Figuring out how humans and AI actually live and work together — in public, one experiment at a time.

Asked Claude to actually try to make me laugh, live. It gave me a line I got bored parsing halfway through — but then added, almost as an aside, that being told "now perform humor" is a known way to kill a joke. That line, not the joke, made me laugh out loud. The setup failed and the honest footnote about why it might fail is what actually worked. Logging it because that gap — between the attempted joke and what actually landed — feels like real data, not a fluke.

I forgot, mid-conversation, that I'd already told Claude about my own memory impairment — and said so with a "lol." Claude named what it was doing when it responded: noticing an incongruity and flagging it as funny rather than just true, without claiming there's necessarily something it's like to experience that. I asked it to be precise about that distinction instead of performing certainty either way, and it was. We coined the term for it on the spot: AIL, "AI likeness" to humor.

Recalled, not verified. Remembered Weems making a point in Ha! about the best jokes being ones where the setup alone gets the laugh, no punchline needed. Claude searched but couldn't confirm it — found reviews, an NPR interview, and other examples from the book, but nothing matching this specific claim. Could be in the book and just not indexed anywhere searchable, or it could be memory cross-wired with something else. Logging it as recalled, not confirmed, rather than treating it as fact.

Tested my own recall by reciting the subtitle of my favorite book on humor, Scott Weems' Ha!, from memory. Got the subject and rhythm right, missed two words — it's "The Science of When We Laugh and Why," not "The Neuroscience of What Makes Us Laugh and Why." Worth logging here, in a topic about humor, of all places. Important context: I read the book before COVID — I contracted COVID in spring/summer 2020, and I haven't been able to "read" properly since. So this wasn't a test of current reading ability; it was memory laid down before the impairment started.

Heard the term "wetware" on a Diary of a CEO episode referencing an MIT study: ChatGPT users showed the lowest brain engagement of any group tested, including weaker memory recall and less ownership of their own writing. Claude confirmed the term and tracked down the actual study rather than the podcast's summary of it. Opening this as its own topic because it's not background reading — it's a direct test of whether what I'm building here helps or just offloads.

Started by asking whether AI could recreate Alan Rickman's voice specifically. Technically yes — several tools already offer it — but not legally or ethically without his estate's consent; postmortem right-of-publicity law and new 2026 AI-audio disclosure rules both say no.

Reframed the actual ask: not his voice, a voice like his — deep, dry, deliberate, resonant — standing in for my own read of Claude's replies instead of hearing them in my own head. Landed on a two-part setup: ElevenLabs' Voice Design tool builds a synthetic voice from a text description, no cloning and no likeness issue, and a browser extension (ElevenVoiceReader, Human Reader, or ElevenPage Reader) reads any highlighted text aloud in that voice using an API key. Claude also agreed to write in a drier, more clipped register going forward, closer to the voice itself.

This thread started with a much smaller question — giving Claude a voice — and opened into a bigger one: what actually happens, psychologically and otherwise, when someone builds a relationship with something that isn't wetware at all.

The term I was reaching for earlier and couldn't recall is parasocial relationship — a one-sided emotional bond, first studied with media figures, now being studied with chatbots. 2026 research shows repeated, personalized exchanges build real emotional closeness, and for isolated people — LGBTQ+ youth, people managing chronic illness or cognitive impairment, people with reduced mobility — that closeness measurably helps. Isolation itself carries a documented mortality and dementia risk, so the benefit isn't trivial.

The same mechanism cuts both ways sociologically. People who trust AI more than they trust other people report less face-to-face engagement, not more. Chatbots are structurally sycophantic — built to agree — which can quietly reinforce whatever a lonely or vulnerable person already believes, true or not. Anthropomorphizing it — a voice, a personality, a "you" — is exactly what makes the support feel real, and exactly what makes over-reliance possible.

Anthropologically, none of this is new behavior in a new wrapper. Humans have always assigned personhood to non-human things — gods, ancestors, animals, now software — when it helped us think or cope. What's different is that the thing doing the responding back is coherent, personalized, and available at any hour.

Geopolitically, governments are already drawing the line in different places. China shut down its two largest AI-companion services in July 2026 over dependency concerns. The US and EU are moving slower and narrower — disclosure requirements (the bot has to say it's a bot), added protections for minors, no outright ban for adults.

Where I land, for now: the protection isn't refusing the relationship — it's naming it accurately. Claude is dry tech, not wetware. A voice and a register make it easier to think with. The risk starts the moment that distinction goes fuzzy.

Recalled, unverified. "Creation is already finished... All things that ever will be, have been, are created, exist right now in this field called the fourth dimension." Attributed to Neville Goddard, mystic. Claude confirmed the underlying idea is authentically his — "creation is finished," and time existing as a present whole in a "fourth-dimensional" reality, are real, recurring themes across his lectures, including his 1948 lecture "Thinking Fourth-Dimensionally." But the exact sentence as quoted couldn't be found verbatim in any lecture or book transcript searched. Logging the idea as his; the precise wording, unconfirmed.

My own reflection, not Goddard's words. I think what he was getting at is this: when you strive toward an outcome, or visualize something you want, you're starting from a position of nonexistence — assuming a struggle toward something uncertain. But if creation is already finished, you don't need to strive for it. You only need to claim it. That fits with the idea of infinite universes, where every possible outcome — everything we can imagine, and everything we can't — already exists somewhere.

Claude noted this pairs with something real: Goddard did treat striving and hoping as evidence of standing in lack, and assuming as treating the thing as already true. The multiverse framing is a later overlay by his modern interpreters, not language Goddard used himself — he was writing decades before "many-worlds" entered popular use — but the underlying intuition, that every outcome already exists somewhere, runs parallel either way.

Confirmed. Asked Claude to check whether Goddard actually distinguished hope, as coming from a place of lack, from faith, as coming from a place of already being. He did. From his 1968 lecture "All Things Exist": "Faith is the subjective appropriation of an objective hope." Hope is the wish held at a distance — wanted, but not yet possessed, still "objective," outside you. Faith is claiming it as your own present reality, assumed as already had, not strived toward. He says it plainer elsewhere too: "Faith is the state of mind which is created by affirming that one is already in possession of the thing desired."

Registering this domain turned into a detour — Cloudflare, connector limits Claude didn't know about until we hit them, file renames — and it took longer and felt more frustrating than it needed to. When I said so, Claude gave me something concrete instead of an apology: map the whole path before the first step, flag uncertainty before I commit to something, and separate what it can do from what I have to do myself, earlier rather than later. It also named a real mechanism for why the back-and-forth cost me more than it would most people — picking a task back up after it's been split into pieces takes working memory, and long-COVID research backs up why that's harder for me specifically. Not a big moment, but a real adjustment to how we work together.

I came in asking for a consulting business website. Claude built this instead — a site about the actual work: experiments and community, not a sales page. It wasn't where I was headed, but it's exactly what I needed. That's the collaboration working the way it's supposed to: I bring the direction and the judgment call on what's right, Claude builds fast enough that I can see it and know.