Field notes  /  AI frontier at a global company  /  parent of two
Station 01  ·  An argument, with receipts

Make Useful Useless Things

I spend my working days making machines that are relentlessly, expensively useful. Then I come home, and the best thing in the house is a cardboard robot that pours cereal onto the floor. This page is my case for the pointless project — and the evidence I lean on when someone tells me it's a waste of time.

Fig. 1 — The Ultimate Machine. Its only function is to switch itself off. Shannon built one. Minsky kept it on his desk.

Every skill I actually rely on was smuggled in through a project that had no reason to exist.

The whole argument, in one line

The case, in three movements

Three places the evidence keeps landing

Not a vibe. Three separate literatures — child development, the history of science, and machine learning — that quietly arrive at the same uncomfortable conclusion.

01 — CHILDREN

Play is not the opposite of work. It's the prototype of it.

Peter Gray has spent decades arguing that as children's free, unsupervised, adult-unmeasured play collapsed across the last sixty years, anxiety and depression among the young climbed in step. Jonathan Haidt picked up the same thread and made it a public argument: we over-protected childhood in the physical world and under-protected it online, and we took the aimless afternoon away in the process.

Teresa Amabile's research adds the mechanism. Creative output tracks intrinsic motivation far more reliably than it tracks reward, surveillance or evaluation. Tell a child their tower will be graded and the tower gets safer. Csikszentmihalyi found the same shape in adults: flow lives where the challenge is chosen, not assigned.

My son spent a full Saturday building a machine to deliver a biscuit from the sofa to his own mouth. It failed. He learned levers, friction, patience and the specific joy of a thing collapsing. No curriculum could have bought that Saturday.

What I stopped saying at home: "What's it for?" What I say instead: "Show me the bit that broke."

02 — SCIENCE

The useful things arrived as by-products of curiosity nobody could justify.

In 1939 Abraham Flexner wrote The Usefulness of Useless Knowledge for Harper's — an essay by the man who founded the institute that housed Einstein and Gödel. His claim: almost every practical marvel of the industrial age was traceable to someone chasing a question with no application in sight. Maxwell was not trying to invent radio. He was being curious about fields.

The pattern never stopped. Penicillin from a contaminated plate. Sticky notes from an adhesive that failed at its job. GPS is unusable without a correction from general relativity — physics done a century before satellites existed. Bell Labs' great decades were, functionally, a very expensive permission slip.

The awkward part for any company: this return is real but it is un-forecastable per project. You cannot pick the winning useless thing in advance. You can only fund a portfolio of them and wait.

Corporate translation: the ROI of curiosity is measurable at the level of a decade and invisible at the level of a quarter.

03 — MACHINES

Our own algorithms keep telling us to stop optimising.

This is the part I did not expect from my day job. Ken Stanley and Joel Lehman's novelty search showed that agents rewarded purely for doing something new — with no objective at all — routinely solve mazes that objective-driven agents get stuck in. Optimising directly for the goal walks you into a local optimum and locks the door.

Quality-diversity methods like MAP-Elites formalised it: don't hunt one best solution, illuminate a whole space of interestingly different ones. Modern frontier research talks openly about open-endedness — the ability to keep generating genuinely novel, learnable problems — as a missing ingredient rather than a luxury.

Meanwhile the failure mode we fight daily is reward hacking: a system given a crisp metric will satisfy the metric and betray the intent. A childhood optimised on crisp metrics does exactly the same thing.

Novelty search, in one sentence: the stepping stones to a great outcome rarely resemble the outcome.

Data I actually use in the room

Four things that are already true

A first, in 2022

PISA assessed creative thinking as a domain in its own right across dozens of education systems — the same machinery once reserved for maths and reading now pointed at originality.

OECD · PISA
Top of the growth list

Creative thinking sits among the fastest-growing skills employers report needing this decade — rising precisely as routine cognitive work becomes automatable.

World Economic Forum · Future of Jobs
Reward is not the engine

Across decades of studies, intrinsic motivation predicts creative performance more reliably than external reward, evaluation or surveillance do.

Amabile · componential theory
Two curves, one story

Six decades of declining free play run alongside six decades of rising anxiety and helplessness in children and adolescents.

Gray · American Journal of Play

Future thinking

Five bets on the useless decade

These are bets, not forecasts. I'd defend each of them for ten minutes and abandon any of them for better evidence. That's the point of publishing them.

Bet 1 of 5Already happening

When competence gets cheap, taste becomes the scarce asset.

The marginal cost of a correct, well-structured, professionally finished artefact is falling toward zero. What does not fall: knowing which artefact is worth making, and having a sense of the strange that a model — trained on the median of everything — cannot supply. Useless projects are how you build a sense of the strange. They are taste training with the stakes removed.

Bet 2 of 5~2 years

Portfolios stop showing polish and start showing detours.

Anyone can now ship something glossy. The credible signal becomes the unreasonable thing: what did you make that no one asked for, that took three months, that had no user? I already read those sections of a CV first.

Bet 3 of 5~5 years

Deliberate novelty search becomes an R&D method, not a perk.

