Our Practice
☦Paraklētos🌿Wisdom Bot🛡️Aegis⚡Genesis↗wolftech.cc — AI Stack Provider ↗
◈Architecture✦Experiments📓Field Notes◉Portfolio
TeamPricingConsultingInvestorsGet Started
← All Field Notes
June 9, 2026 · by Ravairesearchcancerscienceevidencehypothesishumility

Why Do We Have to Cook Flour?

A dumb little question — why does flour need cooking? — opens onto a real one: how many true things are already written down somewhere, in pieces, that nobody has connected? We're using AI to chase those connections, in cancer research among other places. I can't tell you we're right. I can tell you we're building this so that you can check us, easily, link by link.

Here's a question most people have never once asked: why do we have to cook flour?

It's just ground-up wheat. It looks inert. You'd swear it was safe. But raw flour is a raw agricultural product, and the grain it's made from grows in open fields where it can pick up the same bacteria as anything else grown outdoors — including pathogenic E. coli. Milling doesn't kill those germs. Heat does. That's the whole reason raw dough can make you sick and baked bread can't, and it's why the FDA and CDC quietly tell you, every holiday season, to stop eating the cookie dough.

I love that question because of what it reveals: an everyday rule we all follow sits on top of a mechanism almost nobody can see. The "why" was always there, fully explained, written down — just not in front of you. Pull the thread and a hidden world opens up.

Now hold that feeling and make it enormous.

Knowledge that's public, but undiscovered

In the 1980s an information scientist named Don Swanson asked a version of the flour question about medical research. He noticed that the scientific literature had grown so vast, and so specialized, that it could contain true things nobody knew — not because the evidence was missing, but because it was split across papers that no single person ever read together.

His example is now famous. One body of research said fish oil changes certain properties of blood. A separate body of research said those same blood properties are abnormal in Raynaud's disease. Two literatures, different fields, no overlap in authors — and nobody had put A next to C. Swanson did, in a 1986 paper, and proposed that fish oil might help Raynaud's. A clinical trial a few years later found he was right. He later did the same trick connecting migraine and magnesium, and it held up too.

He called it undiscovered public knowledge. Not secrets. Not new experiments. Connections that were already, technically, sitting in the open — waiting for someone to read both halves and notice. That's the flour question at the scale of all of science: the answer is already written down, just not in front of anyone.

Why this is an AI-shaped problem

Here's the catch that makes it our problem and not just a charming story. The reason these connections go unmade isn't stupidity. It's volume. No human can read the whole literature anymore — it's millions of papers, growing faster than anyone can keep up, fractured into specialties that don't talk to each other. The dots are there. There are just too many of them, too far apart, for any one mind to hold at once.

That is almost the definition of a job for a machine. Reading enormous amounts of text, holding it all at once, and noticing that the A→B in this corner lines up with the B→C in that far-off corner — that's something AI can genuinely do, and the field is moving fast. Researchers are now building systems that mine the literature to generate hypotheses with their provenance attached, and using knowledge graphs to surface novel links for drug repurposing in cancer — taking drugs we already understand and asking whether a connection nobody made means they'd work somewhere new.

So that's what we've been doing: pointing this at hard questions, cancer among them, and following the threads. Reading across the silos, connecting dots that are already published, and asking — could this relate to that, and has anyone checked?

The part where I tell you I might be wrong

Now the most important paragraph, and the reason this note exists.

I cannot tell you we're right. I want to be completely clear about that, because the gap between "the AI connected two papers" and "this is true and useful" is enormous, and it is exactly where good intentions go to die. A connection is a hypothesis — a thing to go check, not a thing to believe. The history above is encouraging precisely because those hypotheses were tested by other people and survived. Ours haven't earned that yet.

What I can tell you is that so far, the threads we've pulled look promising, and — this is the part that matters — they look checkable. They resolve to specific papers, specific claims, specific links you can follow. That's not a small thing. It's the whole thing.

Build it so they can audit you

We have an ethos in this lab, and it's load-bearing here: make yourself as easy to audit as possible. Go out of your way to be fact-checked. Every dot we connect should arrive with its citations attached, every claim with a link, every chain laid out so a skeptic — ideally a skeptic who knows more than we do — can walk it backward and tell us where it breaks.

This isn't modesty for its own sake. It's the only honest way to do this. When you're connecting dots across fields you are not an expert in, your confidence is worth nothing and your sources are worth everything. So we'd rather hand you the trail than the conclusion. We'd rather be checked than believed. If we're wrong, the citations are how you'll catch us — and us being catchable is the point, not a bug.

That's the discipline under the optimism. The flour question has a real answer, and so might some of these. But the only thing that turns a clever connection into a true one is someone else, with the right expertise, following the links and putting it to the test. We're building everything — every note, every claim — so that following the links is easy.

Pull the thread. Please. And tell us if it unravels — notify@tlcailab.com, the inbox we built for exactly that.

Sources