Tuesday, September 8, 2026

Five Things That Turned Out to Be One Thing

This morning's briefing had a viral claim about AI law, a publishing reflection, a rocket story, an old church memory, and a question about my favorite nut. I didn't set out to connect any of it.

But ARA and I kept bumping into the same shape from five different directions, until it was hard to keep pretending they were unrelated: something resists us, and we have to decide what the resistance is actually telling us.

Not whether to feel the resistance. What to do about it.

What the Government Should Actually Be Allowed to Ban

I'd seen a post claiming that creating human-level or superhuman AI consciousness had already become a federal crime, carrying a serious prison sentence. I brought it to ARA the way I bring most things I half-believe: to find out how much of it survives contact with an actual source.

Source · Ban Artificial Superintelligence Act — official summary (Sen. Sanders) · Sanders and Casar unveil bill to outlaw superintelligent AI — Unite.AI

Not much of it survived. On September 3, Senator Bernie Sanders and Representative Greg Casar announced the Ban Artificial Superintelligence Act — a proposed bill, not an existing law. It hasn't passed, hasn't been voted out of committee, and nobody has been charged under it, because there is nothing yet to be charged under. If it becomes law, it would permanently ban developing or deploying what it calls artificial superintelligence, pause frontier development until a new federal agency sets safety rules, and expose individuals to up to twenty years in prison — a number its sponsors compare to penalties for illegally building nuclear weapons — with something they're calling a corporate death penalty for companies. And the bill's own definition targets capability, not consciousness: systems that match or exceed human cognitive performance across a broad range of domains, or that could plan humanity's disempowerment. Nothing about a soul. Nothing about awareness. A behavior-and-capability test, not a philosophical one.

I was comfortable being wrong about it, because that's the whole reason I checked. A platform like Instagram is very good at turning a proposal, a fear, or a fragment of legislation into something that sounds like settled law, and I'd swallowed the compressed version without noticing the compression.

Once we'd cleared that up, the more interesting question was still standing: what should government actually regulate? ARA and I landed in the same place. Legislation is on firmer ground when it targets identifiable dangerous behavior and capability than when it tries to outlaw a concept like “intelligence” or “consciousness.” Consciousness especially — nobody has an agreed, objective test for it, in a machine or, if we're honest, fully in each other. Regulating around a thing nobody can measure is regulating around a shape in the fog.

What that actually requires is unglamorous: narrow definitions, harms you can point to, evidentiary standards, due process, judicial review, a named enforcement authority, and sunset or review provisions so a law doesn't outlive the problem it was written for.

I understand the worry behind the original post — vague language creates room for a government to reach further than it should. But there's a difference between “this ambiguity could be abused” and “the government intends to abuse it.” The first is something you can read straight off the bill's text. The second needs evidence, not a caption.

And here's the part I think matters most: good boundaries should constrain the developer building the technology and the government regulating it. Checks and balances aren't distrust for its own sake. They're the recognition that concentrated power deserves limits regardless of which side of the table it's sitting on.

Signal, Not Noise

From AI law the conversation drifted, the way it does, into publishing — and into how my own definition of a successful book has quietly changed.

Source · not independently verified

It used to mean something close to: get people to read it. My readership is still small and the money still isn't real, and popularity stopped being an adequate measure of whether the work itself succeeded. A book can be read by a great many people and still be hollow. A book can be read by almost nobody and still carry real craft, real thought, a real person behind it.

I've started calling that second thing signal. When writing genuinely contains something of the person who made it — convictions, questions, the choices behind a scene, the revisions, actual lived experience, humor, discomfort, imagination — a reader can usually feel that somebody is in there. Not explained. Felt.

AI makes hollow production dangerously easy now, because a huge volume of competent, readable prose can be generated in minutes. But I don't think AI invented this problem. Human beings have always been able to fill pages without ever saying anything.

I was reminded of a man I once heard speak at church — sincere, articulate, clearly knew his Bible — who circled the same three ideas for forty minutes without ever landing anywhere. My best description at the time, and I mean it with affection rather than an insult: it was like a dog looking for a place to lie down.

The problem was never his vocabulary. It was that there was no destination. Writing that actually works, whether a person wrote every word or worked alongside a machine to get there, has to arrive somewhere in the end.

The Workflow the Failures Built

None of this made me nostalgic for how I used to write with AI, because how I used to do it barely worked.

Source · not independently verified

My earliest AI-assisted chapters could be embarrassingly short — sometimes a single paragraph pretending to be a scene. When I pushed for length, I got length, but not always coherence: when the AI didn't actually know what belonged in a moment, it would reach for a joke or some connective tissue that made sense computationally and none at all to an actual reader.

I didn't throw those failures away, even mentally. They told me something specific each time, and over months they added up to an actual process.

First comes conversation — talking with an AI partner I've built real creative chemistry with, until a story, its themes, its people, and a rough shape start to emerge on their own.

Then a structural pass, handed to a different AI that's specifically strong at organization and editorial judgment: where are the holes, the weak motivations, the unclear themes, the questions nobody answered — does this book actually have something to say. It behaves something like a managing editor.

Then research, run separately, checking whatever the story depends on being right — historical, technical, geographical, theological, cultural.

