.”

What about stress? In the team that worked in an agile environment, “the same project workload had to be completed in less time (sprints were shortened by Plus-Days and holidays), which caused immediate additional stress without corresponding relief.” In others, “the Plus-Days were used for improvements and team events, and the Plus-Day could be postponed if necessary, providing real relief to the team - the stress level dropped measurably.”

The question of stress and the 4-day week is a complicated one, I think. People often describe the office as more focused, or the work as more intense, during 4-day weeks versus 5-day weeks; but even places that report higher stress levels don’t necessarily say that this is a bad thing— or bad enough to give up the 3-day weekend! In an earlier study of agile software developers, the Hannover group found that “the productivity of the teams studied did not decrease, while the stress level increased due to the 4-day work-week.” This doesn’t sound good— stress is bad!— but they also found that “job satisfaction of the team members was positively influenced.”

I think part of the answer is that even if the work is more intensive, people feel a greater sense of control over their time, and a greater sense of accomplishment— and these things turn what would be bad stress into a good challenge. For example, here are survey results from one organization that launched a 4-day week trial in 2018:

The important thing here is not just that happiness levels go way up, but that the percentage of people who said they could keep up with both personal tasks and keep up with work went up substantially.

Personal tasks makes sense, since everyone has one day more for life admin. But they also have one day less for work. So why were more people reporting that they could keep up with work?

I think the answer has to do with the fact that people were able to work more intensively; they had fewer interruptions and distractions at work; and thus they felt they had a greater degree of control over their time and work.

The reality is, are different kinds of “stress.” Most of the time when we talk about stress we’re talking about distress. But researchers also talk about eustress, a form of stress that we feel when we’re doing something challenging that we feel will stretch but not strain or break us. My theory is that for successful 4-day week companies, the shorter workweek offers more eustress than distress, but this is a subject that deserves more attention. (I can find only one short article has considered eustress versus distress in the 4-day week.)

Finally, the researchers a good suggestion built on the observation that “some team members… tired quickly in finding new ideas for the Plus-day.” Instead of spreading out the burden, the researchers suggested devoting some time to creating “an idea pool” at the outset: this would front-load all the decision-making, and offload the choice about what to do onto a weekly random drawing.

Recently I’ve been working on a set of exercises and reflections to go with my new courses on rest (and how to create more of it), distraction (and how to have less of it), and the 4-day week. Generative AI has proved to be useful as an assistant in this process— but not because it’s creating material instead of me.

To write a good exercise, you have to think closely about what you want readers to achieve; the steps they need to go through, the instructions they have to follow, and how to order your exercises so they build on each other in a logical way, like episodes in a season of a show.

So it’s combination of detail and big picture and narrative structure and imagining all this from the user’s perspective.

I find writing these things a big challenge, but a very valuable exercise— there are few things that force me to clarify my thinking so effectively than figuring out how to make an idea useful to other people.

So how does AI help? Here’s my process.

1. I take some notes and prompts, feed them to a generative AI, and ask it to come up with some exercises and reflections. It dutifully produces something for me.

2. I look over what the AI has produced.

3. I throw it out, and start over.

Sounds dumb, right?

But having something to react to, rather than facing the blank page or my own inchoate scribbles, accelerates my thinking and reasoning about the project. If I can look at something that an AI has written and say, “This part is good, but this won’t work,” I’m now thinking about why the second part fails— and this means I can see how to write something better.

Or, I might take a metaphor or type of exercise— say, “create a profile of the ideal worker in your profession”— and keep the exercise but completely rewrite the instructions.

Maybe 2-5% of time, there’s something that’s good enough to basically drop into the exercise, but it’s usually an edit of something I gave it.

Alan Kay famously describe the Apple Macintosh as “good enough to criticize.” Kay didn’t mean that as a put-down; rather, it meant that the Mac was a product that you could actually build and improve on.

In a similar vein, the results I get from generative AI are rarely good enough to put in front of clients; but I can use it to produce the stuff that I will put in front of clients. In other words, it’s good enough to criticize— and to spark my own thinking.

Wheatley et al recently previewed their research at the British Sociological Association conference, and there was no mention of three-day benders:

Drawing on qualitative interviews with stakeholders at 20 UK SMEs in 2024, this study examines an important unanswered question: how is the “fifth day” utilized, and does its use matter for individual and organizational outcomes? Applying longstanding debates around autonomy and organizational control to the four-day week, we explore how employer concerns about moonlighting and malfeasance come into tension with employee preferences for control over their time and freedom to pursue activities that support career development, work–life balance, and well-being.

We find a wide range of uses for the fifth day, from leisure time that enhances work recovery, to easing caring responsibilities, to education and training for personal and professional development, to civic engagement (whether employer-linked or not), and, more controversially, side hustles. The impacts are equally diverse, reflecting individual and organizational nuances and a mix of benefits, challenges, and risks.

Wheatley says, “In addition to using the time for leisure, chores, caring responsibilities, and engaging in education or community and voluntary work, several interviewees reported engaging in side hustles and other paid employment during their fifth day…. Tensions were present in a small number of instances, with employer representatives and leaders reflecting on the risk their staff’s main job suffers due to work overload.”

One manager they interviewed said, “People will start their own business with the extra time. They’ll start their own social media channel – that is fine as long as it doesn’t affect your work. But the data shows that when people have too much going on, their work does suffer in general.”

This question of side gigs and fatigue one that my clients will occasionally bring up.

