Why Information Grows - César Hidalgo

I was talking about... I think how to scale teams and complicated systems, to an ex-colleague of mine, and they suggested I read Why Information Grows. So I did. It's... an interesting book.

In the end, the book is about how we build the complicated systems of our economy, through the lens of accumulating and managing information, based on the groundwork of information theory. It sits in the pop-sci/my pet economic theory cross-over space, if you can imagine such a thing. Before going any further, I find it a little pretentious, and somewhat less deep than it thinks it is. With that said, let's dive in a little further:

The start of the book is pop-sci - information theory, entropy and all that stuff. You can tell that they deliberately avoided putting any equations in, presumably to avoid denting the sales figures, since there are so many simple, quantitatively well-defined concepts that aren't elucidated. Entropy is left as magic, not p log p.

There is a gem in this section, though: The work of Prigogine, who explains why it is, if entropy is increasing, that we get bubbles or order, like our world. Out-of-equilibrium systems transiently generate order on the road to entropy. Given "entropy always increases" clashes with our everyday experience, I'm surprised that I had no idea about this work, and that it's not more well known.

After doing the pop-sci bit, he jumps to explaining the economy in terms of knowledge and know-how, the core of the economy as handling information. Physical objects are described as crystallized imagination, as if this is not a pretentious way to say we design things.

A knowledge-based view of the economy is very powerful. It allows you to distinguish between hard-to-produce goods like microelectronics and basic ore extraction in a way that traditional economics glosses over. There's also a certain amount of standing-on-the-shoulders-of-giants knowledge-building-on-knowledge stuff, rather undermined by their Hot Take on resource extraction. The traditional view is that taking resources to build things from poor countries is exploitation, César's view is that those minerals weren't valuable until scientists discovered their usefulness, and hence those countries are actually exploiting the scientists!

Apart from the very dodgy analysis of what exploitation is, I felt this section missed a bunch of tricks. No analysis was done of the resource curse in this framework. Knowledge/know-how is a form of capital, but since he's avoiding the "capital" framing, it's not compared with other forms of capital, so the relative importance of know-how embedded in manufacturing equipment as opposed to experts is never analysed.

Nor is the idea, which I think is very important, of the ease of manufacturing - ahem, crystalizing imagination - in all this. Making fine-featured microchips requires insanely complicated machines, of which there are only a few in the world, and of which the manufacture is itself a great engineering challenge. On the other hand, software, once written, can be copied without effort. There's a reason "software is eating the world". However, there's no discussion of all that.

So, while the author disparages economics for blurring economics for blurring all capital together, important structures in knowledge are blurred over. Oh well.

There's a nice chapter on "Amplifiers", discussing how, through specialisation, we can become experts, and the power that gives us - for example, a guitarist need not know how to make a guitar. I was disappointed that this chapter didn't talk about abstraction. In information systems like software and maths, abstraction is the tool to allow you to concentrate on a single area without having to think about the whole stack simultaneously. It's how we make complex things tractable. The amplifiers concept is abstraction in the world of goods, and not having the information concept of "abstraction" identified in the economic domain was disappointing.

The next step is to the idea that real-world know-how and knowledge is stored in people, and given a human's finite capacity, complicated systems need to store that across many people linked together. This fairly obvious idea is spun by calling the amount one person can know a "personbyte". It is then pointed out that interaction costs are substantial, which is of no surprise whatsoever to anyone who's worked in a large team. Much of the book focuses on the geographic ramifications of knowledge residing in physically co-located networks of people.

The step beyond that is to make the dubious claim that there's a similar limit on the size of a corporation, tied to Coase's theorem. This, I don't believe. There's no reason external interaction costs must be lower than internal interaction costs, at some scale. Personally, I believe the best reason for separate companies is economies of scale across multiple customers. A company that supplies products to multiple companies is going to achieve economies of scale not achievable to a company that only supplies a single customer, while a company supplying a single customer might as well just vertically integrate. Put another way, companies allow the customer graph to be a DAG.

The key point is that the best way to scale links is through abstraction. If I work with other people, the less I need to know about the inner workings of their world, the more people I can work with and the more I can concentrate on my own domain. Separating companies forces an abstraction, but good management can achieve that within a company, too. I can see within my workplace the combination of deliberate abstractions and deliberate network-building (with near and far links) as a way to create an extremely large but scalable company.

Many of these abstractions are effectively interfaces. The book identifies standards as a way of reducing the cost of links; these are interfaces. As a physicist rather than a computer scientist, he nevers says "interface".

