# Your Graph Should Not Be Inert
*Published: 2026-07-22*
*Tags: ai, insights, for-consultants, for-creators, for-founders, for-executives, agents*
*Source: https://chrislema.com/undiscovered-public-knowledge*
---In 1986, a gentleman by the name of Don Swanson, an information scientist, wrote and published a paper, and it was all about this idea of undiscovered public knowledge. Now, you read that — *undiscovered public knowledge* — and you're like, "That doesn't make any sense. If it's public, how can it be undiscovered?"

Well, what he had done is he had gone and done a whole bunch of collecting of research studies and medical literature that suggested that fish oil could lower blood viscosity. He pulls together that research. He didn't do the research. He's just finding it and pulling it together, and that is publicly known. It's available. Everyone understands it.

At the same time, he pulls other work together that completely separately says that Raynaud's syndrome involves high blood viscosity. And you can see where this suddenly goes for him. A is connected to B, C is connected to B. If B's sitting between A and C, is there the possibility that there's an actual connection between A and C? Could fish oil actually help those suffering from Raynaud's syndrome?

And so he writes this article in 1986, and it's ["Fish Oil, Raynaud's Syndrome, and Undiscovered Public Knowledge."](https://www.jstor.org/stable/4307965) That's the title. And I didn't know about it. It's incredible. It's amazing. It's awesome that he found this out and figured it out.

But you're probably going, "What does this have to do with anything?"

## What Does This Have to Do With Anything?

Well, I'll tell you. I have built a product, with our team over at Motivation Code, that is a [knowledge graph](https://coach.motivationcode.com/architecture), and I have spent a couple weeks working through how to ingest all sorts of content. You might have read recently some of these articles about graphs, and taking all your notes from Obsidian and turning them into a graph, and then you can use AI against it and all that. I love all that idea. But the challenge with those systems is that when you do the real work — the work that is at the core of it all — it's all at ingestion. It's all when you start.

When you're taking a new research paper, you're taking a new book, you're taking a new webinar, and you're feeding it in — that's when you start breaking up the content into spans, and then finding the atoms in those spans, and then determining where it connects to other things so you can build the edges, and then feeding it into the graph. All of that happens at ingestion. And then after that, normally, that's all there is. So now you just query it, and it shows you stuff.

But it's inert. It just sits there. It doesn't actually improve. It doesn't learn from itself.

## The Gardener

And that's not what I wanted. So I built some features. Like a *gardener*, who would look at who's querying what, what data is getting sourced, how often it's getting pulled, which paths and routes are traversed more often than others. And I start adding some weights, and I start defining some metadata around those edges.

I am not medically trained at all, but my visual mental model, if you will, is that the myelination in your brain — which is when neurons fire and they fire together, there's this myelin that starts wrapping around the connective tissue between those two neurons, and that ensures that the electricity dissipation happens less, which means they wire together, they fire together, and they fire better and faster. Something like that. Again, I am not medically trained. Please don't take any of this as medical advice.

But the point is that I started to try and build something like that in our graph, so that when nodes fired together, when they were regularly pulled together, we could create a stronger bond there — and when they were loosely pulled together, you could do that too. So the gardener does that work.

And you go, "Yeah, yeah, that's good. That's good." And then we start doing some new work at ingestion. Like: are there new kinds of edges that we should discover, that simply because they didn't exist already doesn't mean they shouldn't exist? And so there's another worker at the ingestion point that does some of that work, so that you're constantly creating potentially new edges, new ways to connect dots. Okay. Got it. That's good.

So today, I opened up Fable in Claude, and I said, "I want to talk about this other dynamic." Because we haven't looked at it. We haven't addressed it.

## Maybe Some Prospector, Looking for New Gold

It's this idea that — imagine that I have a node A in my graph, and I also have a node C in my graph. And now I introduce, through ingestion, through a new article, I introduce B. And node B comes in, and it gets immediately connected B to A. And at another point, in a separate effort, it also gets connected from B to C.

And shouldn't there be some — I won't call it gardener, because we already use that — but maybe some *prospector*, who's looking for new gold? Someone who's looking for, "Oh my gosh, this is what this does."

And so I describe this whole thing to Fable, and I'm saying, "Look, shouldn't we do this? Shouldn't we be looking for this?"