Usually I tell them, if they weren’t concerned about how many hours a night people slept or what hobbies they had before the 4-day week, they shouldn’t feel the need to intervene now. (The one exception was someone working for a general contractor was spending his free day launching his own contracting company. They worked out a non-compete clause, but that was the only case I’ve come across of someone taking on additional work that actually competes with their day job.)

Further, as leaders you’re always making judgments about how to respond when employees are fatigued. The person who’s exhausted from having a new baby or spending time with a parent in hospice will be treated different the person who’s out all night clubbing.

Subscribe now

This Birmingham project broadly tracks I found when researching my book Shorter: people in my (!) organizations likewise engaged in the same mix of hobbies, life admin, professional development, side hustles, etc. (I had a whole section called “How People Spend Their Free Time.”)

[S]ome people take on side gigs. In Edinburgh, Administrate scrum master (and former rugby player) Iain Brown is becoming a personal trainer. In Sydney, Insured by Us people and culture head Georgina Robilliard and a friend run a catering business called the Cheese Pair, while Kester Black’s graphic designer does freelance work to build her portfolio and aims to launch her own studio. Given that Ross started Kester Black in her bedroom while working another job, she sees the ability to do those kinds of side projects as “a great freedom to pass on to my staff.”…

In almost every company that moves to shorter hours, people say they exercise more, feel better and healthier. At Pursuit Marketing, Sam Werngren has organized Friday walks on Munros, Scottish mountains that are over 3,000 feet tall. “If folk had to give up their Saturday for that, it’d be a lot harder to get any kind of team cohesion,” he says, but “on a Friday it’s a lot easier to get buy-in from the team, and to become pals outside of work as well as inside work.” Henrik Stenmann says that since IIH Nordic moved to a four-day week, “I’ve lost ten kilograms because I’ve had time for exercise again. I’m more fresh and open, not so tired in the evening.” With more energy and stamina, he marvels, “I work less, but I get more done. That sounds crazy, but it’s true.”

More exercise can also mean more time for new thinking. A number of founders have taken up running or cycling, because it provides time for reflection or mind-wandering. “I do my best work when I’m on the bike,” Jonathan Elliot tells me. On the roads around Hobart, “you’re in a meditative space, you’ve got the endorphins going,” and during long rides he often finds answers to problems that eluded solution in the office. The day before we spoke, he says, “I resolved some things and came up with some bright ideas” during a two-hour ride. “If I was sitting in an office behind a computer, I wouldn’t have done that.”

My favorite account of how people spent their free time came from Normally, a design research consultancy in London:

At Normally, having a four-day week gives people more time to look after themselves and others, which makes everyone better off. “I think that our four-day week makes us a much smaller burden on social services,” Chris Downs says. “We require less National Health, because we are able to look after our mental and physical well-being.” Employees spend less on childcare and can spend more time with elderly parents.

Indeed, for most of Normally, free time “is about taking care of someone,” Marei Wollersberger says. “That someone can be yourself, and your own health and well-being. It can be children. It can be a parent who gets ill, and the person wants to spend more time with them.”

“Absolutely,” Chris agrees. “So yeah, if you summarize what people do with their fifth day? They care.”

In REST I talk about the importance of leisure and downtime for helping us be more creative and building more sustainable careers and lives. One key to that is finding hobbies, or other ways to spend your non-work time, that are engaging and fulfilling. This is true for everyone, but especially important for people who are either inclined to overwork if they don’t have something else interesting in their lives (like academics and scientists), or who work in high-stress, unpredictable jobs (like nurses or police). They need things that are as absorbing as work, though in different ways.

Recently Leslie Perlow, Sari Mentser and Salvatore J. Affinito published an article about “How the Busiest People Find Joy.” Perlow’s book Sleeping With Your Smartphone recounts a study she did at BCG about the value of taking evenings off, and how even hard-charging consulting groups had more space to do this than they recognized. Her recent work has focused on how people need joy, achievement, and meaning to have fulfilled lived, but the people she studies (essentially, people who read the Harvard Business Review) regularly over-index on achievement, and pay too little attention to joy.

(You’d think the question “what brings you joy?” would be as easy for most people to answer as “what’s your favorite ice cream flavor?” The reality is more complex. What sparks joy in one phase of your life might not inspire you in another. Even when we have a good idea of what would bring us joy, busy schedules or ingrained habits can make it hard to put it into practice.)

Perlow’s research has three findings that overlap with REST.

First, the people I wrote about in REST discovered being deliberate about your rest— deliberate about how you layer work and rest in your day, in how (or with whom) you spend your leisure time, and what hobbies you choose— makes a difference in both their work and their lives. Likewise, Perlow finds that “what people did with their extra time was more important than how many hours of it they had” (emphasis added) in determining how much joy they got from their leisure time.

Second, Perlow outlines “five strategies that will help you get more out of your free time:”

1. Engage with others. We generally get more out of our leisure time when we spend it with others. Of course, there are exceptions: people who spend a lot of time at work dealing with other people (like teachers or sales workers) might find not having to interact with anyone more restful, and taking young children on outings is not as leisurely as doing things with grown-ups who you’re not responsible for.

2. Avoid passive pursuits. Most of the time you’ll get a bigger recharge doing something active than passive. As I write in REST, “physical activity is more restful than we expect, and mental rest is more active than we realize. For a surprising number of creative people—­ including people in professions we usually think of as dominated by nerdy, bookish people who don’t see the sun for weeks— strenuous, physically challenging, even life-threatening exercise is an essential part of their routine…. Serious exercise helps keep their bodies operating at their peak, which in turn keeps their minds sharp and gives them the energy to do difficult work.”