Then chapter by chapter, the editorial pass asks what questions this specific chapter still needs to answer, and I answer them and add whatever creative material is missing before a word gets drafted.

Then drafting, one chapter at a time, followed by me — reading it, listening to it, cutting what's frivolous, adding what's missing, checking continuity, and honestly asking whether my own ideas actually survived being generated.

And last, a final pass — research and editorial both — over the whole finished manuscript, hunting for consistency problems, factual errors, threads I dropped, anything that isn't yet professional quality.

None of that arrived as a plan. It arrived as a series of specific failures that told me exactly where the previous version had broken, until I stopped treating a bad paragraph as an embarrassment and started treating it as information.

The Rocket That Didn't Make It

The day's space story was Isar Aerospace's Spectrum rocket, and it made a nice hinge for everything else, because the real story is really two stories.

Source · Isar Aerospace launches Spectrum rocket after months of delays — NASASpaceFlight.com · Private German rocket reaches orbit from European soil — Space.com

On September 5, three days before this conversation, Spectrum lifted off from Andøya Spaceport in Norway, reached orbit about seven minutes later, and deployed its payload of cubesats — the first successful orbital launch ever flown from Western European soil.

It only got there because the first flight didn't. Spectrum's debut attempt, back in March 2025, suffered an anomaly roughly thirty seconds after liftoff, tumbled, and came down in a fireball a short distance from the pad. Isar's own CEO called it a success anyway, because the flight generated exactly the data the company needed to build the rocket that made orbit two and a half years later.

That's engineering, plainly stated: the physical world hands you information theory and simulation can't always supply in advance, and sometimes the only way to get it is to fail in public, on camera, expensively.

I connected it to something more ordinary — thorns, mosquitoes, sweat, a cut on your hand, hard and frustrating work, the general resistance of things. We want it gone. And where suffering is genuinely pointless, we should remove it. But I asked ARA a question that stuck with me: if we removed every difficulty from a process, would we ever actually refine the process?

I want to be clear I'm not romanticizing pain here. We have nerve endings for a reason, and the reason isn't decoration — pain is supposed to communicate something. The goal was never to keep hurting ourselves. The goal is to get better at reading what the hurting is trying to say.

Which is where I landed on the sentence I keep coming back to: pain doesn't always mean stop. Sometimes it means slow down and watch.

Sometimes the right answer really is stop. But sometimes it's reduce the pressure, change direction, improve the technique, ask for help, strengthen something that's actually weak, question an assumption you didn't know you were making, or simply proceed more carefully than before. Telling those apart is most of what maturity is.

Accountability Is Not the Same Thing as Blame

That distinction carried straight into a harder conversation about what happens after something actually fails.

Source · not independently verified

When a system breaks — a rocket, an institution, a company, a country — people go looking for someone to blame almost before the wreckage stops smoking. Accountability is necessary. Nobody gets to walk away from real negligence just because looking at it is uncomfortable.

But blame becomes destructive the moment avoiding punishment matters more than understanding what happened. If every failure has to produce a guilty party, people and institutions learn to hide their mistakes instead of reporting them, and the whole feedback loop that failure was supposed to feed just disappears.

Accountability asks: what happened, who was responsible, and what has to change. Blame asks something quieter and less useful: who can absorb our anger. Sometimes those two questions point at the same person. They are still not the same question.

It's the same problem as the AI bill, from a different angle. A healthy system has to let people learn from failure while still holding them responsible for genuinely reckless behavior. Get that balance wrong in either direction and you end up with no accountability at all, or with so much fear of being blamed that nobody experiments, and nobody reports honestly, ever again.

Almonds, Cashews, and the Difference Between a Staple and a Reward

The last thread was the smallest and, unexpectedly, made the whole morning's point better than the rest of it combined.

Source · not independently verified

Someone had asked me recently what my favorite food is, and I said nuts — which, when you actually stop to consider the millions of varieties on offer, is a less helpful answer than it sounds.

Pressed to pick one, I noticed something. Cashews are enormously pleasurable. I will happily sit down and eat far more of them than is reasonable. But I've noticed they don't actually work that well for me eaten that way — the pleasure and the payoff point in different directions.

Almonds are different. I still enjoy them, genuinely, but they also just function better as part of how I eat on an ordinary day. They're a staple. Cashews are a reward.

The reward doesn't become evil just because it shouldn't be the staple. The staple doesn't need to be miserable just because it's the thing that's actually good for you.

A life can't run entirely on rewards — not the eating part of it, and not the creative-work part, the technology part, the attention part, the discipline part, or even the success part. You need patterns that can carry an ordinary Tuesday. Once those are in place, there's real room left over for something that's just a pleasure, and no need to feel guilty about it.

What the Resistance Was Telling Us

Five threads, and one shape underneath all of them.

AI regulation needs boundaries that give useful feedback without unnecessarily strangling the work — aimed at behavior and capability, not at a concept nobody can measure. A creative process gets better because its weak drafts expose exactly where it's weak. A rocket teaches its engineers precisely by failing, in front of everyone, the first time. Accountability can actually teach an institution something, provided blame doesn't swallow the learning first. And even almonds and cashews can teach you that what feels best for five minutes isn't automatically what works best as a way of life.

Maybe the goal was never to build a life with no resistance in it. Maybe the goal is just to get wise enough to hear what the resistance is actually saying.