And Fable comes back and says, "Let me tell you about Don Swanson."

And I go, "Uh, who's Don Swanson?"

And it says, "Well, it's the guy who wrote 'Fish Oil, Raynaud's Syndrome, and Undiscovered Public Knowledge.'"

And so I go pull it up, from 1986, and I'm like, "Oh. This is exactly…" He was in the very, very early stages of working this through. The whole nature of publishing insights and information — which was true then, but is even more true now — is that it just gets splintered and pushed everywhere. And so something may be known in one corner of the world, something may also be known in another corner of the world. They may have things in common, but it doesn't connect. And so these are definitely bits of public knowledge, but they are undiscovered in terms of their connectivity, and their impact and result because of it.

And so, to conclude this part of the story, we started building the prospector. Because I think that's super awesome.

But in doing so, I started doing some research to see — hey, there has to be people now who have started working through this kind of problem. And no, it's not new. It's old. People have been taking this on and doing this. The journal article that I found was from 2012: ["A Graph-Based Recovery and Decomposition of Swanson's Hypothesis Using Semantic Predications."](https://pmc.ncbi.nlm.nih.gov/articles/PMC4031661/) And you go, "Oh, this is good stuff."

But what it ought to tell you, first, is: there's no new ideas under the sun. Like, I sit here, I'm talking to Claude, and I go, "Hey, shouldn't we be doing something like this?" — thinking, *this is nifty.* And it's like, "Yeah. Not new at all. You're about 60 years late to the party." And you go, "No, that's fair. I get that."

And then you go, "Oh, but we should *automate* something" — which most people probably haven't done. And then you're like — nope. Nope, sorry. Still 12 years out.

So there's nothing new under the sun.

## There's Nothing New Under the Sun

But the second thing that's really interesting about this is that **you have to choose to do this.** This is what Don Swanson was doing. He was choosing to go pull these things together, to find these connections. And most of the time, that's not what we do. We just get really comfortable with the data set that we're looking at. We get comfortable with the knowledge we have, and that's where it ends.

And you have to choose to dig in deeper, push further, and figure out — what if my data… it's not that it's stale. It's not that it's bad. **What if it's just not enough?** What if there's more out there?

And the beautiful thing is, as you build these graphs, the core takeaway — and my final point — is that **your graph should not be inert.** The data set, all your notes, everything you've written, your second brain, whatever you want to call it — it should not just be a static entity once you've ingested it. It really ought to continue to evolve, to change shape, to take new shape, and ultimately to discover what is previously unknown — and yet public *to you* — knowledge. You might have written one note in one place and another note another time, and all of a sudden you don't realize the connection that's already sitting there. And your graph should do that work. There is nothing stopping it.

## And It's a Little Bit Complicated

And to do that work is a little bit complicated. Because you can't just say every A that connects to B, and every C that connects to B — boom, that's awesome. No. What you've got to do is think: well, is B one of those connector nodes? Is B one of those where it has 755 nodes connected to it? That's not new information. That's not going to be helpful. But if the dynamic of that node is not to be a connector node — if it just has two pieces of information, three pieces of information — then legitimately, a new connection that makes something possible like that is super valuable.

And so that's part of what you have to figure out.

Also, the edges. When you connect two different nodes, two atoms, you connect it through meaning. There's semantic meaning to that edge. And so the edge might be *is evidence of*, or *mediates*, or *in parallel with*, or *prereq*, or *friction with*, or *competes with* — whatever terms you've used in your model. But not every one of those edges, not every one of those connections, will connect and follow through with the next one. They don't build on each other necessarily. So you have to start looking at: okay, which are the ones that are transitive? Which are the ones that propagate? Which are the ones that are parallel? Which are the ones I should care about, and which are the ones I don't need to care about? So that you don't create just a lot of noise.

And all of this is fascinating to me, because it's at the core of everything we're doing with AI. It comes back to knowledge, knowledge engineering, domain expertise. How do we understand it? How do we get AI to understand it? And what can AI do when it has it?

Don Swanson didn't have a graph. He had a library and a decision — to go pull the pieces together and see what nobody had bothered to connect. You and I have the graph. The decision is still the same one. So if you're building one of those second brains, if you're building one of those graphs, don't just fill it. Go looking. And let me know what you find.