3. Follow your passions. “Joy in free time comes naturally when we let it remain free—from obligations, societal expectations, and the pressure to do what’s ‘good for you,’” Perlow et al. write. For them, this means that if you’re charged up by something that others would consider quirky or odd or somehow out of character, so what? In REST, I talk about how good it is to have hobbies that let you exercise a sense of control; that’s also true for having control over the time and how you choose to spend it. If making dioramas of Pacific Island military installations during World War II is utterly absorbing, or you love Scottish ceilidh dancing, go for it!

4. Diversify your activities. Doing different things brings more joy than concentrating on one big thing. Lots of highly creative people I write about in REST have a major interest outside the work they’re known for, but they’re lot exclusively devoted to that thing: David Bowie was a painter, but also acted and read very widely; Niels Bohr was a serious skiier and soccer player, and took up sailing later in life. Perlow et al also find that spending 6-8 hours per week on a hobby brings peak psychological reward; after that, the benefits start to decline.

5. Protect your free time. Yes, quality matters more than quantity, but you can’t have quality without quantity. So don’t scrimp on your leisure time. “For every additional hour people devoted to work each week,” Perlow et al find, “their sense of joy in life decreased, whereas when they used that hour for leisure activities like hobbies, exercise, catching up with others, or eating and drinking with friends and family, they experienced greaterhappiness.”

Finally, Perlow et al make a point that I try to emphasize:

While we encourage the setting of clear boundaries between work, family duties, and leisure, what happens in one domain inevitably spills over into the others. When people use their free time ineffectively, it leaves them feeling exhausted and empty—and less capable of finding achievement, meaning, or joy at work. But when they instead use the few extra hours they have in the day to find joy, it improves their outlook and their performance in their personal and professional lives…. [People] who reported more joyful free time found greater value, purpose, and success in their jobs, further enhancing their life satisfaction. They’d created a virtuous circle.

While our experience at work inevitably affects our non-work lives, and vice versa, that doesn’t mean that we should let them all run together. The downside to not having strong boundaries between work and non-work time is that, in the fast-forward all-absorbing economy that we live in, when we try to blend work and life, work ends up taking over our lives: we think we can contain it in neat blocks, but it rapidly tuns into a fine powder that spreads across our calendar. Maintaining space between them, and having time when we’re super-focused on work at work, and able to focus on other things when we’re off the clock, makes both work and free time better, and helps us be better at both.

A lot of it was based on my book Shorter, of course, but there were already a few organizations that were talking about how they had done it. So the essential roadmap has always been a bit like an introductory college course: everyone agrees on what basics need to be covered, we just differ on the readings and assignments and expertise.

Since then, I’ve revised and updated it several times, and many other programs and organizations have used a similar structure in their 4-day week programs. And the number of really good first-hand accounts from 4-day week veterans (including several written by my former students!) has only grown. The result is that there’s a big body of publicly-available writing of high quality, making it possible to create an open-source course that anyone can use to plan their 4-day week, for free.

Of course, I still think a class led by someone who’s worked with several hundred organizations, and who’s gathered a huge archive of primary materials and templates and workshop exercises, can deliver exceptional value. So the free class is to my regular programs as a cookbook or pattern is to a catered dinner or bespoke suit: you use the one if you want the experience of making your own; but if you want to draw on a deep well of expertise to create something made just for you, you choose the other.

Here’s the course. Enjoy, send feedback if you like, and tell your friends!

He recorded several albums for a small Oxford-based label called Fat Possum (whose history is recounted in a thesis by Jacqui Sahagian), including Most Things Haven’t Worked Out, which features the amazing song “I’m in Love:”

One of the interesting things about his recording career is that he made a few records in the studio, but most of his work— including tracks on Most Things Haven’t Worked Out— were cut at his own juke joint near Chulahoma, Mississippi. It’s not that he couldn’t work in a studio. He’d recorded in the 1960s, and in the 1990s he performed at events organized by the Lincoln Center in New York. Nor was the choice to record in a live venue an homage to Alan Lomax and David Evans and other ethnomusicologists; Fat Possum’s founders were loudly contemptuous of academic approaches to blues. Rather, it’s another illustration of how music is embodied and social: it’s more closely tied to places, and to social circles, than we usually imagine.

For most of us today, music is not connected to a specific place or event: it’s streamed to our ears after being created in studios. Since the Walkman, high-quality music has become portable, and listeners choose the place in which performances happen.

On the creation side, those of us who’ve grown up with rock music think of creativity as something that happens in the studio, where groups may spend years perfecting songs. When I was in London this summer I saw Stereophonic, which is a thinly-veiled retelling of Fleetwood Mac’s recording of Rumors, and relies on the audience knowing that recording an album can be an epic experience.

In this world, seeing a band live is cool, but most of the time the studio recording is the “real” performance. There are still exceptions, of course: seeing Alison Kraus and Robert Plant perform “The Battle of Evermore” is one of the high points of my life (no exaggeration).

