Seems like "trust that scales" is an important social technology that most of us have lost. Perhaps the population got too large and disconnected, and the mechanism didn't scale enough. (Two random medieval Islamic merchants probably had a friend in common. Two random people in New York City probably don't.) But sometimes you can still use it, if you limit yourself to a small social group such as the Amish, or the upper class.
I already know from experience that trust is not transitive. There are people whom I trust, but I don't trust their judgment about whom else they trust. Sometimes this seems correlated; some people seem too naive to lie, but also too naive to spot a liar. Sometimes they place too much trust in their group, but the group is a mixed bag of true believers (who probably are trustworthy, when it comes to helping their fellow believers) and opportunists (who are mostly there to exploit the true believers and their connections). Sadly, at least once I have introduced an untrustworthy person to my group of friends, because I failed to notice the red flags. Similarly, I got burned by trusting a person that multiple(!) trustworthy people have recommended to me. (I still trust the people who made the recommendation, and they regret the mistake they made.)
I think there is a great difference between N:N networks (the Amish; the Islamic merchants) and 1:N networks (Johnson and his contacts). The latter are relatively easier to build, but they will not survive you. I wonder whether a mixed model is possible, consisting of an N:N connected core, and a set of people who are individually connected to the core but not to each other. This seems like a possible intermediate step between a 1:N network and a N:N network; you gradually keep adding people to the core.
How did the Islamic merchants resolve situations where merchant X claimed that merchant Y violated the rules, but merchant Y claimed that merchant X violated the rules? The one with more friends / better verbal skills won? What if there was an external influence, like everyone thought there was 90% chance X was telling the truth and Y was the trust-breaker, but Y happened to be sultan's cousin and you didn't want sultan to hear that you were badmouthing his relative? (An analogy today could be that Y has lots of money and great lawyers, and is going to sue you for slander if you talk about things that happened between you two without witnesses.)
Compared to individuals, companies tend to be more deliberate about whom to trust. Successful companies spend large resources trying to predict how people will behave if they are included in the company trust network. Recruiters look at resumes, do tests, interviews, ask for referrals – in short, gather data points that help them model the future behavior of the person applying. Are they trustworthy? Will they give mutual aid to their colleagues? Are they skilled?
Yet, companies are famous for being moral mazes. The referrals are often limited to "yes, this person worked here, in this position". Managers seem mostly unaware of which team members contribute most, or which help each other. So I'd say that companies wish to have information about whom to trust, but their actual mechanism to find this out is a cargo cult.
Trust often not being transitive is an important point that I had overlooked. Thanks a lot for that.
And I am fascinated with the pros and cons of N:N networks and 1:N networks at the moment. My model for thinking about them have been networks with a common border vs boundryless networks. (I say boundryless networks because 1:N networks will overlap with other 1:N networks creating a sort of diffuse bounryless whole). An upside of N:N is coordination. You can get a lot more done by forging a more definite community - say union, that negotiates on your behalf. But there is a lot of administrative overhead, and a higher risk that the network gets coopted by someone with a hunger for power. 1:N networks are more flexible, and it is easier to shed sociopaths even when large parts of N are fooled by them. And as long as trust is slightly transitive they might scale better, thanks to not having to deal with increasing coordination problems and consensus processes etc.
And yes, my description of companies is highly idealized (sort of based on Lazslo Bock's descriptions of how they do recruiting at Google). I don't think most companies fit the description. But compared to individuals, they at least has a mental framework for deliberately thinking about who to grant trust.
One obvious reason this might not work as well for you as for Islamic merchants is the scale of the threatened sanction.
Even assuming that the merchant would not face legal penalties for reneging on his note, and even assuming a lack of extrajudicial violence against him, he is still potentially subject to extreme levels of social pressure not available to you today. You cannot prevent a grocery store from selling him food, or a job from employing him. Social networks in a smaller society could potentially do that. You can...threaten to say mean things about him on the internet?
