Jamie on Software

Links, August 2022

Richard Rhodes’s The Making of the Atomic Bomb was the highlight this month, a detailed and incredibly scientifically literate story of the agglomeration effects of genius and the construction of the atomic age. A long and detailed book but interesting ideas and stories on every page.

Continually impressed by Stripe Press and thoroughly enjoyed Where Is My Flying Car? by J. Storrs Hall. We need more techno-optimistic books. Hope and ambition are technologies too, and should be developed and promoted.

Also finished The Anarchy by William Dalrymple. Had the chance to chat to Dalrymple last month and his enthusiasm and perspicuity for this subject is astounding. The EIC was a violent and predictable and transformative institution, but it was at the same time path-dependent and its success subject to utterly unpredictable chance. Every time I read history I’m taken aback by the fragility of the world we’ve inherited.

I wrote a short post on state-adjacent institutions.

The US approved a small nuclear reactor design.

Noah Smith on CHIPS and chips, and four reasons why GDP is a useful number.

Vision papers in science.

Regress Studies on how to choose books and why you should allow yourself to be dragged into books you suspect you won’t finish.

An older thread from Arnaud on books on the history of modern computing.

The best of Scott Aaronson.

From Knowledge spillovers in Silicon Valley:

Our IV approach estimates substantial returns to face-to-face meetings with overidentification tests suggesting we are capturing the returns to serendipity that play a central role in the urban theories of Jane Jacobs.

Types of barcodes

Some good Visa threads on a cluster of related ideas: do what you want to do, do it a lot, and tell others to do what they want to do. I had a great chat with Visa at Future Forum, and his temper and mode and writing is lovely in a happy-go-lucky, auto-optimistic kind of way. His blog is also excellent.

Nuño Sempere on decomposing quantitative problems into simpler questions.

2:43pm. September 1, 2022.

State-Adjacent Institutions

From Dominic Cummings’s recent ‘Snippets 5’:

NSN searched for people who wanted to start new schools that we intended to allow under new legislation then worked with them on their application.

It developed specific ideas for how the system should work. Unlike normal think tanks, it worked at the extreme detail end of the spectrum, like drafting actual forms that the DfE could use.

We thought that if we helped people develop specific projects and could actually bend the bureaucracy, we could get ‘Free Schools’ going.

This is interesting, because it suggests a model for state-adjacent institutions that support the functions of the state without being tied to the bureaucracy or subject to the incentives. Something somewhere in the midpoint between a management consultancy, a think tank, and a training institute.

It seems like this sort of model is beneficial in specific circumstances:

  • Where there is considerable state funding for a certain industry / project / goal, where ‘considerable’ is relative to the industry / project / goal. (You need more money to make a new nuclear power plant design than you do a school.)
  • Where access to that funding is reasonably competitive. This means it needs to be accessible to SMEs or small teams, rather than just Serco, G4S, the major universities, and the other big boys.
  • Where the government department distributing the funding is receptive to introductions, advice, support. A lot of the benefits seem to come from new ideas being introduced revisited through a feedback loop, so the department must be flexible enough to integrate the state-adjacent institution into its decision making – or, at least, leave its door open to such input. Generally, I suspect, departments are only really flexible in this sort of way during their first few years and for a year or two after a major shake-up (eg DoE under Gove).

A few ideas where this could be useful:

  • An ‘ARIA Attractor’: finding individuals / teams with audacious projects, helping them refine their proposals, and using the knowledge gained from the search to help shape the department’s broader research focus.
  • An analogous attractor for the UKRI / NIH / other ‘normal science’ funding bodies, with an emphasis on speeding up the funding application process.
  • Standards setting and guidance for navigating the planning system, but beyond a simple planning consultancy. A team of people who work hard to befriend and effectively lobby local planning teams, giving developers large and small the opportunity to boot up to the activation energy required to get anything built. It would be useful to focus on local authorities with a recent change in governance, or with an especially acute housing problem, where the demand to do something is greater.
  • A company to abstract away the medical licensing and compliance process and help guide new teams / University spin-offs to apply for medical licenses. Medical software, in particular, is notoriously hard to get adequate licensing for. A team dedicated to building solutions and increasing governmental awareness of these issues, working with funding bodies and NHS clinical commissioning groups, could meaningfully move the needle on what innovation we permit in the medical space (Scarlet are doing interesting things adjacent to this idea.)

Creating and supporting these sorts of state-adjacent institutions could be an important focus area for a movement like Effective Altruism.

But it could also be a powerful new framing for start-ups themselves: many of these ideas could grow into profitable businesses as well as institutions with positive-sum state/business interactions.

