unhosted web apps
freedom from web 2.0's monopoly platforms
26. Decentralized reputation systems
Decentralized reputation systems
When communicating with pre-existing contacts, trust is simply established by making sure the other person is who they say they are. As discussed in episode 17, this can be done using a combination of DNS, PGP, and checking whether the other person sounds like themselves in how they interact with you. But when interacting with strangers, trust can be based on reputation.
Basing trust on reputation is fundamentally unreliable, since a person's past behavior is no guarantee for their future behavior. But a certain game theoretical assurance can be achieved when people not only prove that they have a certain reputation from past behavior, but also risk this reputation. Assuming the other person is a rational agent, they will not destroy a valuable reputation in order to trick you once in a relatively small interaction. A seller on ebay or airbnb is unlikely to treat you unfairly if they risk getting a negative review from you.
This means the person being reviewed, in turn, assumes the reviewer will be fair. It is not always clear what the game theoretical advantage is of reviewing people fairly, but there is also no advantage in reviewing people unfairly, and I guess people are just fundamentally nice to each other when they have nothing to lose by it. In any case, in most reputation systems, reviewing is a voluntary act of altruism, both (in case of a positive review) towards the person you are reviewing, and towards the other future users of the system or website to which you are adding the review. Examples of such systems are Hacker News, Stack Overflow, Twitter, Linked In, eBay, Amazon (reviewing books rather than humans), and AirBnB.
On your Indie Web site, you will probably link to some of your "claims to fame", so that people who look you up online can see how many Twitter followers and Github projects you have. This way, you can bootstrap your reputation from your reputation inside these centralized web2.0 walled gardens. You could also post PGP-signed recommendations from other people on your website, for instance a few customer reviews of products you offer.
Reputation is not a scalar. You can be a good JavaScript programmer on Stack Overflow, but a poor host on AirBnB. If we are to automate decentralized reputation on the web then we would have to take this into account.
The most reliable way to let people review each other would seem to be if reviewers take a certain risk when inaccurately reviewing someone - for instance their own reputation. Humans dedicate a considerable part of their brain activity to reputation of themselves and others in a group, and a lot of human activity is dedicated solely to establishing a good reputation, status and fame. Even people who don't have enough money to buy proper food, health, and education for themselves, often spend a lot of money on jewellery and flashy cars which basically act as proof-of-work to increase their social status within the village or group.
We have only just started the era in which people tout their prosperity online instead of in the material world, and I think a lot will change in how the economy works when people care ever more about their number of Twitter followers, and ever less about the brand of their car.
Reputation spam
A very interesting system implementing reputation on top of pseudonymity is gnunet. It uses the principle that a newly created user account should have a value of zero, otherwise an adversary can simply destroy their reputation and then create a new user account.
Reputation in gnunet is built up using a sort of tit-for-tat scheme, which they call an excess-based economic model, since newcomers can only use the network's excess capacity. This boils down to: don't trust your neighbours a priori; at first, you will only answer their requests when you're idle. Over time, allow them to invest in their relationship with you by answering your queries, while not sending too many queries themselves. As their score increases, you will eventually allow them to send requests at higher effective priorities as well.
Peer education
Universities have an often centuries-old reputation, giving them the power to judge which students pass and which students fail their tests, even though there is a big monetary incentive to give out false diplomas. If we want online peer education to become bigger, we need a system like that, where a teacher can gain reputation, and a student can become a teacher. Good teachers could charge money for their classes, while teaching material can be (and already is) available in the public domain.
Likewise, a student can get a "diploma" from a renowned teacher when they pass a test. This test can even be highly automated as an online service to improve efficiency, but the value of the diploma would come from the reputation of the teacher who (digitally) signed it.
A lot of space for future research is open in this area, and I expect a lot of exciting developments over the coming years. This episode concludes the second part of this blog series, about theoretical foundations for building freedom from web2.0's platform monopolies. In the third and last part, we will look at practical issues related to building unhosted web apps. Comments welcome!