I'm glad this approach has worked for you, but I don't think it's game-theoretically sound, I think it is relying on good faith.
As an aside, I'm somewhat skeptical of the story of the merchants' notes being traded so far afield. How do you avoid forgery? If a note is given and never traded, the merchant can remember to who a favor is owed - if you stay in the same area, people can see who is owed and keep him honest. How can a trader in Zanzibar know whether this is a genuine favor note from a trader in Algiers?
Good points.
Can I double click on: "I don't think it's game-theoretically sound"?
And I don't know enough about the Islamic merchants to answer that question. I guess the notes where countersigned along the way and so on. But true, it does sound a bit outlandish to modern ears (which is why I thought it'd make a cute opening).
Attempted expansion:
If the person you are dealing with is willing to do the right thing even if it costs them something, a lot of different systems will work.
If you do not trust the person you are dealing with to do the right thing in the absence of incentives, you want a system that will impose incentives that ensure it is in their interest to do the right thing. I'm calling this 'game-theoretically sound' to mean 'the system accomplishes the intended goal even when one or both parties are the kinds of sociopath that prevail in game theory problems'.
This could be formal law (do your end of the contract or the government will punish you).
It could be informal threats (uphold the bargain or me and my mates will clobber you).
In some circumstances it could be reputation (honor your note or no one will deal with you again? refund my purchase or I will review you badly online?).
However, for reputation to actually work as a system that is protected against adversaries/sociopaths (as distinct from working as a system for nice people who are already pretty much trustworthy), you need the damages caused by reputation to exceed the benefits gained by defecting. This is plausibly true for ancient merchants living in small societies, or for companies that have large numbers of customers. I don't think it's true for your networks of trust.
That is not necessarily a problem for you! If you're just dealing with your neighbor, you don't need your system to be defensible against sociopaths. But if you want to scale up your system, it will become more and more relevant.
Hope this is clearer, apologies for length.
I appreciated the length!
Its true, building a network with high barriers to entry is hard to scale, and will never compare to the scalability of a functioning justice system or something like blockchain. And it relies a lot on weeding out sociopaths to function; though the value of belonging to a high trust network can work as an incentive to play fair and not get excluded.
And it relies a lot on weeding out sociopaths to function;
Like, if that note you signed gets cashed in on after your death, who pays it?
Also, what if someone forges your signature and makes fake notes?
Those are the questions I have about the written promises.
Here are some questions to note against "networks of trust":
These are some of the more interesting questions this essay has provoked.
Sorry for the stream of consciousness. I need a cup of coffee.
What a generous, personable post! Good-natured and earnest in the best Scandinavian tradition.
(Notes to self:
Let me stress, in case it's not clear, I'm saying 👍🏼 excellent essay on a subject we (as a society of individuals) badly need to address ASAP; in a way that scales into extant power hierarchies, without simply abstracting to create another corrupted homogeneity.
Something I really like about this is that it is babbling. I think too much time is spent on LW doing the opposite: pruning. Saying "this is a bad idea" or "this won't work". I'd like to see more "what if we did this?". Especially when the upside is large.
Yet it was said that its merchants shunned enforceable contracts, preferring to seal transactions “with a handshake and a glance at heaven”.
I wonder if this worked because they thought god would punish them if they defected.
If he does, God forbid, I'm going to take a page from the Islamic merchant's playbook: I'll slander him in verse. And then I'll email it to all the connections we share.
Hm, I wonder how they used to get around this risk. What if someone threatens to slander you unless you pay them some amount of money? I guess we face the same risk in our society, but their society depended more on trust.
The best thing about having a trusted connection is that you can access that person’s network now.
Good point. I never really recognized the power of this until now.
I didn't notice this post was in the Review until it was over, but, I wanted to say I liked it.
It's been about 15 years since I read Samuel Delany's Times Square Red, Times Square Blue, but I read it multiple times and wrote a paper or two which were heavily influenced by it.
