I think a lot about lines, and most of life is lines (queues, if you’re from the UK). They are everywhere: merging onto a freeway, on-phone hold times, why a laptop is so hot, chatbots at online retailers, weekend dim sum, a ski lift will open on a powder morning, and so on. These are all lines, with many similar and important properties.
It is less well understood than it should be, but one of the most important developments of the last few decades is that of queuing theory, a body of research that allows us to engineer systems to deal with stuff showing up, stuff waiting for service, and then said stuff leaving. The “stuff”, of course, can be human, can be cars, can be packets on a network — it can be many things — but they’re all often in queues. And like so many things at which we can throw algorithms, the cost of engineering a queue has collapsed in recent decades: we know more, and we can manage them less expensively. To a first approximation, queues are cheap and everywhere.
An example will help. Most queues can be characterized via two parameters: arrival times and service times; the rate at which things show up in the queue, and the speed with the queue is processed. Both of these parameters have distributions. The simplest version comes when people (let’s use that example) show up a constant, predictable rate, and when service rates are also fixed. Most real-life scenarios aren’t like that, of course, and both arrival rates and service rates vary wildly, but can be assigned distributions, like the exponential, that make the problem tractable.
So, here is a simple example. Both arrival and service times are exponentials, and I’ve set them equal. What’s interesting inthis simple example is how queues build and dissipate, even though the arrival and service rates are equal. All it takes is a few complex cases, or a few extra people showing up — both of which are predictable given the underlying distribution -= and suddenly people are waiting longer than expected.
Things get much messier with no underlying parameter changes. In the following case, lines ballooned, even though the arrival rate and the service rate are still the same.
What’s interesting about queuing theory is how useful it is, and how easily you can model real-world situations in ways that make important problems kinda go away. But the problems haven’t really gone away. All it takes is a temporary change in the parameters to make everything go bananas — an accident on a freeway, a stuck process on a computer, a powder day, a market crash, etc. — and the queues run wild. There is a kind of hidden wildness in systems that disappears, until it doesn’t.
Look around you for queues. You will find them everywhere, and we are obsessed with making them more efficient. This has consequences, and only subtle changes will throw us into entirely new regimes.
A few papers worth thinking about:
I have long argued that the primary benefit of doing startups in dense cities (to a point) is that it increases the likelihood of “collisions”, of people, by chance, running into other people. That can lead to an exchange of ideas, which provokes new ways of thinking, and sometimes turns into innovations. I am pleased to see this being born out in a new study, that shows that overall innovation is flatter than is often modeled, but “atypical innovations” are more closely associated with dense urban areas where collision foment them.
People are funny about how they feel about robots and jobs. For example, they prefer that other people’s jobs aren’t replaced by robots, but they would prefer their own job, if it gets eliminated, be replaced by robots than by humans. Why? The authors argue that “being replaced by machines, robots or software (versus other humans) is associated with reduced self-threat”. This is intriguing, and not at all what we usually think happens.
We are becoming increasingly aware of health problems tied to inflammation, and the problems apparently start even earlier than previously known.. According to a new paper using erythrocyte sedimentation factors (how quickly red blood cells fall to the bottom of a test tube as a proxy for inflammation) in adolescence are highly predictive of death due to cancer and cardiovascular disease.
A provocative, rich, and fascinating talk by Danny Hillis on how complexity emerges from simplicity, whether we are talking about life, technologies, or almost any other common system.