Reputation systems work because people are mostly good

Economist Tim Harford writes about holidaying in prosperous Bavaria, where hotels let you run up bills of €1000+ without a credit-card and all room-keys are stored in a cupboard where any guest can get at them, and asks how this can all work without being destroyed by dishonesty?

Trust is key to a prosperous society, and it seems like the causal arrow points in both directions: a society without trust wastes resources on walls, bodyguards, and inefficient monetary strategies (hiding cash in mattresses or wall-safes); a rich society has fewer problems with trust.

We build reputation systems into platforms like Airbnb to allow us to treat strangers like neighbors, with all the good and bad that this entails: on the one hand, it lets a whole lot of people feel like they can trust randos to stay in their house, with a surprisingly low level of problems. On the other hand, it also enables systematic racism by Airbnb hosts, who stir their own, biased trust-metric into the mix of who they'll let stay in their homes (Harford notes that in trusting Bavaria, shopkeepers ban Arab kids from touching sunglasses, but make an exception for white kids).

Reputation systems like Airbnb replicate all the problems of other rules of thumb — including wealth inequality and the delusion that wealth correlates to merit and worth. Just as cab drivers will foolishly pass by upstanding black fares to pick up white paroled felons, so too will Airbnb hosts deploy their biased heuristics to their own detriment.


But still, Airbnb hosts and cab drivers mostly don't get robbed, because most people aren't robbers. Even in fields that seem dominated by bad guys — say, email, where nearly everything you get is a scam of some sort — the actual number of scammers is vanishingly small (it's just that technology allows them to be fantastically prolific). Even there, the scammers might be upstanding citizens in desperate circumstances for whom the con is the least-worst option, or they might be coerced into participating in the con.


All complex ecosystems have parasites, and the rare bad actions loom large in our perceptions (think of how Donald Trump is able to sell America's all-time low-crime moment as the most dangerous moment in American history). But the use of reputation metrics is, often as not, just a way for us to justify our bias as it is for us to winnow away the bad actors — and since most people are good, virtually any system to sort the good from the bad will "work" — from the perspective of the person deploying the system.


Personalised trust has never been fairly distributed. When Harvard Business School researchers Benjamin Edelman, Michael Luca and Dan Svirsky (pdf) conducted field experiments on Airbnb, they found that both hosts and guests were discriminating against racial minorities. Other researchers have found evidence of discrimination in places from Craigslist to carpools. New online tools are giving us the ability to treat faraway strangers as though they were neighbours — and we do, in good ways and in bad.

Trust is as unfairly granted in Bavaria as anywhere else. While browsing for shades in Garmisch-Partenkirchen, I warned my young son not to play with the merchandise: a sign forbade children to touch the sunglasses.

The shopkeeper bustled over and reassured me that the rule did not apply to my son. "It's for the Arab kids," she told me, beaming. "They just drop the sunglasses on the floor."

Ah. My son is adorably blond but he is as capable of snapping a pair of designer sunglasses as any other four-year-old. Trust is sometimes given to people who do not deserve it. And it is often withheld from people who do.

The meaning of trust in the age of Airbnb
[Tim Harford]


(Image: Ukrainian girl scout from Plast gives a Scout Salute, Viktor Gurniak, Yarko, CC-BY-SA)