This post reminded me of it. You might find it an interesting read. Times Square Red, the second half, is about strong and weak social networks (particularly in urban environments) and how they function.
Just a heads-up: the first half is quite sexually graphic -- it's a pretty explicit memoir of Delany's time in the Times Square porn theaters and the subculture of gay/homosocial sex and relationships that flourished there. It's pretty raunchy : )
Benjamin Franklin, famous for his ability to forge vast networks of trust (consisting of mentors, business partners, fraternities, voluntary associations…) was equally famous for his bad judgment in choosing friends.
Maybe he had a lot of friends, so people got to see a lot more of the distribution than was normal? (There's also a joke about antifragile in here.)
The cost of these experiments is capped, but the upside is unknown and may well be priceless. If you lend your neighbor a motor saw, you can't lose more than a motor saw. (In the US, I guess you could get sued as well if your neighbor had watched too much Evil Dead and decided to saw of their hand.) But the upside of motor saw loans can be enormous – high trust transactions for the rest of your life or your neighbors. Whichever is shortest.
And that's assuming the transactions end there.
When you get people to extend help, they sometimes continue to help just to live up to their self-identity of being someone that helps you. I'm not sure about the psychology here.
Would it make more sense if it was reciprocated? (i.e., when you give people opportunities to help you, they give you more opportunities to do the same? With regard to say, 'babysitting', maybe there's something to be said for efficiency if there's enough trust.)
If he does, God forbid, I'm going to take a page from the Islamic merchant's playbook: I'll slander him in verse. And then I'll email it to all the connections we share.
This part wasn't clear. If the neighbor moves? Defects? - then you'll curse his name?
Or if he continues to be an amazing neighbor but moves, you'll write songs about how he's such an amazing neighbor?
… it's possible that most people just aren't interested in [tools that help them leverage their trust networks], and that the whole network would devolve into a network of banks that abstract the trust of their account holders, just like today.
It's also possible to imagine a network which may or may not be about making money off loans, but instead provides members with something like liquidity (like an insurance company), but it's just a cooperative network.
On Franklin: in Walter Isaacson's biography, he makes a fairly convincing case that Franklin approached building trust networks deliberately and with great success, but choose friends based on more romantic notions that repeatedly ended in disasters. So I don't think its a variance thing, but to separate categories of people in his life.
"This part wasn't clear. If the neighbor moves? Defects? - then you'll curse his name?" – Sorry for being unclear. I was trying to be funny. What I meant was that one of the things that keeps my neighbor from defecting is the fact that it would cause other people to lose trust in him, which would be costly. (At the moment though, I want to sing his praise. If you ever live close, I can't recommend him enough.)
Franklin approached building trust networks deliberately and with great success, but choose friends based on more romantic notions that repeatedly ended in disasters.
That's interesting. I should study Franklin. (The stories sound entertaining, if nothing else.)
The humor came through alright. (Even if the part about that being a merchant tactic didn't - that's context the reader is missing, though can pick up on well enough.)
Franklin has a lot to teach. Isaacson's biography is not detailed enough to be super useful, though. I'd want something more in line with what Robert Caro does for LBJ. If anyone has a suggestion, let me know.
employing me to do your data visualizations
When I read this, I chortled so hard that I aspirated coffee.
Say you want to buy my house, but you're out of money. What can you do? There are some obvious things, like getting a job or taking out a loan, but those things bore you, so here's an interesting solution: you do something extraordinary that convinces me to trust you more than my wife. Then you sign a piece of paper saying: "I owe you one."
Assuming what you did to make me trust you was public enough, I might not even need to cash in your promise. Instead, I can go to someone else and hand them your note in exchange for a first edition of Newton's Principia. Now your promise has turned into currency. And I get to cry in silent awe.
Some of the most successful examples of people doing these types of transactions come from the early Islamic caliphate. Arabic papyri from the 9th century documents that merchants would write letters of credit (suftaja), saying, basically, "I owe you."
