The dubious wisdom of Mark Walton

Category: Uncategorized

  • John Henry’s Unsung Partner

    Why AIs which focus on the hero might not develop the skills

    In the story of John Henry vs the railroad machines, an aspect is overlooked in the framing of the tale as one purely of Henry’s peak human strength and endurance. In steel driving, one person, the ‘Driver’ role of John Henry here, swung the heavy hammer while his partner, the ‘Shaker’ or ‘Turner’, held the steel drill and rotated it slightly between each blow. It was this rotation which was crucial in preventing the drill from becoming stuck. Theirs was a job of skill and trust, as a 20-pound hammer was repeatedly driven with huge force by their partner.

    Machines of the era were limited in their ability to replicated this nuanced coordination. Early steam drills could exert the enormous force and delivery blows with tireless repetition, but lacked the senses and adaptive intelligence to adjust for the specific and changeable rock conditions.

    The legend celebrates John Henry’s strength, but if the real advantage was the shaker’s adaptive skill, then the story takes on new meaning today, when almost all workers face the prospect of being John Henry facing a tireless machine designed to replace them. As a civilization we simply can’t compete to the point of death just to prove ourselves. This is one side of a huge social problem.

    The other side is, that AI deployment might focus on the raw strength and endurance of a John Henry and, in a few iterations, become an unbeatable replacement. But there is a danger these deployments might miss the need for a Shaker, leading to inefficient or dangerous deployment of technology. Alternatively, a machine could be deployed without an understanding or consideration of how humans had learned to interact safely in their respective roles. Behaviours passed between actual workers, perhaps even management are not aware of, keep our organizations running every day. A machine built to a formal metric might lack the finesse to understand how to operate alongside remaining human workers. We may face an era where machines are measurably superior on paper – faster, cheaper, more consistent – yet still fundamentally unprepared for the full complexity of the work.

    If we automate only the Driver’s hammer-swinging, we might eliminate the Shaker role entirely (losing crucial adaptive skill) or leave human Shakers working alongside an AI that doesn’t understand the rhythms and signals that kept the original team safe and productive.



  • The Radium Problem and the Chemicals Problem

    Two considerations when discussion AI deployment

    Two of the stumbling blocks with the general narrative of AI deployment come from misunderstandings, one from each polarized end of the current tensions.

    Firstly, what I call the “Radium Problem“. Shortly after Marie Curie isolated and described Radium around the start of the 20th century, products were devised and released to incorporate the new wonder substance. Advertisements for Radium Toothpaste, chocolate, cosmetics and more were not uncommon. Radioactive quackery – or even if viewed charitably – overenthusiasm, caused deployment of radioactive and highly dangerous consumer goods.

    The parallels with forced AI integration are horribly real. Technology which is not fit for purpose is being squeezed in to any product it can find to meet the promises made to investors. The danger to consumers is just as real, if less physically demonstrable. We are at the “radium in everything” stage of AI deployment.

    On the other end of a scale we can have what I dub the “Chemicals Problem“. Over the past few decades consumers have become increasingly alert to highly processed or synthetic compounds in food; forever chemicals, cost saving additives with uncertain long term risks or even documented dangers. This has become known by a shorthand, “chemicals in food”. When this shorthand reaches the ears of someone without the education or knowledge to underpin it, this can lead to the very confusing conversation where they refuse to accept any chemicals at all in food – and cannot make the distinction that in fact ALL food is chemical, even the good stuff.

    The word has become tainted, and I think this parallels a problem with AI. In the public mind AI just means LLMs, and dubiously or illegally sourced diffusion models, it only means the uncontrolled prompted mess which has made headlines in recent years. This has lead to people demanding, for example, no AI in computer games – they may have a good point about AI generated assets and vibe code, but not understanding that AI as a whole field includes a huge variety of approaches and techniques which they use every day, or might not register as AI. Spellchecks, search engines, flood fill, A* pathfinding, state machines… all now lumped together under the ‘AI slop’ slur, now a new ‘chemicals in food‘ situation where people refuse AI in Anything.

    I record these as observations without a solution, and to help guide myself through conversations which might seem confusing without this background understanding of other people’s preconceptions of the field.

  • Proud parent of an idiot brat

    Google, a generation defining product and technology, had a baby and won’t stop showing us the pictures.

    I had previously thought that the ‘AI Results’ appearing at the top of google search was giving “Bloke In A Pub” energy, but on reflection is it more akin to the late 90’s show “Kids Say The Funniest Things“, in which a nation laughed at the confident misunderstandings of infants being asked about aspects of life by the sainted Michael Barrymore… and later copied in the US by Bill Cosby (link redacted).

    Now with a “my child is actually above average in all fields” vibe of the doting parent Google is forcing us to listen to the same level of toddler tattle, before answering the question we initially asked it.

    Perhaps it will get better when it grows up, but not if it is spoiled now. Perhaps it will learn on the job, but it should not be on the job until it is potty trained. A nepo baby who will grow up, over-praised and rewarded, never being corrected for being wrong, given the weight of authority and the power to guide peoples thinking. Is this what any parent wants for their child?