Asking the question “how can the private or non-profit sector smooth over the relationship between the state and others?” is a subtly different question – and one with (in many cases perhaps) more leverage than the simpler question “how can the state do things more effectively?”

Why? Because it refocuses the conversation on what can be done practically, today, without needing large-scale systemic reform, without needing to push through the ossified systems of government purchasing and permits that sit in deeply suboptimal equilibria.

2:59pm. August 27, 2022.

Books I Wish Existed

For a while I’ve been collecting a list of interesting book-length projects. Since I’m never going to write the bloody things, I figured it’d be better to throw them into the light of day and see what daylight makes of them. Here is a preliminary, first stab at a list of books I wish existed: books that haven’t been written yet, but could be.

A collection of short books on the history and interpretation of U.S. constitutional amendments

One under-explored feature of the US constitution is its deep cultural, as well as judicial, role in modern American politics. Each amendment has its own motivations, historical context, and judicial precedent; but each amendment also serves as the starting point for contemporary arguments for or against certain policies: even, today, the policies of private companies.

I’d like to see a series of short, concise, focussed books – think roughly the length and depth of the OUP Very Short Introductions series – with each volume focussed around an amendment to the U.S. constitution. Each book could discuss the amendment’s historical context, important cases in its subsequent judicial precedent, and the moral and legal and institutional justifications for the amendment and how they have changed.

Most amendments would have their own volume, while some of the more arcane amendments might be bundled together, where appropriate. The 18th & 21st are a natural pairing; the 13th, 14th and 15th sit snugly together in terms of their shared historical context, but are perhaps each significant enough, with their own rich set of continuing precedent and relevance, for their own volumes; perhaps the 3rd, 4th and 5th. It might also be interesting to conclude the series with a volume on the amendments that didn’t get passed: amongst many others, the ERA, balanced budget amendments, the We the People amendment.

Seeing Like A Startup

Scott’s Seeing Like A State is an excellent piece of political epistemology, not because he makes a powerful moral argument to curtail the absolute power of the state – which he does – nor because of the trenchant analysis he applies to the material and sociopolitical conditions under which the tools of statecraft are likely to be abused – which he gives – but because it grounds the analysis in a Weltanschauung, an all-encompassing frame of reference, a set of spectacles that underpin the identity of those who wear them. To see like a state is not just to see the world a certain way, to plan with a specific framework, to write with a specific dictionary, but also to be somebody.

Isn’t the same true of startup-land? Isn’t working in a startup with its techno-optimism and its studied disregard of conventional wisdom and Disruption with a Capital D a form of world-view? Weren’t we decades ahead on remote work and Agile / Lean Startup approaches to product development? Don’t startups, especially tech startups, have a distinctive set of incentives and respond to a distinctive set of internal and external cues? Isn’t this weird (physical or virtual) Bay Area we inhabit a conduit for a specific mode of thought, a Weltanschauung, a pair of spectacles?

A full-length biography of Évariste Galois

Évariste Galois died aged 20, after being shot in the stomach with a pistol. He died a gregarious yet unlikable, angry young man, but he bequeathed us a small elliptic body of mathematical work that has proven to be incredibly fertile.

The short biographies that accompany discussions of his work are useful and evocative, but focus almost exclusively on either his precociousness, or the Potemkin-romanticism of his death. His life was short but full of activity, sadness, anger, intense adolescence, mental illness and revolutionary politics.

The best biography of him so far (fr) focusses on Galois-as-mathematical-figure (‘personne’ vs ‘personnage’). I’d like to see a full-length biography of Galois-as-boy and Galois-as-man, as well as Galois-as-mathematician: something that draws out the dynamics of a Republican and Bonapartite household in restoration Paris, the stability of his mother and bipolarity of his father (who himself committed suicide when Évariste was 15), the friends and foes, real and imagined, that shaped this troubled young boy.

I’ve been trying to write this book for a while, but have put the project on hold. Perhaps I’ll resurrect it one day.

Uses and abuses of popular science

The effective communication of science is incredibly important. What the electorate understands and values about scientific output can translate meaningfully into policy outcomes, on the one hand, and our continued ability to discover more about the world on the other. (At its limit, it can cause deadly incentives failures when the scientific bureaucracy needs to reengage a science-saturated public).  Simplifying without talking down is a tough job, and the very best writers do it with elegance and wit and humanity. But so much of it is reductionist, factually incorrect, statistically ignorant, sensationalist drivel.