If the merchants had good reputations, these letters could be passed on from trader to trader – making their way from one end of the Sahara to the other. Even, on occasion, crossing the Indian Ocean. Traders in Zanzibar or Sri Lanka would trade pieces of paper backed by nothing but the reputation of a person living in Algiers, a person they would probably never meet. How could that possibly work?
David Graeber writes of one of these trading posts:
How was it possible to scale trust so far?
In another essay, a few weeks ago, I proposed that people in the developed world rely too much on markets. We can save time and money if we invest a larger share of our resources in developing networks of trust, I argued.
Though I also noted that "I don't know how far you can scale that until you hit diminishing returns".
Well, it seems like the Islamic merchants have explored that question before me. And they took it pretty far.
How can you scale your network of trust in cost-effective ways? How can you find highly competent and trustworthy people willing to engage in win-win-transactions, without relying on institutions that limit defection?
My current best answer – which I will unpack in this essay – is this:
Form predictions about who is trustworthy, competent, and willing to engage in mutual aid.
Test those predictions by initiating small transactions.
Scale up transactions that beat the market by playing tit-for-almost-tat.
Expand through referrals.
1. Predictions
Whom should we trust?
We trust our friends. When we need someone to babysit our daughter, we call our mother. We ask our best friend for feedback on our essay. Our ex is a designer, she can do the logo. This is a good enough heuristic for everyday transactions.
But it doesn't scale.
As ChristianKl pointed out in the comments to the last essay, we have a bias toward thinking our friends more competent and trustworthy than they are. I, for one, do not optimize for trust in friendships; I optimize for wildness of conversation. And having an overcooked brain does not correlate with trustworthiness. It correlates with substance abuse and being on a first-name basis with the receptionists at the psychiatry ward.
Benjamin Franklin, famous for his ability to forge vast networks of trust (consisting of mentors, business partners, fraternities, voluntary associations…) was equally famous for his bad judgment in choosing friends. Hugh Meridith, a close friend turned business partner, took to drink as soon as the printing press was up. At one point, Franklin threw him off a boat when he refused his turn at the oars. No matter how long they kept him swimming behind the boat Meridith wouldn’t relent. Another close friend reneged on a large loan – 27 £! – and disappeared after Franklin brought him along on a business trip to London.
Being a great party invite is not the same thing as being a great business partner. Mixing those things up – I've tried – sets you up for Shakespearean betrayals.
Selueen recounts in the comments:
So an important thing to remember: a network of trust is not the same thing as your friends. It is a group of trusted connections. People you know can engage in mutually beneficial transactions.
You are looking for complementarity. An ideal connection is someone that can solve problems you can't – say a plumber, if you're a programmer – and someone for whom you, in turn, can provide value. (This leads trust networks to be more heterogeneous than typical friendships.)
But if you are a programmer, how do you know which plumber to trust? You don't.
But you have a trump card. You can make a Bayesian prediction.
To keep things simple, I’m going to pretend that trust can be reduced to two dimensions (whereas in reality, it's a mess of different factors). When you trust someone, you predict that (a) they will do a good job and (b) they will treat you well if you play fair. Let's call those two dimensions skill and willingness to give mutual aid.
Saying that you trust someone in that case means that you predict that they belong above this vertical line:
And we’re especially interested in the upper right corner. That’s what we’re trying to fill in when we scale our network of trust. How can we find people that are both skilled and helpful? How can we find more people that act as the hippies did in the last essay?
As I pointed out there, companies are trust networks – and looking at how they operate can help us scale. Compared to individuals, companies tend to be more deliberate about whom to trust. Successful companies spend large resources trying to predict how people will behave if they are included in the company trust network. Recruiters look at resumes, do tests, interviews, ask for referrals – in short, gather data points that help them model the future behavior of the person applying. Are they trustworthy? Will they give mutual aid to their colleagues? Are they skilled?