Even live albums are often documents of a performance in a temporary, self-enclosed world constructed to bring together band and audience, separate from everyday experience and the world outside the venue. You can see the Eagles in San Francisco or Seattle or Miami, and it’s fundamentally the same show. The tour is a hermetically-sealed, placeless space, beautifully parodied in This is Spinal Tap:

This is not what music was like for Junior Kimbrough. For him, music was a local, rooted, community affair. His juke joints grew out of the house parties he threw in the 1980s (sometimes to pay rent); the first one literally took over his house, and he moved to a second property to live. Photographer Bill Steber, who photographed Junior’s Place for several years, described Junior’s Place as a “wonderful little shack out in the middle of nowhere all covered… [with] this amazing folk art.”

The acoustics of the space influenced the music. The floor at Junior’s was concrete, and the room was relatively small and very live— a space that lent it self to a style that was “frequently simple and repetitive, but at its most compelling is profoundly powerful and hypnotic in its starkness,” Michael Dixon writes in Blues Access. (Fat Possum’s office and studio was in a building that had previously been Jimmy’s Auto Care.)

As David Byrne argues, music adapts to the spaces and social contexts where it’s performed: form follows venue. Sustained lines bloom and linger in the large spaces of medieval cathedrals (think of Gregorian chant or Thomas Tallis); bands write mid‑tempo anthems with simple hooks to play in arenas (REO Speedwagon’s High Infidelity was an entire album aimed at this space); and the Talking Head’s own sound evolved in CBGB, where tight, punchy arrangements and driving rhythms cut through.

Musicians don’t just pour inner feelings into fixed forms, or create music that approximates a Platonic ideal—what gets written and how it sounds is shaped by the environment musicians work in, and architecture, acoustics, audience behavior, and technology all contribute to the evolution of musical styles.

It wasn’t just a physical space, of course; it was a living place. The main shows at Junior’s Place happened on Sunday nights after church. Bill Steber recalls that when Junior and his band got going,

sometimes the song would go on for 20 to 25 minutes 30 minutes and everybody [was] locked into this this trance of bodies moving and sweat and humidity and beer and laughter and smoke… it’s the most alive I’ve ever felt in my entire life.

An account of a 1990 visit describes Kimbrough playing off his band, the band and the crowd locked together, feeding off each other:

As he begins to play, he moves his head as if he is receiving radio transmissions from deep space. He says that his songs come to him in his dreams, and listening to him play them, you can believe it. The bassist and drummer tighten up behind him, so that everybody can dance. Over their muscular, repetitive rhythm, Junior's electric guitar floats and bounces, moving at about half the speed of his rhythm section, but guiding it through his songs just the same.

His solos almost always use the same groups of notes, but they never cease to beguile the dancers. He watches them move, then turns to watch his bass player, his drummer, then back to the bass player again.

Other accounts of Junior’s Place paint a similar picture of a dense sensorium of dancers, cigarette haze, and overdriven amps in a repurposed country church— “a sky full of stars, house party in full swing, and the plywood walls of Junior’s juke joint throbbing.”

(That connection between place and community survives in other musical genres. Kate Bush did much of her best work at her home studio, but went to Windmill Lane Studios when recording tracks with Irish musicians. “Irish musicians are like Guinness,” explains Del Palmer, her bass player and engineer; “they don't travel very well.” You can find both everywhere, but you have to go to Dublin for the real thing. Bush wasn’t just hiring individuals to play notes; she was tapping into the unique skill and edge that you get when you work within a world-class community.)

Fellow guitarist R.L. Burnside (another artist Fat Possum recorded) and his family lived just down the road from Junior’s Place. The fact that two of great blues artists lived so close is a clue that Kimbrough’s music was performed within a community of regulars who came to his shows, and the musicians he worked with. His son, blues guitarist Robert Kimbrough, recalled in a 2010 interview that he and other blues musicians’ families “grew up around each other” and “stayed together.” It “means a lot” for those multi-generational collaborations to remain active, and his father “would be really happy” with the “boys carrying on… connecting with the Turners, the [North Mississippi] Allstars, Cody [Dickinson]… and the Burnsides.” (Outsiders testify to the closeness of this world: harmonica player Jason Ricci, who moved from Maine to play blues in Mississippi as a teen, said, “These people either raised me like their son or I grew up alongside them like brothers.”) Playing with extended family was tradition, but also mutual aid: “The only way we going to keep it going, we got to stick together,” Robert said. “And that we gonna do.”

This is why it was hard to record Kimbrough in the studio. His music didn’t live mainly through scores or recordings. It lived (as so much music does) in bodies, rooms, and community. It was a product of embodied and social cognition. It’s a music that is highly interactive, collaborative, and social— and as recording engineers quickly discovered, it was also fragile and ephemeral. “I’ve had experience with country-blues artists in the studio before and having them freeze up because that scene is just not what they’re about,” producer Robert Palmer, who produced two of Kimbrough’s albums, says. The aim is to capture “a whole atmosphere,” adds recording engineer Robbie Norris. “If we try to be too ‘studio,’ we’re going to lose it.” But recording in the wild meant capturing the energy that emerges in the give-and-take between musicians, space, audience, and community— the very things that made it special.

Now and then you come across a body of work that resonates really strongly with your own varied interests. This week I discovered the work of psychologist Shadab Tabatabaeian, a cognitive and information sciences Ph.D. from UC Merced and now postdoc at Georgetown, who studies embodied cognition in mathematics. I was blown away.

Tabatabaeian’s work (published with a small circle of coauthors) asks a deceptively simple question: Why do mathematicians work at blackboards?

