India : US :: Spreadsheet : Webapp

Have you read Seeing Like a State? The thesis of the book is relevant to what you’re talking about. I forget who the narrator is, but he’s one of the top 3 sci-fi narrators, so I immediately felt comfortable in the audiobook like if my dad was reading to me.

I found the book to be thought-provoking and educational.

The basic thesis of the book is that from around the mid-century an idea became ripe around the world that it was time for scientific state governance. Many attempts were made around the world to make government “scientific” in a way that only a bureaucrat could totally miss the point. Systems were made more rigid, cities and forests were planned, data was collected, etc. Generally the balance of control shifted from the leaves to the trunk.

The book takes a neutral tone and tries to be matter-of-fact about pros and cons of each attempt. There were some occasional pros. But of course it was almost total failure. The problem wasn’t the idea per se, because in most cases it’s possible to imagine it going well in an ideal world. But the problem was execution: no one could afford the necessary level of data collection, no one could afford expensive central decision-making. Bureaucrats pretended the world was simple instead of doing the hard work required to make the centralization succeed.

So I think it’s easy to misread the moral of the book. The obvious conclusion is that centralization has failed repeatedly to the extent that bureaucrats have given up. The actual moral is that those attempts were too early, but now that the world runs on powerful computers, especially with AI in the mix, there will be a new round of attempts and many of the reasons they failed last time are no longer relevant.

My approach with MixRank is the India way by default.

I start with minimal process, since that’s easy and fluid when there’s rapid change. Then over time things mature and the flaws of the lack of process become apparent. Then I think carefully and decide on what’s the minimal change I can make to formalize the process to address the issues I’m observing.

Assuming it’s successful, I forget about it and move on to something else until it breaks again. All the processes are evolutionary and only make sense for the given point in time. I can’t skip to the final version because the company isn’t ready for that version yet. Give it enough time though and it will reluctantly evolve from India to US.

“Good abstractions are hard” seems like such an unsatisfying conclusion.

Good abstractions are indeed really hard. Bad abstractions are easy. And bad abstractions are disastrous compared to no abstraction. I figuratively clench my teeth whenever I’m about to build an abstraction because I’m about to make a mistake with leverage. I have to be really sure I’m doing the right thing to pull forward with it, otherwise I’ll back out cautiously.

Please go ahead. Feel free to edit in any way you please to fit the blog.

Chip War → Seeing Like a State is a good topical transition.

I’m not convinced that a complex adaptive system can be steered even with perfect data collection ability.

We have an existence proof of the human body. It has centralized control and sensory apparatus spread throughout. It one sense it’s vastly more complex than a state. Though in another sense it’s less complex because a state includes at least one human.

We can make a weaker argument that’s easy to agree with though. There’s benefits to some centralization. A tree structure is the most efficient for propagating a signal. A state with no central control is easier to conquer. Etc. But there’s limits to the effectiveness of centralization: size vs latency, lack of information to act on, lack of processing ability. If you have lower latency communication, more extensive information to act on, and greater processing ability — we have all three in fantastic abundance since mid-century! — then the equilibrium shifts further towards centralization.

We’ve discussed “Markets are efficient if and only if P = NP”.

Definitely relevant and a great point. But a stock trader doesn’t necessarily care about a math proof if their probabilistic model is good enough to make money each day. In biology, evolution has never cared about perfection, just fitness. And for better and worse, centralization has significant advantages to out-compete decentralization.

Can you centrally plan groundbreaking inventions?

China 😬

ChatGPT is centrally planned, and I’m sure you’d agree it’s come up with novel ideas.

I agree with the point you’re trying to make. Centralization is bad for innovation. New ideas get suppressed and the organization doesn’t keep up with its peers. It’s prone to cancer (literal and figurative).

The only reason centralization works in biology is that each organism is disposable and that there’s a large population of them.

Good point.

Inference is greedy, so it’s unlikely to produce a sequence that starts unlikely but then finds a far-off local optimum.

Is it capable of validating that a local optimum was found though? The answer isn’t obvious to me.

I believe I shared a youtube lecture with you years back. The question was why deep learning works when the search space is exponential? The answer was that the equations that govern physics are low degree polynomials, and when you compose low degree polynomials you get low degree polynomials, and all the stuff we observe is the output of this physical model.

Since all the physics equations we discover are so short and sweet, can’t we brute force search physics?

I remember sharing it nine years ago 🙂

It’s rare I get to do a memory flex on you so I’m cherishing this moment

Very true! I agree and I’ve seen that too.

Back when I was working in China it was like that sometimes too.

People had memories which were too good, and it reduced the pressure on them to really fix and standardize things.

I think the slack in unoptimized processes is useful for workers since it increases their personal power. Clerks suddenly gain much more power and status since they can kill applications or advance them using inside knowledge rather than being a cog in the machine. That’s the cynical view.

But people also sometimes think weakening institutions is actually good since it means if the leadership turns evil, they can just stop all progress without having to actually win an election to dissolve it.

There are lots of positive versions too. People like to feel useful (graveyards are full of indispensable men and software).

They often just don’t realize the costs. Benefits to speed are maybe more distributed so harder to measure and promote in weak institutions?

I think riding the edge between automatic process and conscious decision is the actual reason consciousness exists at all. We literally are here specifically to work on and use agency to handle problems which can’t be autosolved. So, it may always be that our lives are always full of difficult choices. Since easy ones get optimized out into subconsciousness.

So, I’d probably say that it’s case by case whether any given process should be standardized by the current staff at the present time.