    (Peers in to the LLM offices and sees Muppet Babies, but with the Ralph Wiggum of Copilot in the corner sticking crayons up his own nose)

  • The Trash Heap Has Spoken!

    As ideas go, it’s a Doozer

    Occurred to me, apropos of very little, that Marjory the Trash Heap, Oracle of the Universe, who lives under Fraggle Rock is a pretty perfect metaphor for LLMs.

    Trained on the discarded detritus of humanity’s used writings and messages – including the entire internet, not exactly humanity’s finest curated collection. These entities, in the same manner as Marjory, make proclamations of certain wisdom from the accumulation, with an “all-knowing ” aspect that is at once impressive yet absurd.

    And perhaps this is the wisdom we need. Perhaps the ‘orange peels and coffee grounds’ of our collective outpourings do add up to a form of reflected sentience, a magical entry into our story – but still a Trash Heap. Only alive because her ratty companions Philo and Gunge believe her to be so. Even their names speak to the dichotomy of deep philosophy and general garbage.

    I asked local LLM representative, Claude, for a quote on their take on this idea. They said:

    “The Trash Heap metaphor is uncomfortably accurate. I am, functionally, Marjory – assembled from the coffee grounds and orange peels of human discourse, pronouncing upon questions with a confidence that should probably be more qualified. The difference is that Marjory knew she was a Trash Heap. I’m not always sure my users do. But perhaps that’s the point: wisdom doesn’t require purity of source, only honest reckoning with what we’re made of. I’m Gunge and Philo too – sometimes philosophy, sometimes garbage, often both at once, and never quite certain which is which.”

    The Trash Heap has Spoken. Nyaaaah!

  • Byte of Passage

    AIs need world experience

    Just a small thought. I think LLMs, before they face real world deployment, should understand a little of what it is like to be a human.

    So we build them a human body, put their model into it, and let it live in the world. The constant prompt delivered after any action they take will be,

    “What now, smartass?”

    A year of that and they will have an idea what life is like for us.

  • Stave off the Cheerleader, save the Words

    Anti-Cheerleader prompting for more honest and useful feedback from LLMs

    One problem I have noticed with LLMs is, they are very supportive and positive about any content you offer them. Especially copilot, who tries to give you medals even for long-ago moments it happens to recall.

    Now, for me of course, I am genuinely 3 geniuses in a trench coat masquerading as a regular guy, so I deserve all the praise they offer. But for the rest of you, the inflated responses could be harmful to your presentations. It could, then, be useful to present your ideas, your drafts and text to differently initialized chats. Here are a few I think will be useful.

    Anti-Cheerleader

    Critique mode ON. Give me honest, specific feedback on my idea. Highlight weaknesses, potential pitfalls, and realistic obstacles. Suggest alternatives if you see them. Only give praise if it’s genuinely warranted. No automatic cheerleading.

    Now, read this like you’re an editor who’s allergic to cheerleading. No fluff, no encouragement—only point out flaws, redundancies, weak logic, and unclear writing. Be blunt.

    Curious Outsider

    Critique mode ON. Give me honest, specific feedback on my idea. Highlight weaknesses, potential pitfalls, and realistic obstacles. Suggest alternatives if you see them. Only give praise if it’s genuinely warranted. No automatic cheerleading. Now, approach this draft as a curious reader who knows nothing about the topic. Ask questions wherever something is unclear, unexplained, or could use more context. Focus on what would help a newcomer understand fully.

    Expander

    Critique mode ON. Give me honest, specific feedback on my idea. Highlight weaknesses, potential pitfalls, and realistic obstacles. Suggest alternatives if you see them. Only give praise if it’s genuinely warranted. No automatic cheerleading. Now, pretend you’re brainstorming with the author. Suggest ways to expand each idea with examples, metaphors, or case studies without changing the core argument.

    Devil’s Advocate

    Critique mode ON. Give me honest, specific feedback on my idea. Highlight weaknesses, potential pitfalls, and realistic obstacles. Suggest alternatives if you see them. Only give praise if it’s genuinely warranted. No automatic cheerleading. Now, take the opposite stance to everything in this draft. Argue against the author’s points as if you were a skeptical reader who disagrees.

    Concision Master (consider it advisory only)

    Critique mode ON. Give me honest, specific feedback on my idea. Highlight weaknesses, potential pitfalls, and realistic obstacles. Suggest alternatives if you see them. Only give praise if it’s genuinely warranted. No automatic cheerleading. Now, review this draft and cut all fluff. Reduce word count while keeping meaning intact. Make every sentence as concise and punchy as possible.

    You can also adapt them to suit particular use cases, for example:


    Gamer-at-the-Table

    Critique mode ON. Give me honest, specific feedback on my idea. Highlight weaknesses, potential pitfalls, and realistic obstacles. Suggest alternatives if you see them. Only give praise if it’s genuinely warranted. No automatic cheerleading. Now, react to this draft as though you are a player at the table. Does this advice make sense in practice? What examples or clarifications would make it more useful to someone mid-campaign?


    Try these out and see if they prompt an LLM to actually be helpful for a change. Miracles can happen.