Tim J. Davey
I study measures that compress systems into something we can act on, without stripping them of the nuance which makes them reliable.
Our civilisation is complex, and small decisions cascade into large effects. Yet those decisions are based only on the information we have access to and, more importantly, what we can make sense of. Measurement is a powerful way of compressing our messy, illegible world into something we can concretely do something about.
But it's risky. In the late eighteenth century, foresters in Prussia and Saxony wanted to predict how much timber their forests would yield, so they started managing them as numbers. They cleared the undergrowth, carted away the fallen wood and replanted in tidy rows of a single fast-growing species, Norway spruce. The first harvest was a triumph, but by the second generation of trees the forests were failing. Stripped of the tangle of species, insects and soil life that never appeared in the tables, the trees sickened and growth fell by nearly a third. The Germans coined a word for it: Waldsterben, forest death.
This happens in messy ecosystems, but also in simple human interactions. Replacing a doorman with automatic doors removes a huge amount of cost, yet it also removes the added security, someone to hail your taxi and a friendly face to greet you.
Assuming a measure accurately represents everything that matters is dangerous. Yet measurement is still the foundation of all modern technological progress, so we shouldn't ignore it or throw it away. Instead we need to double down on the science of measurement itself.
This is all the more true as we get closer to recursive self-improvement. Once AI is improving itself, in theory "benchmarking is all you need" (or more realistically, all we'll have). So we need to get this right soon.
Incoherence · Entropy, 2024 · Paper
If you can't measure a system's complexity, you can't manage or diagnose it. So I published a new measure of complexity, based on information theory, which captures how well the average represents a typical instance of a system.
Cohesion · Entropy, 2023 · Paper
Some complex systems are easier to reason about than others. Here I proposed a measure which helps guide how you make decisions about when to collect more data vs simply preparing for the unknown.
Most people know me for the things I've helped build.
Founded 2018 · lis.ac.uk
A degree built around tackling complex problems rather than siloed disciplines. "The most radical new university to open in decades", the Times.
As you'd expect at first nobody got it. "Who needs more generalists?" Then just one year after launching, covid took over the world and demonstrated how brittle and interwoven our global system is. It moved complexity science into the mainstream and showed the value of system thinking.
A year after that, OpenAI released their first LLM, and the speed of development also proved our thesis that having high agency is fundamental to the future of what it means to be human.
The world is becoming more interconnected and complex. Specialisation in the workforce is becoming automated by AI, fast. Both of which mean we need more people who can stitch ideas together across boundaries to get stuff done.
Once the first two years of early ground work was complete, I moved into teaching Systems thinking and Data science. I went deep into creating the measures of complexity and self-organisation. The whole thesis of the university is about tying perspectives together, and nowhere is this more acute than at the boundary between quantitative and qualitative, creating measures for things which typically defy articulation.
Founded 2009 · The largest distributed hotel in the world
A tight knit organisation, in both senses.
The team was amazing and have gone on to create many more successful companies. But it was also tightly coordinated across a huge number of moving parts. To service that many homes we had to become one of the largest domestic cleaning businesses in London, one of the largest plumbing businesses, and so on. Then holding a consistent level of quality for guests, when the product itself was about staying in someone's unique home somewhere across five continents, was a very real world lesson in legibility.
We ended up selling the company for a healthy nine figure sum.
Founded 2007 · A recruitment ad network, and a very early YC batch
I was only 19 and it was long before any exits. What blows my mind is how wonderful that early hacker group of people were, and how basically everyone has gone on to do amazing things.
As I further my research, I'm building Concentric.
First it is a tool which makes sense of the work being done in your business. Organisations are complex: a small change can lead to a large outcome. For good and for bad, making these relationships legible will transform what these organisations are able to do. In particular, it'll enable teams to self-organise, adapt and coordinate at the speed they'll need in the future.