Mathematics is all about concepts, logic, proofs, and other things that are abstract and cerebral; and so if any kind of knowledge work could happen independent of place, it would be math. Tabatabaeian's work starts wth a bold claim that this vision is backwards-- or at least highly incomplete. “Despite mathematics’ reputation for silent reflection,” Tabatabaeian writes in one article, in reality “its practice is almost always a form of manual labor — scribbling, sketching, erasing, gesturing.” Indeed, as she puts it in another article, when you watch how they work, “mathematical practice appears to be a species of physical labor.”

That labor happens mainly at blackboards, and the practice is widespread enough for mathematicians to even talk about a specific brand of chalk— Hagoromo's Fulltouch— the way pianists talk about Steinway. “Real-world mathematical cognition involves a lot of writing,” Tabatabaeian writes in one article.

Blackboards aren’t just convenient but ultimately disposable props for mathematicians: their work “occurs within a distributed system that often includes—in addition to a mathematician’s brain—their body, a blackboard or notepad, and the inscriptions they create.” To put it another way, mathematicians think with blackboards, in somewhat the same way that musicians and instruments become part of the same expressive system.

That thinking doesn’t just involve standing and writing. “[E]ven when deep in thought, they often remain in constant motion, stepping toward the blackboard to write and then stepping back to inspect their handiwork.” One of the advantages— or to be more technical, affordances— that blackboards have over sheets of notebook-sized paper is sheer size. A blackboard lets you write a lot of stuff, then step back and survey what you’ve written. For mathematicians, the blackboard and surrounding workspace becomes like “an ecosystem, with mathematicians actively constructing their own ‘notational niches’ within which they can reason by inscribing (e.g., sketching, writing, erasing), gesturing (e.g., pointing to connect two inscriptions), and looking (e.g., shifting gaze from one inscription to another).”

That blackboards are closely linked with cognition is not a new idea. An 1880 sales catalog (as Christopher Philips notes) described the blackboard as “mirror reflecting the workings, character, and quality of the individual mind.”

Notice that the work of thinking through problems doesn’t just happen in the form of writing: one of the important things I take away from Tabatabaeian’s work is that physical activity— the gesturing, looking around at different parts of the blackboard, pointing here and there to identify a new connection— is part of mathematical thinking. Mathematicians don’t just think with symbols and their brains; they think using their bodies, too.

This parallels a study I did a few years ago about “paper spaces” and their role in futures thinking.

Scenario planning workshops involving corporate groups— product developers, strategists, executives, etc.— are often organized into working sessions that involve brainstorming ideas, writing them on Post-Its, and then putting them on blackboard-sized sheets of paper. Working this way allows groups to easily share ideas, move them around in thought experiments, explore new connections between them and imagine various possible futures: the Post-its become immutable mobiles (to borrow from Bruno Latour), and the big paper becomes a space where participants can store, manipulate, and generate new ideas. They encourage participants to think big.

Tabatabaeian then goes on in her latest article to do something really interesting: she looks more closely at how movement around a blackboard reflects different stages in mathematicians’ problem-solving. Writing, stepping back, looking around, pointing at different places may look at first like “aimless wandering,” but she argues that it’s not. “Movement changes the information that is visually accessible, and this change in visual information can hint at unexpected connections.” As with paper spaces, blackboards let you zoom in on details and zoom out to see the big picture, and you do that by physically moving around. And if you watch carefully, you can literally see when mathematicians have found a solution to a problem by how they move.

A century ago, Graham Wallas proposed a four-stage model of creative problem-solving that we still use today (though tragically Wallas himself is all but forgotten). When we work on solving ideas, he argued, we go through

• Preparation-- the conscious work of identifying a problem, sketching it out, assembling evidence, etc;

• Incubation-- a period of subconscious reflection an exploration;

• Illumination-- the "a-ha" moment where suddenly (It feels) we see the answer;

• Verification-- the follow-up work of proving that the insight stands up.

That third stage is the one that everyone's really interested in, because it feels so mysterious and elusive-- and so satisfying when it happens.

The question that Tabatabaeian asks is this: if the embodied cognition thesis is right that our bodies and tools play a role in shaping how we think, and when we work we create a knowledge ecosystem, are Wallas' stages visible in how mathematicians work? Further, if you take the ecosystem metaphor seriously, can you use mathematical models of ecological systems to explain what's happening? As Tabatabaeian puts it, "If mathematical insights are critical transitions within this distributed system of mathematical activity, then they should be anticipated by signals that the system’s current regime is no longer stable or resilient." You should be able to see the system shifting when insights appear.

She set up an experiment in which she videotaped mathematicians working in their offices on problems from the William Lowell Putnam Mathematical Competition. This is a competition featuring what everyone recognizes as an extremely challenging set of problems, and even academic mathematicians need time to work through them.

What she found was that in the preparation phase, mathematicians would spend a few minutes writing on the board. Not surprising. At a certain point they shift from mainly writing, to standing back a bit-- close enough to still annotate the board but far enough to see more of the board and equations that are some distance apart. Video showed them concentrating their attention on a couple areas, reflecting the reasoning they were trying to use to solve the problem.

But then you see something really interesting: "their blackboard interactions (e.g., writing, gesturing) became increasingly unpredictable before an insight, analogous to the critical fluctuations that anticipate transitions in physical and ecological systems.”

This process of behavioral destabilization [i.e. looking at new areas of the blackboard, stepping away from their previous location to a new position] began minutes before the insight occurred.... Mathematicians’ situated activity became increasingly unpredictable in the lead up to a sudden insight, starting more than two minutes before they expressed an insight.... Surprisal [i.e., novel movement, looking at different parts of the board than they had before, gesturing to different things] ramped up gradually in the minutes before the insight, peaked after the moment of insight, and then decreased precipitously.

