Ryan Baker

Karma: 0

I’m a systems thinker. It started with software. Old enough to have had an TRS-80. Built a small-town startup out of my first programming job at 15 (1993). Later my interest in systems escaped the computer world and has reached economics, philosophy and philanthropy. So, that’s how I got here.

These days, my first outlet is my Substack. Perhaps unsurprisingly, AI has become a central topic. My years in software took me into architect roles, with security, reliability and development processes being core there, though you’d probably call me a generalist as I can’t resist learning about everything.

I see policy as very important, but I see a big gap between the public and the internals of the IT industry, much less the AI industry. I first became interested in EA when I heard about it from some local Chicago groups (2010′s). Climate change would have been top of my mind at that time.

In the background, I’m very interested in global development. I didn’t get concerned about climate change because I’m an environmentalist. The typical environmentalist has a much deeper prioritization of nature than I do. I respect that view, but my path was always that humanity depends upon our environment, and it’s inescapable that we have to take care of it, so it can take care of us. Mostly those views end up at the same place in terms of actions, but they do involve different ways about talking about the why of things.

Ryan Baker 6 Oct 2026 17:17 UTC
1 point
0
on: Post-AGI Economics As If Nothing Ever Happens
It’s a brave project. I agree that the narrow views are deceiving. I wonder how helpful economics can be in understanding a post-AGI world. It can extrapolate from some initial conditions, or at least make a good effort at that. But the initial conditions are uncertain enough that you have to maintain many models that have lost correlation with each other. When those scenarios have lost their correlation, you can’t average them, or smooth them, you simply hold them apart. I’d say you’re gaining knowledge in the pursuit, but will it ever be useful knowledge?
If you pick one scenario, you can start to act as you believe it. But that’s not very useful if it’s one of many and the pick was arbitrary. You probably cannot find any compromise when correlation is absent. It’s the same problem you have with all the narrow views. They vary one condition, and hold others constant, when it’s unlikely anything holds constant. Would you average all of their conclusions? You really can’t. Moving from 1 dimensional scenarios to multi-dimensional scenarios doesn’t reduce the intractableness of the problem.

Ryan Baker 6 Oct 2026 15:03 UTC
1 point
0
on: Expecting Short Inferential Distances
I think this is in the right territory on the difficulty of communication. Have some doubts about the ancestral link.
Two counterexamples in the ancestral environment. Children and secret roles. Children still need to be explained and start out lower. But maybe this doesn’t change things much as the order of presentation to children can be somewhat fixed, introducing one inferential step at a time. In many ways, we still use that. “Educating” has always seemed a misnomer to me, as educated and education are imperfectly connected. We become educated by a self-act, and teachers at best arrange concepts in a convenient order to be consumed. It probably made more sense before books.
Secret roles, like shaman, sometimes chiefs, would have also afforded this path.
What’s hard about “the public”, is they aren’t on any path. You don’t know where they are. The obvious options then would seem to be start at the beginning or the end. Both would fail for the reasons highlighted. At the end, will miss some connection, which fails to communicate. Unless the audience decides to build the necessary connections, the communication is ineffective. It’s also alienating. Either you appear as disconnected, or superior. The first is worse than the second, but both are damaging.
If you start at the beginning, you’ll undoubtedly repeat things some or maybe all of the audience knows. Here you look insulting, presuming that there is no knowledge. Plus, you’re wasting their time.
I think the tendency tends to be to start at the end, because given two bad options, might as well do the one that’s less work. I think that’s probably a bigger influence than anything about ancestral environments.

Ryan Baker 6 Oct 2026 13:56 UTC
1 point
0
on: Say Not “Complexity”
If you feel that complexity or emergence is being misused, to avoid examining at depth, then the substitution of magic is a useful psychological trick to substitute “magic”. But magic is less descriptive. Emergence or complexity are definitively a different things. They do have descriptive power.
What I expect is the key point here, is that the gap between magic and emergence in terms of descriptive power is far less than the gap between emergence and fully explaining anything that resides in the interior of the process of emergence. In the right context, using that gap to avoid pursuing understanding you’re not ready to pursue, is the right move. But when it becomes a pattern that ignores there is another context, that the interior is describable, or at least may be, it becomes a blocker.
If it is used to allow you to intentionally turn your focus elsewhere, it’s useful. But when it forces you to turn away, to, to lose interest, to lose hope, of a deeper explanation, this is where it becomes an impediment. Emergence or complexity state our acceptance of our ignorance. They give us license to retain that ignorance. Useful in the moment, but a loss as habit. In this sense, this is exactly what we worry about, or even ridicule, under the phrase magical thinking.

Ryan Baker 3 Oct 2026 22:11 UTC
1 point
0
on: The Twelve Virtues of Rationality
Surrender to the truth as quickly as you can.
I find myself wanting to improve upon this statement. It’s not easy, but it seems desirable. Surrender strikes me as too strong. Constructing defenses against evidence is the failure. You also want to be an active participant in removing resistance. If you’ve built a fortress around an idea, and discover the foundations are unsound, you should be an active participant in deconstructing the defenses built on the unsound foundations. You should not force others to remove those structures for you, nor should you attempt to repair the gaps.
Surrender implies this, but it implies more. You have to recognize truth to surrender to it. If you stand in a fortress and a wall falls, you should wonder about the foundation, but can you assume that without inspection? If you don’t know the state of the foundation, you do not surrender prematurely.

Ryan Baker 25 Sep 2026 18:09 UTC
1 point
0
on: Swarm Scaling
With this in mind, it’s reasonable to expect that once you’ve proved in 88 hours the ability for a 10,000 agent swarm to produce a result, your next step would not be to spawn another 10,000 agent swarm to solve one new problem 3 days later, but instead to spawn 100, 100-agent swarms to produce 100 results of similar complexity and importance in 30 days. I guess we’ll see in time if this bet is placed.
You could also do 10,000 in 1 year, but then you’d have the problem of identifying 10,000 candidates, and the model scaling would reset the accumulated value midway, so doubt we should wait for that version.
I wonder to what degree this trend line continues, as it would seem likely given the current newness of swarms that there is room to optimize inter-agent communications in a way that’s beneficial to scaling.