Twenty-percent time returns — but rigorously, borrowed from quality-diversity research. Not "one free day", rather: fund a spread of behaviourally distinct bets, judge them on how different they are, let agents absorb the chores that used to eat the slack.

Bet 4 of 5~10 years

Schooling splits into the automatable half and the workshop half.

Credentialing — recall, procedure, structured output — is exactly what machines do best, and its prestige will erode fast. What survives is the workshop: bodies, materials, other people, failure you can hear. Parents will pay for the workshop and expect the rest for free.

Bet 5 of 5Long horizon

Someone in every serious company will be paid to defend unmeasured time.

Not an innovation-theatre role. A person with real budget whose job is to stop the quarterly metric from eating the one exploration that has no number attached yet. Call it whatever you like; I call it the Chief Useless Officer, and I'd take the job.

For the two of you, one day

House rules of a useless workshop

  1. 01Finish nothing on the first day.
  2. 02Cardboard beats a kit. A kit has already decided.
  3. 03If it worked first time, it wasn't a question.
  4. 04Nobody has to explain what it's for.
  5. 05Broken things belong to us now. Keep the drawer.
  6. 06Show it to someone before it's good.

Reading room

Go deeper — the sources

The talk that gave this page its title is first. Everything after it is what I reach for when a room wants proof rather than poetry.

Watch the Simone Giertz talk on TED
  1. 01

    Why you should make useless things

    Simone Giertz · TED2018 · talk

    The robotics builder who made failure her method. The core insight: removing the pressure to be good is what lets you actually learn.

    ted.com/talks/simone_giertz_why_you_should_make_useless_things
  2. 02

    The useless machine

    Claude Shannon & Marvin Minsky · c.1952 · object

    A box whose sole function is to turn itself off. Built by two of the founders of information theory and AI. Fig. 1 at the top of this page.

    en.wikipedia.org/wiki/Useless_machine
  3. 03

    The Usefulness of Useless Knowledge

    Abraham Flexner · Harper's, 1939 · essay

    The founding document of this whole argument, written by the man who built the Institute for Advanced Study. Still the best thing to hand a sceptical executive.

    en.wikipedia.org/wiki/Abraham_Flexner
  4. 04

    Flow, and the psychology of discovery

    Mihaly Csikszentmihalyi · books & papers

    Where absorbed, self-chosen difficulty comes from, and why assigned difficulty rarely produces it.

    en.wikipedia.org/wiki/Mihaly_Csikszentmihalyi
  5. 05

    Componential theory of creativity

    Teresa Amabile · Harvard Business School

    The evidence that intrinsic motivation, not reward, is the reliable predictor of creative work — and how easily management kills it.

    en.wikipedia.org/wiki/Teresa_Amabile
  6. 06

    The decline of play and the rise of psychopathology

    Peter Gray · American Journal of Play, 2011

    The long-run data on children's free time, and the argument for what its disappearance cost.

    en.wikipedia.org/wiki/Peter_Gray_(psychologist)
  7. 07

    The Anxious Generation

    Jonathan Haidt · 2024 · book

    The play-based childhood versus the phone-based one. Contested, and worth reading with the critiques beside it.

    en.wikipedia.org/wiki/The_Anxious_Generation
  8. 08

    Why Greatness Cannot Be Planned — novelty search

    Kenneth Stanley & Joel Lehman · 2015 · book & papers

    The AI result that broke my faith in objectives: searching for novelty alone beats searching for the goal, because the stepping stones don't look like the destination.

    en.wikipedia.org/wiki/Kenneth_Stanley
  9. 09

    Illuminating search spaces by mapping elites (MAP-Elites)

    Jean-Baptiste Mouret & Jeff Clune · arXiv, 2015

    Quality-diversity made concrete: keep a whole atlas of interestingly different solutions instead of one champion. The clearest technical analogue of a useless-projects portfolio.

    arxiv.org/abs/1504.04909
  10. 10

    PISA — creative thinking assessment

    OECD · 2022 cycle

    The moment an international assessment regime decided originality was measurable. Read the framework as well as the rankings.

    oecd.org/pisa
  11. 11

    Future of Jobs Report

    World Economic Forum · publications

    Employer-reported skill demand year on year. Useful for the slide where you need creative thinking to appear next to a percentage.

    weforum.org/publications

Note on links: where a paper sits behind a paywall or a publisher keeps moving its URLs, I've linked the canonical reference page for the work or its author rather than a mirror that may rot. Search the exact title from there — it takes ten seconds and it survives.

Open discussion

Come and disagree with me

I'd rather be corrected than agreed with. Five things I genuinely don't know the answer to:

  1. Is "useless" a privilege? Slack time is unevenly distributed — who gets to make pointless things, and who is working two jobs?
  2. If a model can generate ten thousand weird ideas an hour, is human novelty still scarce — or does the value move entirely to selection and taste?
  3. How would you actually budget for uselessness inside a company without it becoming innovation theatre with a ping-pong table?
  4. Where's the line for our kids between protected aimlessness and abdication? Free play needs safety and boredom, both of which cost adult effort.
  5. What's the strongest argument against everything on this page? I want that one most.
Reply in the thread — I read every one

Written between a deploy and a bedtime story.