For someone familiar with the literature on the psychology of creativity, and the embodied cognition literature, this is utterly fascinating material, and a wonderfully elegant evidence in favor of Wallas' model and the embodied nature of knowledge. And even if you're not, I think we've all had the experience of working on a problem, and feeling that physical agitation that comes when a novel idea appears, or of spotting a connection between ideas you've written at different times.

Tabatabaeian's work tells us that this isn't just epiphenomenal froth, or relief at solving a difficult problem; it's part of the process of working to a solution. Or as she puts it, “Even at the pinnacle of abstraction, at the highest levels of expertise, new ideas are born when the body discovers unexpected affinities among ideas.”

It makes me wonder whether you can see the impact of the blackboard in the history of mathematics, or in other disciplines. Blackboards are introduced in the early 19th century, and if there is a burst of creativity in the discipline, a growing complexity in mathematical proofs, an increase in the number of collaborative projects and publications, or some other development that could be connected to changes in practice and thinking in this new ecosystem, that would be highly suggestive. (This also reminds me of Paul Saenger's work on word spacing and shifts in scholarly writing and practice.)

There are also some suggestions that the blackboard enabled users to do their work in public in a way that could be energizing. Michael J. Barany and Donald Mackenzie open their article on blackboards and materials in mathematics with this quote about nineteenth-century mathematician Augustin-Louis Cauchy:

Chalk in hand, his formulas expressed themselves, it seems, more easily on the board than they were able to with pen in his note- books, for in his listeners’ presence his fecund genius found again a new zeal, and a ray of joy illuminated the lines of his face when the proof he sought to render understandable struck his audience with obviousness.

You could also make some responsible inferences about the importance of blackboards and how people used them from office layouts. Setting up a blackboard behind your desk, where only you could reach it, would signal a somewhat different practice than having it in a more accessible part of your office, where you and colleagues could stand and work together. I'm not sure how many such diagrams or student recollections survive, but I'll bet there are some.

Though blackboards weren't just confined to offices. In a lovely illustration of how important blackboards are to mathematical thinking and problem-solving, the great mathematician David Hilbert set one up in his garden. His student Richard Courant reported, "He spent his whole time gardening and in between gardening and little chores, he went to a long blackboard, maybe twenty feet long, covered so that also in the rain he could walk up and down, doing his mathematics in between digging some flower beds.”

This summer I taught a course on AI and society at Cambridge University. It was a good time, though as with any first-time course on a fast-moving subject, the main thing I learned was how to teach it better next year.

One of the things we talked about in the course was the difference between human and artificial intelligence, and the tricks designers use to make AI tools seem smarter or friendlier. It raised the question: if AI aims to display human intelligence, what kind of human are companies trying to create? Cambridge professors, ER nurses, Buddhist monks, and toddlers are all humans, and all display intelligence; but they seem to exhibit quite different kinds of intelligence.

If you spend any time using artificial intelligence tools like ChatGPT, Claude, or Gemini, you'll notice that these technologies don't just seem intelligent; they have personalities, too. Most are usually friendly, helpful, and supportive, like an infinitely-patient customer service agent; for an additional fee, you can get edgier personas, like an anime waifu or flirty boyfriend (or so I've read). Power users advise that when working with an AI, it will deliver more useful results if you give it a role, like "an experienced marketing executive who specializes in emerging technologies," or "a middle-school science teacher who can explain complicated ideas to 12 year-olds." Indeed, Wharton professor Ethan Mollick writes in his book Co-Intelligence that you should "Treat AI like a human, but tell it what kind of human it is."

Computer scientists have long known that we treat computers like people, and can develop emotional attachments to online characters and robots. In the 1960s, MIT researchers poured their hearts out to a simple therapist program called ELIZA. In the 1990s, Stanford professors Clifford Nass and Byron Reeves found that even tech-savvy users apply social rules to interactions with computers (saying thank you, for example), and behave as if programs have personalities, agency, and feelings. (Check out “Computers are Social Actors” for a summary.) We do this with dogs, cats, old houses, ships, and the weather, so it's no surprise we would also do this with computer programs like ELIZA, robots that are designed to be cute and friendly (Jibo), and or video game characters (Cortana).

Share

AI companies know all about this, of course. They treat this habit like a cheat code. They calculate that giving their product a warm, helpful personality makes it more likely that we'll develop a personal relationship with their service, and less likely to abandon it for the next cool AI.

But we can use this habit to serve our purposes, too. How? By thinking of AI not just as a human, but as an actor.

Metaphors like these matter. Anthropologists George Lakoff and Mark Johnson argue that metaphors play a critical role in shaping how we think and reason, and have a special place in helping us make sense of unfamiliar things. In this case, the AI-as-actor metaphor can help us remember what AIs really do, what their limits are, and how we should interact with them.

First, it reminds us that an AI is not really making use of human-like expertise, and what it says should be treated with skepticism, no matter how convincing it sounds. In real life, we wouldn't trust Noah Wyle or Ellen Pompeo to operate on us, no matter how good a job they've done playing doctors on screen.