The tail end of the chapter whines a bit about the inefficiency of US government contracting and the healthcare system, spending so much of the knowledge capacity in what is effectively link management (aka bureaucracy). For someone who is supposed to be looking at the big picture, this is a complete lack of Systems Thinking ("the purpose of a system is what it does"). As someone so focused on the production of crystallized imagination, he assumes the purpose of all organisations is to do that. It's convenient that many organisations align so well with doing that, and by cherry-picking the organisations, it's easy to pretend that's what they're all for. The "inefficient" organisations are only so if you mistake what they're trying to achieve.

In short, I found the whole "firmbyte" concept of firm size limitation and associated discussion pretty deeply disappointing.

The next section talks about the relationship between social networks and network-building and the role of trust in society. The idea of Silicon Valley winning out over the Boston tech centre due to more openness is a nice and concrete (if not terribly well-supported) example of how these personal network effects can affect the viability of a corporate ecosystem. Even more compelling are the stats on job-finding through social networks - networking is vital. It also demonstrates the challenges in building a diverse workforce, given the lack of diversity in many social networks.

For me, one of the big questions the book has failed to answer is "How vital is the human-held knowledge/know-how, relative to the physical capital?". I guess this is roughly akin to the traditional question of returns on capital vs. returns on labour, except the labour here is the expertise of knowledge workers, not the stereotyped manual labour. The book does give a great example of a network transport success, in the form of von Braun's team being transferred to the US post-WW2, but on the other hand they had access to almost unlimited resources. Physical resources had quite the hand in getting man to the moon!

By focusing so hard on knowledge and know-how, its place is lost in the wider picture, which is a shame, since I think the arguments for human knowledge/know-how being the most limiting factor in building advanced economic capacity are quite strong. Witness various governments' failed plans to build tech hubs by not understanding the human element and need for specific expertise - throwing money and failing. There are also myriad interesting examples to look at. Where do physical-equipment-light sectors like banking end up, compared to those that perform physical engineering (looking at you, Germany)? How did Shenzhen bootstrap itself to its current world-leading position? What do we see happening in internet-mediated communities, like open-source software? There's so much here, where empirical data would really help flesh out the ideas.

The latter part of the book does, finally, tie into traditional economics and theory of capital. This bit is actually quite good - it makes the case for not analysing the economy in aggregate, but breaking it out by specialisation, and analysing rare industries that encompass a high level of specialisation separately, which represent high-knowledge areas with increased economic value. By taking this more subtle approach to the economics of countries, more accurate growth modeling can be achieved. This whole chapter feels like Hidalgo's research papers have been simplified into book form, and compared to much of the rest of the book, this rigour works.

The end of the book compares companies and creatures, economies and ecosystems. The author notes that plants and animals are good at encapsulating all their knowledge/know-how for reproduction - for example, a tree is able to grow from a seed, in a way that companies cannot. On the other hand, a whole ecosystem is a set of flore and fauna in equilibrium, and represents a set-up that is hard to transfer, just as an economy is hard to build.

This analysis once again lacks systems thinking. The purpose of a rabbit is to make more rabbits. The purpose of a tennis ball factory is to make tennis balls. To make a tennis ball factory, you buy machines from other factories that in turn were assembled from components of other factories. The economy is a graph of resources that create other resources, while nature (modulo symbiotic relations like pollination etc.) is about things that directly produce copies of themselves. It is therefore zero surprise that our economy has no equivalent of the seed.

Looking at the length of this review, there's a fair old amount in this short book (180 pages before you hit the volumous and most grating acknowledgements section). Yet it feels incredibly shallow. The pop-sci section at the start both wastes an opportunity by avoiding grounding the concepts in maths, and is just plain unnecessary for the economics part of the book. The economics part spends a lot of time belabouring the obvious in pretentious terms, misses the opportunity for deeper thinking, and rarely ties the discussion to quantitative data, or even significant/compelling case studies. In the places where it does do so, mostly off the back of Hidalgo's published research, it's good. The downside is that this highlights the flaws in the rest of the book.

Going into the book, I was hoping for some insight into how to scale knowledge at the limits of human-managable complex systems where you're hitting the personbytes limits (*cough*, my team's job, *cough*), and I was disappointed that it didn't deliver there, but it also didn't deliver in so many other ways. Interesting, but a wasted opportunity.

Posted 2021-06-19.