The dubious wisdom of Mark Walton

Category: Artificial Intelligence

General thoughts related to AI creation and deployment

  • Autotunes of Innocence

    Current AI uptake has more in common with the U2 iTunes debacle than genuine adoption. Presence is not the same as value.

    In 2014, Apple pushed U2’s album directly into 500 million iTunes libraries without consent. Downloads soared. By every metric of distribution, it was a historic success. By every metric of reception, it was a disaster.

    Current AI spread seems more a triumph of marketing, puffery and leverage than actual intelligence. Most enterprise AI deployment isn’t happening because a CTO evaluated it and found it transformative. Private users are not seeking the tech and might not even be aware it has been injected into their device.

    Anthropic, OpenAI, Google and anyone else who wants to play, all are burning capital at rates that require narrative momentum to sustain valuation. So we are being forced into friction-free adoption of something which might not even be ready for prime time. Even those who might want such services can’t be sure they are getting the strong technology they need.

    That’s not adoption driven by intelligence. That’s adoption driven by distribution muscle and procurement inertia over product conviction. Counting seats and suggesting the model must be proportionally better accordingly, a distribution story dressed as an intelligence story.

  • Gamorrean Toilets

    I installed Gemma4 locally for experimentation – for now it will be a local roleplaying game assistant and oracle++, later I’ll make SillyTavern plugins for handling actual RPG game rules.

    The idea of giving the agent the ability to websearch for lore and ideas is a tempting one but I immediately hit a paranoid concern – what if I am running a RPG session where the players are tackling a terrorist threat, and the agent decides to search for ‘bomb making’, just for verisimilitude, and I end up in jail!

    (It’s a serious concern, in the UK the Terrorism Act 2000 / 2006 covers collecting information useful to terrorism, not just intent to use it.)

    Or even worse, the agent decides to search “elf mating rituals” or “gamorrean toilets” because it thinks they would add something to the current game. There’s some things I do not want on my permanent record.

    I can’t very well rely on the Hull Defence when I have specifically blogged against it.

    The conclusion must be, that for any agentic search, limit it to known useful resources. In the case of my RPG tasks, limit to relevant fandom sites and official lore sources. For broader tasks? I think I will not allow it to search at all, but instead create a plugin which allows it to tell me what search it would have liked to make, to aid its task. Then I can decide whether to allow it.

    I don’t want someone gaining access to my search history an rubbing ‘Gamorrean toilets’ in my face.

  • That’s So Cratic

    LLMS help me get over the “halo problem“, a term I coined for a fail state of self-education in the wikipedia age. Which is, I have a desire to learn about subject ‘A’ and look it up. Then when I encounter a phrase or concept ‘B’ which is foreign to me I explore that link, and anything problematic in ‘B’ I do the same… there was a Simpsons scene where Homer start with an advanced Marketing guide, drops that for a Dummies type guide, bins that and goes to look up Marketing in the dictionary…

    Well, that works a lot of the time, but sometimes ‘B’ includes a baffling concept ‘C’, when I try to understand that I trip over alien concept ‘D’, and so I read up on ‘D’ and… that article is contingent on already understanding my initial topic ‘A’. It circled all the way back around

    So instead of finding stairs up, the whole topic exists in a self supporting halo above me and out of reach.

    I won’t trust LLMs to understand things FOR me, I’ve no desire to activate an Electric Monk1. But it really gets me over the hurdle of not quite knowing how to ask another human (or redditor, close enough) what it is I don’t know about what I don’t know! Even hedging the comment to explain that’s what I am doing is painful.

    If LLMs do nothing else well, they do this.

    1. “The Electric Monk was a labour-saving device, like a dishwasher or a video recorder… Electric Monks believed things for you, thus saving you what was becoming an increasingly onerous task, that of believing all the things the world expected you to believe.”
      ― Douglas Adams, Dirk Gently’s Holistic Detective Agency ↩︎

  • Reflections on the Mirror Test

    I am annoyed, you see, by the mirror test for sentience.

    Considering how many perceptual illusions a human could fall for, like the rubber hand illusion, suggesting that any one test based on decoding a perceptual duplicity is a test for self awareness seems too ridiculous to even contemplate for any serious moment.

    To my mind, any creature which develops a survival mechanism must have a sense of ‘self’ otherwise they could not know which parts of their sensory experience are the parts they need to protect in order to survive.

    Denying an octopus or a corvid the dignity of accepting they are thinking beings, when we observe their problem solving in real time, is just rudeness disguised as science. Other creatures whose sensory priorities are different to ours will not have developed the relationship with mirrors that we have.

    Can we find a better description of the phenomenon we wish to understand, and drop this headline-baiting nonsense once and for all?

  • Emu was Innocent

    We cannot allow puppeteers to blame their puppets.

    From the 1970s until his passing, Rod Hull was partnered by a silent menace called Emu, a puppet operated by Hull’s own hand.

    Even without the reported use of the puppet as an excuse to touch people in inappropriate ways, the antics were presented as not Hull’s fault, because the bird Emu was responsible for the havoc and harm.

    We can’t allow bad faith operators to design puppets just so that they have the frail excuse that it is not their fault if their creations perform acts against individuals, groups, organizations or humanity as a whole.

    AGI is a perfect excuse for very selfish people to exert power via a puppet, and showing us that their hands are not inside it at the time. It’s a smug obfuscation of what we all plainly see.

    We can’t allow AI owners a ‘Hull Defence’ for any actions their machines perpetrate.