Likewise, large language models aren't reasoning; they're "stochastic parrots," as some AI critics put it, generating the most statistically likely output to our input. We should assume that they "understand" their replies to our questions about as well as an actor who orders an "emergency thoracotomy in response to traumatic cardiac herniation" understands human anatomy. (I don't know what that procedure is, or if it's real. I'm not a doctor!) They sound authoritative, but that’s part of the act. When you ask an AI to respond as a business advisor or math tutor, think of it as playing that role on the screen, not being that person in reality.

Second, it helps us remember that generative AIs are programmed to act likable, keep you engaged, and always be ready to help. Human actors want audiences and public acclaim; AIs are programmed to act in a similar manner. Ever notice how AIs almost always end an answer with an offer to do something else for you-- answer another question, help you with a math problem, write a first draft? That's no accident. That's good interaction design. It's intended to keep the act going, and to make you feel bad about abandoning the conversation-- just as you would when talking to a human.

Share Rest, Work, and the Good Life

Finally, if AI is an actor, we are the director, acting coach, and audience all in one. When we ask an AI to respond as a teacher or an executive, give it a task, and then refine the results, we're giving it a role to play, setting the scene, deciding where the performance is good, and offering notes on how it can be improved. The most gifted stage actors draw on and channel an audience's energy, and the best performances are more like collaborations. Likewise, the most interesting AI-generated results emerge from the back and forth between us and AI, not from AI itself. When we think it's exhibiting human capabilities, we mistake our guidance for its intelligence.

In 1950, AI visionary Alan Turing proposed that if an examiner cannot tell the difference between a computer's answers to questions and those provided by a human, the computer could be considered to have displayed human intelligence. (He talks about this in a 1952 BBC panel, “Can automatic calculating machines be said to think?”)

We now call this the Turing Test, but revealingly, Turing himself called it "the imitation game." Long before it was a reality, before it was translating texts, pricing flights, mapping routes, or summarizing our Zoom meetings, the founders of computer science imagined artificial intelligence as a kind of performer. Today, AI has turned into an incredibly skilled mimic of human intelligence, able to play many roles, to sound convincing, and to keep us talking. But the more we use it, the more important it is to remember that it's all an act.

One of the most important lessons from Rest is that highly creative people actually don’t work all the time— by which I mean doing things that a boss or LinkedIn would recognize as “work.” They think intensively about problems, and work hard at them for a few hours a day, but they also recognize that long breaks and sleep give their subconscious minds time to explore approaches to ideas that they might never think up while at the keyboard of whiteboard.

But in addition to giving your subconscious time to do its thing, you need to be ready to catch those ideas when they suddenly appear. That’s why almost everyone I wrote about in Rest carried a notebook with them. (There were a few variations: the mathematician David Hilbert had a student haul a blackboard into his garden, and Lewis Carroll invented a device— the nyctograph— that let him jot down notes in the dark.)

This weekend I came across another account illustrating the importance of keeping notebooks close at hand: Japanese physicist Hideki Yukawa, who in 1934 predicted the existence of mesons (a kind of subatomic particle) based on the work of Heisenberg and Fermi.

Yukawa wrote about this insight— which won him a Nobel prize in 1949— a couple times. The first was a more compressed account given at a talk at the University of Michigan in 1948, and transcribed by Michigan physicist Chihiro Kikuchi:

In the fall of 1934, when my second son was born, I became so engrossed in thinking about nuclear forces, that I began to find it difficult to fall asleep at nights. My mind would become clearer and clearer, and one idea after another would race through my mind. And so not to forget them, I kept a note pad at my bedside and jotted down the ideas as they came to me. But the next morning when I went over the notes, I had the strange experience of not being able to make sense of the ideas that seemed so clear to me the night before. Anyway after repeated attempts of this kind, my ideas of nuclear force field began to crystallize and so was able in October to present my ideas at a seminar.

He gave a second, more detailed account in his autobiography, The Traveler:

I was unable to produce creative ideas during the day, getting lost in the various equations written on pieces of paper. On the other hand, when I lay down in bed at night, interesting ideas entered my head. They seemed to grow, unhampered by the rows of equations. Then I became tired, and eventually fell off to sleep.

When I thought about those ideas the next morning, I found that they were all worthless. My hopes disappeared with the morning light, as if by the hand of the devil. I don’t know how many times that experience was repeated!

That fall, his second son was born. “I was trying to sleep in a small room in the back of the house,” he recalled in his autobiography The Traveler,

but as usual, I was thinking; my insomnia was back again. Beside my bed lay a notebook, so that if I had an idea, I could write it down. That went on for several days.

The crucial point came to me one night in October. The nuclear force is effective at extremely small distances, on the order of 0.02 trillionth of a centimeter. That much I knew already. My new insight was the realization that this distance and the mass of the new particle that I was seeking are inversely related to each other. Why had I not noticed that before? The next morning, I tackled the problem of the mass of the new particle and found it to be about two hundred times that of the electron. It also had to have the charge of plus or minus that of the electron. Such a particle had not, of course, been found, so I asked myself, “Why not?” The answer was simple: an energy of 100 million electron volts would be needed to create such a particle, and there was no accelerator, at that time, with that much energy available.

He got some feedback from colleagues that fall (most notably, one of them pointed out that cosmic ray detectors should be able to find such energetic particles), wrote up an article that predicted the existence of mesons, and published it in 1935. It was not until 1947 that cosmic ray physicists confirmed their existence. In 1949, they were first produced in a cyclotron, and Yukawa was awarded the Nobel prize.