These predictions can then be compared to the actual outcome. And that feedback can be used to improve the recruiting heuristics to make future predictions more reliable.
When recruiting people to your personal network of trust, you might not be able to be as data-driven. But the underlying aim should be the same. Look for things that will help you predict how competent a person is, and how likely to defect.
Before asking a group of hippies to redo our roof, my wife talked to a neighbor that knows them to see if he would trust them. He would. Then we watched two documentaries about them to study the constructions they had designed, and see them in action. It looked OK to me, but more importantly – my wife was impressed. She has weird enough hobbies to be able to tell if a carpenter follows industry standards, or actually understands what he's doing.
All in all, this took us maybe 50 hours (including the books on carpentry and the history of construction that my wife reads while breastfeeding). It saved us something on the order of a year of salary.
If you want to try this at home, it worth pointing out that we live in the Danish countryside. That means the underlying incentive structure is pretty solid. And the incentive structure is what gives the trust diagram it's distribution. If you live in an environment where people only engage in one-off transactions, and bad actors aren’t punished, you have to deal with a gravitational force pulling everything toward the Prince of Nigeria.
2. Testing the predictions
Employing someone to redo your roof is a risky way to start a relationship.
If you have located a potentially trustworthy person, the safer way to test that prediction is to make a small transaction. Figure out some way you can help – lend them a motor saw if they need to release some anger on a tree; point out a missed opportunity in one of their essays – and then see if they look for ways to help you in return.
The cost of these experiments is capped, but the upside is unknown and may well be priceless. If you lend your neighbor a motor saw, you can't lose more than a motor saw. (In the US, I guess you could get sued as well if your neighbor had watched too much Evil Dead and decided to saw of their hand.) But the upside of motor saw loans can be enormous – high trust transactions for the rest of your life or your neighbors. Whichever is shortest.
In 1812, Michael Faraday, who at the time was an unschooled bookbinder, was looking for a way to become a scientist. It would have been tricky even today, him not having any formal education, but it was even more daunting in an age when only the landed gentry and some reverends had the luxury of doing science. What did he do? He made a gift. A book where he had bound his notes from Sir Humphry Davy’s lectures on chemistry. Davy was so touched – and impressed – that a year later, after having hurt his eyes experimenting with nitrogen trichloride, he asked Faraday to step in as his assistant.
If you find the idea of helping strangers repellent, you can also do it the other way around. Asking for what you want is a solid and under-used strategy.
After your kids have played together with your neighbor’s a few times, ask them if they can drive you to the hospital when your kid starts throwing up and fainting. That can be the start of something big. (The element of crisis might help cement the bond.)
When you get people to extend help, they sometimes continue to help just to live up to their self-identity of being someone that helps you. I'm not sure about the psychology here.
3. Scaling transaction sizes
So you got someone to drive your vomiting kid to the hospital? Good. The next step is to return the favor. But don't just return it, do it with suger on top.
The crucial thing about trust relationships is that you do not want them to even out. You never return an egg for an egg. You return a rhubarb pie.
Not tit-for-tat, but tit-for-almost-tat.
If someone buys you dinner and you wire them the exact amount, that is a signal that you don't want the relationship to continue. If you instead give back by inviting them to your midsummer celebrations – well, that's a whole other story.
When the person you are dealing with is trustworthy and competent, tit-for-almost-tat creates a ratchet. You return slightly bigger favors every time, taking turns scaling up the transactions.
I and my closest neighbor are locked into one such escalating arms race of helpfulness at the moment. Every day at dusk, I'm looking out across the field to see what he's up to. I'm trying to figure out if there is some way I can help him. We fenced in his orchard; he offered us to use his car when we go grocery shopping; my wife and daughter fed his horses when he was away; now, he's helping the hippies remove the old roof for us. It's a lovely little arms race, and it has saved both of us countless hours (compared to having to buy services in the market).
