Real-World Coordination Problems are Usually Information Problems
post by johnswentworth · 2019-06-13T18:21:55.586Z · LW · GW · 2 commentsLet’s start with a few examples of very common real-world coordination problems.
- The marketing department at a car dealership posts ads for specific cars, but the salespeople don’t know which cars were advertised, causing confusion when a customer calls in asking about a specific car. There’s no intentional information-hoarding, it’s just that the marketing and sales people don’t sit next to each other or talk very often. Even if the info were shared, it would need to be translated to a format usable by the salespeople.
- Various hard problems in analysis of large-scale biological data likely have close analogues in econometrics. The econometricians have good methods to solve the problems, and would probably be quite happy to apply those methods to biological data, and the bio experimentalists would love some analytic help. But these people hardly ever talk to each other, and use different language for the same things anyway.
- When the US invaded Grenada in the ‘80’s, the marines occupied one side of the island and the army occupied the other. Their radios were not compatible, so if an army office needed to contact their counterpart in the marines, they had to walk to the nearest pay phone and get routed through Fort Bragg on commercial telephone lines.
- Various US intelligence agencies had all of the pieces necessary to stop the 9/11 attacks. There were agencies which knew something was planned for that day, and knew who the actors were. There were agencies which knew the terrorists were getting on the planes. There were agencies which could have moved to stop them, but unfortunately the fax(!) from the agencies which knew what was happening wasn’t checked in time.
- There are about 300 million people in the US. If I have a small company producing doilies, chances are there are plenty of people in the US alone who’d love my doilies and be happy to pay for them. But it’s hard to figure out exactly which people those are, and even once that’s done it’s hard to get them a message showing off my product. And even if all that works out, if the customers really want a slightly different pattern, it’s hard for them to communicate back to me what they want - even if I’d be happy to make it.
- Just yesterday I was looking for data on turnover time of atherosclerotic plaques. I know plaques increase with age, but is it the same plaques in the same places growing, or is it an increase in equilibrium number of plaques (each appearing and dissipating quickly)? There’s probably thousands of people who can easily answer that question and would be happy to do so, yet finding a clear answer is still nontrivial.
Obviously these are all specific examples of problems which happen all the time.
To some extent, coordination problems are universal and have always been with us. But humans evolved to solve coordination problems in Dunbar’s-number-sized groups (plus or minus an order of magnitude) regularly talking face-to-face. Even just two hundred years ago, most people operated in relatively small communities. It’s only since the rise of cheap long-distance communication that large-scale coordination problems have crept into everyday life. The cheaper and more ubiquitous long-distance communication becomes, the more coordination problems are going to be a bottleneck. Not all coordination problems look like this, but these are the sort of coordination problems which we’d expect to become more common over time. (See “From Personal to Prison Gangs [LW · GW]” for a more fleshed-out version of this argument, and related problems.)
Look over the list of coordination problems above, and a few major themes jump out:
- Matching problems: Doily-makers know there are customers out there who want their product. Bio experimentalists know there are analysts out there who want their data. I know there’s someone out there who can answer my plaques question. But finding those people, in a world of 6 billion, is a needles-in-haystack problem. Just figuring out who to talk to is hard.
- Lack of communication channels: Cheap, fast communication channels just don’t exist between company and customer or between departments of an organization. Even if you know who to talk to, you still need a way to talk to them.
- Language difficulties: Econometricians and biologists use different language or even different frameworks for similar systems. Different departments use different data formats. The army and the marines had incompatible radios. Even when you know who to talk to and have a channel, communication can still be hard.
Standard discussions of coordination problems tend to focus on cases where a dictator could easily solve the problem. Need to meet up with someone in New York City at a specific place and time without communicating in advance? The dictator can declare “Empire State building at noon is the official meet-up spot and time”, and there we go, we’re done. But the harder sorts of real-world coordination problems usually aren’t that easy. Having a designated dictator on hand doesn’t help a doily company find enthusiastic customers, or help a biologist and an econometrician realize they should collaborate, or help translate data from one format to another (assuming they do in fact need different formats).
The biggest problem is that there’s a combinatorially huge space of possible coordination problems, and any particular coordination problem won’t happen many times. How many people have asked my exact question about atherosclerotic plaques? In order to be useful, a coordination mechanism has to address a very wide class of coordination problems in one fell swoop - e.g. the question-answering site Quora. But simply declaring “this is the canonical question-answering site” doesn’t solve the problem - in order for it to actually work, we still need a good matching engine, so that askers and answerers can find each other without having to search through the haystack themselves.
A combinatorially huge space of problems directly leads to a more insidious issue: humans have limited processing ability, so there will inevitably be coordination problems where nobody involved even knows what’s possible. The biologist and the econometrician don’t know that their fields complement each other. In order to solve that sort of problem, a third party has to proactively look for opportunities to coordinate. Once the opportunity is found, actually connecting people is the relatively easy part - lots of academics are interested in opportunities to collaborate across fields (I hear grantmakers love that stuff).
2 comments
Comments sorted by top scores.
comment by Dagon · 2019-06-13T21:59:29.657Z · LW(p) · GW(p)
I don't see much justification for the word "usually" in the title. And while I'd agree that many real-world communication problems include information issues, I think they're almost always mixed with alignment (not all participants are fully onboard with the goals) and trust (even if we agree, is it worth my effort if you're going to let me down) issues.
It's very hard to tell which of these issues is most important for any given failure, but I'd argue that the alignment and trust issues are causes of the information issues.
Replies from: johnswentworth↑ comment by johnswentworth · 2019-06-14T00:17:43.919Z · LW(p) · GW(p)
“From Personal to Prison Gangs [LW · GW]” is my main foundation here. I say real-world coordination problems "usually" look like this, because these are the kinds of problems we'd expect to increase over time, based on the ideas in that previous post.
That said, Personal to Prison Gangs attributes both the information problems and the trust problems to the same root cause: I interact with a larger number of people, with fewer interactions per person. On the one hand, fewer iterations means less penalty for "defectors", and common knowledge of this fact means less trust. On the other hand, more people + fewer interactions per person means both less time and less mental resources to customize my interaction with each individual person. Thus, in a large company, people are forced to rely more heavily on job titles - and in a larger society, people are forced to rely more heavily on identities more generally.
All the examples listed in the OP are the sorts of things you'd expect in a world with more people, more specialization, and fewer interactions between any given pair. (In some cases, this means zero interactions between a pair which would really benefit from interacting, as in several of the examples.)
I don't disagree that alignment & trust have a role here. But I do think that the large majority of real-world coordination problems could be solved by sticking the right two or three people in a room and just letting them talk for a full day. And in most cases, I think the relevant people would actually like to talk! The problem is finding the right two or three people and building that communication channel.