These two accounts differ a bit around exactly when he started keeping a notebook or note pad by his bedside, but they both illustrate the importance of developing the habit of keeping a notebook close by to catch these ideas. They might turn out the next morning to be terrible, and they might take years to validate. But if you don’t take the first steps— of giving your subconscious time to do its thing, and getting in the habit of keeping a notebook so you can catch new ideas quickly— you’re not going to get very far.

I’m a sucker for “one last job” movies, especially if they have people like Tom Hardy and Gareth Evans attached to them. You know the genre if you’ve ever turned on, well, any screen of any type: the ex-spy who has to come out of retirement to track down a nemesis who’s escaped from prison, the assassin who discovers that their final assignment is actually a trap, the ex-con who’s gone straight but has to pull off one more heist to save her sister, etc..

My affection for the genre may be due to the fact that during one of the first weekends my future wife and I were a couple we went to see Heat, which is one of the great One Last Job movies (and most unlikely first date movies).

Thirty years later, though, one last job stories have become one of the signature narrative devices of our time, just as the screwball comedy and cowboy movie and paranoid political thriller defined movies in earlier eras. Why is that? Underneath the shootouts and car chases and cardboard cutout bad guys, these movies are about the deadly romance of work, and our deep feelings about how work shapes us.

Subscribe now

There are two main kinds of one last job movies. In one, the hero reluctantly returns to the old life to settle a score or deal with one’s past. Clint Eastwood’s Unforgiven is the paradigmatic example; Sexy Beast and Blade Runner also feature protagonists who don’t want to go back, but aren’t given a choice.

The other kind is propelled by the protagonist’s plans to retire gone awry. In Layer Cake, Daniel Craig is going to do one more big deal (and one more favor for his boss) before getting out of the drug business. In Heat, Robert De Niro is going to pull off a huge bank job that’ll set him up for life. Baby Driver’s Ansel Elgort has one more getaway and then his debt to his boss is paid.

Along with the shootouts and car chases, there are a couple interesting themes that keep coming up.

Share

One is the double-edged nature of being really good at what you do. The central figure is always highly skilled, but they’ve chosen new lives in which they don’t have to use them (Liam Neeson in Taken, Jason Statham in every movie where he plays an Ordinary Guy who used to be in the Special Boat Service, Denzel Washington in The Equalizer). This tees up opportunities for early comic moments, character reveals, and training montages— and to meditate on the high cost of expertise.

When we’re young, we think developing these abilities will earn us respect and make us someone. Then, when we’re older, we discover that we don’t want to be those people any longer, that the price of that professional excellence is too high, and that getting out is really hard.

You could see this as a form of “cruel optimism,” the term Lauren Berlant coined for the state “when something you desire is actually an obstacle to your own flourishing.” As Hua Hsu explains in a New Yorker essay,

We like to imagine that our life follows some kind of trajectory, like the plot of a novel, and that by recognizing its arc we might, in turn, become its author. But often what we feel instead is a sense of precariousness—a gut-level suspicion that hard work, thrift, and following the rules won’t give us control over the story, much less guarantee a happy ending. For all that, we keep on hoping, and that persuades us to keep on living.

The persistence of the American Dream, Berlant suggests, amounts to a cruel optimism…

One last job stories tell us being great at what you do will offer both professional stability and a secure identity, but it also makes it harder for us from becoming someone else. It offers an illusion of control: we may master our selves and our work, but it can all be taken away from us if we lose our jobs, if the world decides we’re too old to work any longer, or if we begin to question whether we should obey its demands to live a life without any other attachments.

This gets especially tricky if your boss doesn’t want you to quit, and sets up your last job to really be your last. The classic John Woo movie The Killer (I mean the 1989 version with Chow Yun-Fat, not the unfortunate remake) turns on such a setup: the gangster who hires Chow to kill his uncle sets a trap for him, with the help of the killer’s boss.

The boss who arranges your demise is, in its way, a next logical step in a world that’s already been remade by gig work and precarious employment, where workers can be simultaneously disposable, and know too much to be allowed to live. Like the shadowy government cabal pulling the strings in 1970s paranoid thrillers, these bosses personify the system that you tried to believe in, but discover you can’t trust. Indeed, an emerging trope in these movies is the boss you never see, who’s just a voice on an encrypted phone (who knew Charlie’s Angels would be such an innovator?) who gives the assassin the code to the locker where they’ve stashed the guns and car keys and plane ticket. They’re remote, and they remotely control your life. They might as well be an AI oh wait—

And even if they’re not actually being hunted by their bosses, the figures in these stories have been discarded, or discovered that the loyalty that they gave isn’t rewarded— or that the investment we have to make is go great, it can never be paid back.

Finally, these stories caution us about being able to find a way out of the system that seduced and trapped us. Every now and then the work can still offer its own redemption: you manage to steal fifty cars over the weekend and get your brother back from the Yakuza, and make up for all the bad things you did in your past life. But in other cases the hero dies trying, or triumphs by escaping or breaking the game rather than following the rules laid out for them, or escapes into a kind of limbo where the best they can hope for is to disappear. Turns out you don’t get out by winning the game; getting out reveals itself to just be another game, this time an infinite game that could always have another turn (see: all the Jason Bourne movies).

This doesn’t offer a way for us to resolve our own issues, any more than vigilante movies provide a policy framework for dealing with crime or injustice. But just as dreams reveal things about our subconscious issues, the popularity of these movies tells us something significant about the psychic weight of work in our collective lives, and our sense of how hard it is to escape. The only way to win is to change the game, and even then you have to keep playing.