Of course, if we escalate too far we might reach a point where the expected value of future transactions is less than what we owe. At that point, the incentive structure shifts; my neighbor might decide to defect when he’s on top. The same thing can happen if he decides to move – at which point he’s no longer constrained by considerations about future transactions.
If he does, God forbid, I'm going to take a page from the Islamic merchant's playbook: I'll slander him in verse. And then I'll email it to all the connections we share.
4. Referrals
The best thing about having a trusted connection is that you can access that person’s network now.
Often the access is indirect. You don't need to know who your connection has to trust to be able to help you. I don't know the suppliers that the hippies rely on. The hippies are trust abstractions, a little bit like functions – since I trust their output, I don't need to see the implementation.
At other times, you want to refactor the function and get access to the connections directly. This is called asking for a referral. If I need plumbing, I ask the hippies whom they would use – and since I trust them, I can transfer that trust and gain a new trusted connection. I can also do chains of referrals, where I ask them to refer me to someone that refers me to someone else that refers me to… alas, the further you go, the more you dilute the trust so we can’t go on forever.
In an interview, anthropologist Joseph Henrich describes using these chains of referrals to establish trust with groups he wants to study:
That gave him the trust needed to convince the Mapuche to participate in game-theoretical experiments. Making use of referrals this way, allowed him to expand his network of trust far beyond Dunbar's number – 150 – which is generally held to be the number of people you can coordinate through informal processes. It is one of the processes that makes it possible for a suftaja to travel across the Indian Ocean.
Referrals are a blast. The last time I traveled to a place where I didn't know anyone – Switzerland – I asked my friend Viktor to recommend me to people in Zürich and Bern. His connections in turn referred me to others. That immediately, as soon as I got off the train, planted me into a caring social network. Russian horror surf-bands. Ping pong in a stable. A Game of Thrones analysis marathon with some french literature students. They even bailed me out when I ran out of money, and got caught ticket-less on the train home.
To scale a network of trust you need to engage people you don't know in small transactions. You have to observe how they behave and see if the transactions are worth your time. When they are, you have to increase the transactions by returning ever-larger favors. And then you reach out through referrals.
Some people will be better at this than others.
Lyndon B. Johnson was an obsessive collector of trusted collaborators. From his college years on, he would do everything he could to make people he could trust stay in his orbit. He would place them in agencies he controlled, he would make them partners in law firms he trusted, or in his radio studio, or at subcontractors he favored. That way, when he decided to run for Senate, he had a minor army of lawyers and organizers that could be mobilized at short notice. And then, once elected, he proceeded to place them in federal agencies, where they would wait patiently for his bid on the presidency.
People like Johnson accumulate piles of trust and can make unfathomable numbers of referrals. They are massive trust abstractions. Having them in your network can vastly increase what you can accomplish.
But controlling the flow of trust also gives them power over you. They might take a cut of the transactions they enable, they might impose demands; Johnson perfected this. In the Senate, he would sit through the night with his phone in the dark, making call after call, pulling on an intricate web of connections until, by morning, the votes he needed had been switched. He would also humiliate the people that depended on him by making them take his orders while he shat in front of them.
When I discuss this essay with Ryan Fugger – a person who, having invented the concept behind Ripple, knows way more about trust than me – he points out that corporations usually play this trust abstraction role. Banks, for example, abstract trust, and then extract value from the transactions that enable. That's great. People who do not trust each other can buy trust from a corporation. But why are so many people unable to create their own trust?
Ryan writes in the chat:
For those willing to grow their networks of trust, rather than relying on bought trust, though, there is leverage to be had.
Take the Amish. In 1965, as Johnson (by then President) was expanding social security to include Medicare, the Amish leveraged their network power to renegotiate their deal with the State. Having high levels of trust, they were able to coordinate and get themselves exempted from the social security system – and the taxes.
Figuring out how you do that, however, I will have to leave for another day.
(Cross-posted from my blog.)