Writing about science poorly harms us all. Being excessively confident about scientists’ predictions – “toast causes cancer!” – shifts our focus onto the wrong things, or erodes trust in the output of science when it turns out that, you know, the world might be a little more complicated. Being excessively cynical about science’s output is so often a tiresome postmodern ploy to import political solutions to yet-understood social issues.

I’d like to see a book on popular science and the popularisation of science: what good it can do when it’s good, what harm it can do when it’s bad, and how we can get more of the former and less of the latter. I’d also like to learn more about how science fiction fits into all this. We will never get to a stage where science is not weaponised in one direction or another – discovery is, as the physicists of the Manhattan Project discovered, the beginning of the moral story, not the end of it – but with a better understanding of how science is reported, we might be able to give people the tools to at least discount the views of the most egregious of offenders.

What could science look like?

The way that modern science is structured – the categories and classifications of physics, biology, chemistry, computer science, mathematics, philosophy, the social sciences – forms a reasonably arbitrary and path-dependent structure. A few changes in how humans organised themselves at various stages, how projects got funded, and which questions happened to be salient (for cultural or contingent material reasons), and we have a very different body of knowledge, structured along different lines, today. What, for instance, would modern AI look like if the centre of gravity in computer science hadn’t drifted away from cybernetics and the HCI-focussed research tradition during the ARPA golden years, and toward applied mathematics and algorithm design? What could biology look like if our best mathematicians were more interested in biological systems rather than physical systems? What would Newton have done if he hadn’t spent so much time pursuing alchemy?

A good moral, economic and psychological investigation into paternalism.

I have a set of libertarian-ish (which is to say, mostly negative) aesthetic reactions to paternalism, and, in a trivial sense, ‘paternalism is bad’ seems true by definition – at least on a normative reading of ‘paternalism’. Naturally, these intuitions have come into much sharper focus throughout the pandemic. But state interventions in private lives are nothing new, in many cases they are basically uncontroversial (e.g. seat belts) and there are a whole host of moral and economic arguments in favour as well as against.

Perhaps paternalistic reasoning is our default mode of thought and respect for individual freedom only gets bolted on in certain contexts? If you really believe that such-and-such a lifestyle is immoral, harmful materially and spiritually to he who practices it, why wouldn’t you want to intervene? Liberalism is a position most have to contort themselves into. I’d like to see a modern book-length treatment of this subject, exploring the changing relationship between individual and society, ideally within a framework that make sense of big data, the long death of privacy, and crypto- or techno-libertarianism.

Aesthetics in politics.

Hume never got to finish his ‘examination of morals, politics, and criticism’, but if he had, I imagine much of the project would be spent grounding political discourse in terms of human sentiments like approval and disgust. Jonathan Haidt offers a modern-day version of this story, arguing for the centrality of psychological states in understanding politics and religious discourse.

But one thing that often gets ignored, I think, is how much aesthetics play a part. People find views they dislike not just disagreeable but ugly, and often detached logical reasoning takes a backseat to matters of taste. I’d wager that a lot of opposition to virtue signalling, for instance, is simply that it seems distasteful, or uncouth, or something like that.

To what extent do we elevate matters of taste to matters of shared social importance? (There’s an interesting Twitter thread here on conservatism and aesthetic sense, which might begin to address these issues.)

The House of Uncommons: the rise and fall of excellence in politics

Why has politics lost its cultural cachet? Why do we pay our governors so little relative to other, arguably better run, countries – and certainly less than a lot of private-sector high-status jobs? Are our politicians getting more incompetent and panderous, as they do indeed seem to be? When was the golden era of the politician? What characteristics should we try to select for? Given the unpredictability of democracy and the epistemic credentials of the average voter, how can we reshape our institutions to better encourage the selection of these characteristics?

The Aesthetics of Programming Languages

One thing that often gets lost amongst the computer science jargon and expediencies of writing functional software is that there’s an important aesthetic dimension to programming, a concern with the beauty of the code and algorithms we write. We throw around words like ‘beautiful’ when we talk about code, but we’re usually just gesturing toward some muddy intuitive notion, something like ‘clean’. There’s been little attempt to define these words more rigorously, or explore other aesthetic or aesthetic-adjacent virtues, such as simplicity, or parsimony.

It’s not merely syntactic, either. Much of what a programmer does is invent abstractions, extract out pieces of a system into reusable and more generic chunks. Some abstractions are intuitively better than others. But on what grounds? It’s not just “how widely applicable is this thing”, or “how performant is this thing”, or “how few lines of code does this thing take to implement or call”. There’s a notion of expressivity, the capacity for the abstraction to open and close the right set of logical doors, that is crucially important, and, crucially, misunderstood.

