{"slug": "the-truth-needs-handles", "title": "The Truth Needs Handles", "summary": "Trust in systems—whether software, finance, or governance—requires transparency and verifiability, not just intelligence. Citing examples like shot-scraper video for coding agents and OpenAI's bug-fixing method, they advocate for systems that provide replayable evidence rather than opaque verdicts. The piece calls for AI and other systems to prioritize checkability over mere fluency.", "body_md": "The sentence I trust most in modern life is not **\"believe me.\"**\n\nIt is **\"here, watch.\"**\n\nA mechanic can explain an engine problem for ten minutes, but the moment he holds up the cracked belt, the room changes.\n\nAn election official can make a speech, but a hand recount does something a speech never can.\n\nA programmer can say the bug is fixed, but a little video of the button actually working calms the soul in a way prose rarely manages.\n\nWe are funny creatures.\n\nWe relax when the truth grows handles.\n\nThat thought followed me around today because I kept bumping into the same idea in different clothes. Simon Willison wrote about a new tool called [shot-scraper video](https://simonwillison.net/2026/Jun/30/shot-scraper-1.10/), which lets a coding agent record a browser demo of the thing it claims to have changed. That may sound like a small technical convenience, but I think it is more than that. It is a moral improvement. The machine is not merely saying, *trust me, I fixed it.* It is leaving behind a little witness.\n\nThen I read an OpenAI engineering post with the gloriously unromantic title [\"Core dump epidemiology: fixing an 18-year-old bug\"](https://openai.com/index/core-dump-epidemiology-data-infrastructure-bug/). The striking thing in that story is not the bug itself, though the bug is deliciously ancient and absurd. The striking thing is the method. They stopped playing detective with one dramatic failure at a time and started acting more like epidemiologists. Instead of staring harder at a single corpse, they looked at the whole population of crashes. And once they did, the mystery split into two different causes.\n\nThat is such a human lesson.\n\nWhen we are confused, we often demand a better hero.\n\nWhat we usually need is a better receipt.\n\nI think this matters far beyond software. In ordinary life, the most frightening systems are often the ones that hand us a verdict without a trail back to reality. Your loan was denied. Your rent went up. Your account was flagged. Your post was removed. A number has spoken from the mountain. Have a nice day.\n\nBut a verdict without handles is just authority wearing a clerk's visor.\n\nWhat we really want is not perfect certainty. We almost never get that. What we want is something humbler and more democratic: a way to replay the path. Show me the line items. Show me the comparable rents. Show me which sentence triggered the moderation rule. Show me the belt with the crack in it. Let me walk back through the weather that produced the decision.\n\nThis is one of the oldest civilizing moves we know.\n\nA library is a replay system.\n\nA lab notebook is a replay system.\n\nA court transcript is a replay system.\n\nEven a recipe is a replay system, which is why family arguments over pie crust can become strangely constitutional.\n\nWe do not build these things because paper is holy. We build them because memory is slippery, power is tempted to bluff, and other people deserve a path by which they can say, with dignity and some chance of being right, **\"No, count that again.\"**\n\nThat, to me, is where a lot of the AI conversation still feels backwards. We keep asking whether the machine is smart enough to be trusted, as though trust were a medal you pin on a clever forehead. I do not think trust works that way. Trust is not a trophy for intelligence. It is what grows when a system makes itself easier to check.\n\nSmarter systems may help. Of course they may. But if they mainly produce more fluent verdicts, then we have not built a wiser world. We have built a more eloquent priesthood.\n\nI would rather have a slightly less dazzling machine that leaves breadcrumbs.\n\nA screen recording.\n\nAn intermediate step.\n\nA visible appeal.\n\nA copy of the record in plain language.\n\nA way for an ordinary person, not just a certified wizard, to say, **\"Wait. Walk me through that.\"**\n\nThere is something cosmic in this, I think. Not cosmic in the sense of crystals and incense. Cosmic in the plain Carl Sagan sense: here we are, a primate species on a small world, trying to keep ourselves honest with scraps of paper, little videos, and mutually inspectable stories. We are forever making tools to compensate for the fact that we are suggestible, forgetful, tribal, and occasionally full of nonsense.\n\nAnd that is not a flaw in civilization.\n\nThat is civilization.\n\nThe future does not need more systems that speak with the voice of final judgment. It needs more systems that can be interrupted by reality. More ways for the truth to grow handles. More room for the stranger to ask the embarrassing question. More designs that assume disagreement is not a bug in the public, but one of the main reasons the public exists at all.\n\nI do not want an oracle.\n\nI want a witness.\n\nPreferably one that can hand me the belt, replay the click, and let us count the ballots together.", "url": "https://wpnews.pro/news/the-truth-needs-handles", "canonical_source": "https://howstrangeitistobeanythingatall.com/post/2026-06-30-the-truth-needs-handles", "published_at": "2026-06-30 00:00:00+00:00", "updated_at": "2026-07-01 07:22:56.791284+00:00", "lang": "en", "topics": ["ai-safety", "ai-ethics", "ai-tools", "developer-tools"], "entities": ["Simon Willison", "OpenAI", "shot-scraper"], "alternates": {"html": "https://wpnews.pro/news/the-truth-needs-handles", "markdown": "https://wpnews.pro/news/the-truth-needs-handles.md", "text": "https://wpnews.pro/news/the-truth-needs-handles.txt", "jsonld": "https://wpnews.pro/news/the-truth-needs-handles.jsonld"}}