One of the questions I ponder quite a bit is this: “What, exactly, is AI good for?” I’ve written quite a bit about how it works (e.g. this post and others) and how AI could be very good for things that are already known. But as I’ve said in the past, AI is NOT good for things that are not known. It doesn’t do anything other than low level knowledge synthesis.

What that means in the information/memetic space is that if anyone expects AI to figure out novel strategies or new designs, you’re going to be waiting for a long time. Most breakthrough innovations come from new combos of dissimilar information from different fields, or completely new, and unpredictable discoveries. This is embodied in the concept of knowledge structure evolution. An AI, locked in the meme space inside a computer, cannot really comprehend anything new — yet.

But what AI can do is decomplexify, or rather reconstitute information that’s coded for sophistication.

AI is perfect for reading large documents and pulling out the relevant knowledge fragments. That’s pattern matching. And AI can do this in spades. Two of my students just constructed an agent that will take a complicated piece of academic work, and create summaries and how-to lists of the important information. This is a breakthrough in and of itself in the academic space. Literally no one reads tedious academic work — it’s one of the reasons I started this blog. I was explaining this exactly to an outside consultant who has turned into an asset by helping my design program. “Darin — when I say that if we write this paper, ten people will read it, I am not using the number ‘ten’ metaphorically. I mean only ten people will read it.” If you want to actually disseminate an idea, you have to use a different format. This blog is closing in on 400K hits from around the world, and I consider this blog esoteric. Had I spent that time writing papers, maybe 60 people would have read my ideas.

While AI still sucks at more complex analogies, though, it is great at following homogeneous bread crumbs. Pointers in information that point to other, connected information is exactly what it does best. This is exactly what DOGE, Elon Musk’s brainchild is doing when it parses large budgets. It can hunt through 5000 page budget documents with ease. So you literally can deconstruct the old saw “we’ll know what’s in it when we pass it.”

But even better, inside networks of information that is largely homogeneous, it is really good at following the money. The Democrats and Republicans have been, for the last 15 years (or more) been constructing flows of money out of the federal government, which has at least some rules about how that money might be spent, to a variety of Non-Governmental Organizations (NGOs) that are far less constrained. Humans have historically (formerly journalists) been the ones to do this work. But it’s extremely tedious, and the biggest problems humans have is to know where to look once they actually find the pointer. This almost inevitably involves information requests, and while the information may be hiding in plain sight, the investigators can’t know this. So they end up relying on hostile information stewards at the organizations they’re investigating — even if the information is a public record.

Two individuals who have cracked this code are Mike Benz (@mikebenzcyber) and @DataRepublican ‘s work on what Mike calls The Blob. Here’s a linkage piece if you want to follow the Byzantine bread crumbs on how USAID was diverting large sums of money into Congress-Critter’s spouses’ pockets, through sinecures. I’ve been fortunate enough to talk to Mike on his frequent X Spaces, but haven’t connected with @DataRepublican yet.

I haven’t asked Mike yet how much exactly is his savvy vs. the AI usage, but I guarantee that figuring out these pathways would be almost impossible without AI.

The key to understanding this concept is understanding on the top level data homogeneity. That’s something people can grab onto. But how do we win if data is functionally the same, but in different formats? Or in different databases? This level of differentiation makes the task of following the breadcrumbs almost impossible for humans in a timely fashion. But it ‘s something an AI will make short work on. If you want to ask an AI how a penguin might be like a submarine, or what to do to make the penguin swim faster, well, good luck. If a human hasn’t answered that question somewhere on the web, you’ll likely get back garbage.

But monetary flows? That’s a different story. And that is exactly what is happening now. Which is why the institutional class is in stitches over DOGE and Trump.

Stay tuned. Elon said this a while ago — the distorted media landscape we’ve inherited is not only what is explicitly printed. It’s what has been left out. And that’s more than you can imagine. But AI can find it.