It runs deeper than just the code that actually gets written. Different language design decisions force us to think about our code in different ways, and to structure our programmes along different fault lines. Type systems force us to think about our domain before we think about the processes we apply in that domain. Pure functional languages force us to think about the flow and transformation of data. Different languages, sometimes, though by no means always, designed for different tasks, start with different mental primitives which change both how we write code now, and how the norms of the broader language ecosystem evolve.

The great irony of programming: instructing computers can be a deeply human thing. It would be fun to see a thoughtful little book exploring these questions in more detail.

4:16pm. May 16, 2021.

Why I’m long on Visa

Last month, Visa announced their intention to buy Plaid for $5.3 billion dollars.

Plaid provide an abstraction around bank accounts. They offer a developer-friendly API to query one or more of a user’s accounts. This allows startups and consumer products to offer financial analysis and services in a bank-agnostic way.

On the back of this purchase, and alongside my general growing interest in fintech, I’ve been becoming more and more bullish about Visa.

But why did Visa pay so much for Plaid?

$5.3b is a lot of money. It’s somewhere between a 25x and 50x multiple of Plaid’s revenue.

The Plaid team are impressive, but at $5.3b, it’s got to be something strategic. As my friend Rich put it, when you’re counting in billions, it’s not an acquihire.

So if it’s not revenue, and it’s not (just) the team, what’s the strategic value in Visa’s owning Plaid?

I can think of three big reasons:

1. It solidifies Visa’s core business.

Visa is a three-sided network. It provides the infrastructure to move money between consumers, merchants, and banks.

When Visa works well, everyone benefits:

Consumers get instant access to credit, and can buy products from anywhere the card is supported.

Merchants can accept payments from anyone, and no longer need to run back-office operations responsible for credit and payments, nor handle cash.

Banks can offer credit to consumers more easily, at higher interest rates, and collect fees from merchants to provide the aforementioned credit management services – and, in doing so, reduce merchants’ exposure to credit risk.

For arranging this service, Visa charge merchants and banks a percentage of each transaction. The underlying economics of the business are excellent: they have over a 50% profit margin, steady revenues, a long pedigree, and revenue still grows 10% year-on-year.

Plaid can contribute to this core business, since an increase in fintech innovation is likely to increase transactions simpliciter:

Banks can offer more, and better tailored, financial products to consumers. Visa can better integrate identity and security services with payments, reducing rates of fraud. More broadly, and more importantly, fintech can, and will, bring finance to the under- and unbanked.

Thus, Visa can leverage Plaid to shore up the existing network, make being a Visa customer more attractive, and create whole new demographics of customers in developing markets.

All of this means more transactions, and more transactions means more transaction fees.

2. Plaid is a good business in its own right.

A 25x to 30x multiple on revenue is large, but Plaid are still a relatively young company.

Plaid offers Visa the opportunity to take fees from the “other side” of the consumer’s relationship with the bank: in GraphQL terms, the queries against the bank account rather than the mutations on it. This is somewhere Visa currently aren’t able to capture value.

Plaid’s target customers are software engineers, but the institutions at the bottleneck to Plaid’s growth are banks. Now Plaid has the weight – and half-century of personal relationships – of Visa behind it, the product itself can grow substantially more.

Plaid are also, so far, focussed heavily on the U.S. Fintech opportunities are global, so now Plaid can lean on Visa’s global reach to expand internationally.

All this said, Plaid could represent a significant income stream for Visa in its own right.

3. Plaid is part of a broader Cambrian explosion in fintech.

Finally, this purchase reflects a wider trend.

As Stripe, Twilio, Algolia, and now Plaid have shown, making developers happy is big business.

But making developers happy and productive also has serious downstream effects. It reduces the amount of time and money it takes to create new products. It encourages the development of new tools which themselves make developers happy and productive, effecting a Cambrian explosion of new products and tools.

Just like the set of norms and tools developed around open source, and composability in smart-contract technologies, making it easy, cheap, and secure for developers to access financial data will only accelerate the potential of technology –– and, crucially, along dimensions we can’t easily predict.

In short: it increases the amount of innovation possible in the world.

If I am at all correct about this, then Visa have shorn up their excellent business model, giving themselves access to an attractive new revenue stream, and now control a resource which has already, and will continue to be, central to sustaining and growing the fintech space, and, eventually, the global economy.

Thanks to Jessica Cooper, Richard Burton, Peter King, JS Denain, and Jonny Corrie for their notes and comments.