by David Strayhorn
STATUS: UNFINISHED FIRST DRAFT
I propose five desirable features for a web of trust-based ratings and reputation system of the future: flexibility, simplicity of the user interface, privacy, monetizability, and portability. These features are unified by the single organizing principle that the end-user must be the ultimate steward and owner of his or her own data.
Ratings and reputation systems of contemporary social networks like eBay, Reddit, Amazon, Yelp, etc leave much to be desired. In this article I present five highly desired features that are lacking in the current state of the art but which must be envisioned in the next generation WoT.
These five features -- flexibility, simplicity of the user interface, privacy, monetizability, and portability -- can be understood as flowing from and being inspired by the single organizing principle: The end-user must be the ultimate steward and owner of his or her own data. One of the best tools to design and envision putting this principle into practice is to imagine recreating contemporary social networks so that they exist on a purely peer-to-peer network, eg the reincarnation of eBay as OpenBazaar where no centralized authority exists to whom stewardship of user data could possibly be yielded. However, there is no reason that these five features could not be implemented by an old fashioned centralized website. For the sake of illustration, I will discuss how each feature might be implemented in the specific case of a (centralized) Reddit-style bulletin board to address one of the most vexing problems of many social networks: trolls. For the sake of this article, this hypothetical bulletin board will be named "arranarr," short for R 'n' R ("ratings and reputation").
1. Flexibility / Customizability: The end-user must have exquisite control over the cagorizations, labels, algorithms for generation of composite scores, and other aspects of the ratings and reputation system.
Reputation systems of today are defined centrally, are inflexible, and create scores using algorithms that are frequently opaque. But the end-users' needs will vary over time, from community to community, from user to user, and even within the same user but from one context to another. In the WoT of the future, the end-user will have the ability to create new ways to rate products/users/content – new ’labels’ – as well as new ways to analyze the data (create composite scores for sorting purposes).
Bulletin board example: Suppose a user comes up with the idea of three new cagories of users: Trolls, who do what trolls do; Troll Spotters, whose job is to flag users as Trolls (or label them not-Trolls); and Blacklisted, which are users whose content is blocked. Each user maintains his or her own personal lists of Trolls, Troll Spotters, and Blacklisted Users. The purpose of these categories is to establish an automated system to filter out content generated by Trolls. But instead of simply flagging each and every Troll by hand, each user maintains a conceivably relatively small and simple list of trusted Troll Spotters and sets up an algorithm so that anyone flagged as a Troll by a Troll Spotter will be Blacklisted. Flags generated by non Troll Spotters are ignored.
If desired, the algorithm to place users on the Blacklist can become more complicated. One user may change the algorithm so that at least two Troll Spotters must flag a user as a Troll before that user is Blacklisted. Another user may temporarily Blacklist any user account that is not more than 24 hours old.
Another user in another community may be less worried about Trolls and more worried about large scale sybil attacks. Therefore this user may come up with categories: Sybils and Sybil Spotters which function in fashion similar to Trolls and Troll Spotters.
These concepts (Troll, Troll Spotter, Sybil, Sybil Spotter, and the associated algorithms) are broadcast publicly so that other users will hopefully start using them. Users who choose to do so place add these concepts to their default settings. Just like a subreddit, these concepts either gain widespread usage or they don’t. If not, no harm done.
2. Simplicity of User Interface: The WoT needs to be simple for the casual or beginner end user to use, despite the complexities introduced by the various other desired features (primarily flexibility).
One may think it will be impossible to make a WoT both simple and flexible, because flexibility will introduce too many complications. The solution to this problem is summed up in two words: default settings.
Bulletin board example: One of the most important jobs of the creaters of arranarr will be to do their best to populate default settings of new users with useful concepts like those of the troll system described above. New users can start using the network without necessarily having any idea that making new concepts is even possible. For those users who are interested, they can make new concepts which other users can incorporate into their default settings. New users have the option of adopting either the default settings, adopting from a list of "most popular" settings, or adopting the settings used by someone they know or trust. Casual users inherit default settings. Only the sophisticated and interested users need get under the hood to come up with new user categories of users, new algorithms to generate composite scores, etc.
3. Monetizability: The end user will be the one to monetize his or her data, and will have a myriad of strategies to do so.
The primary business model for many if not most online social networks (Facebook, LinkedIn, eBay) depends in one way or another on the centralized control of your data. LinkedIn offers a more detailed view of connected user profiles, more details about local WoT topology, and more in exchange for a paid membership. Product rating sites including Amazon, eBay, Netflix, Yelp, AngiesList make use of user-supplied ratings to enhance the overall utility of their site and attract users. Facebook mines your personal data to make money through targeted advertising. The WoT of the future must turn this business model on its head. By giving the end user control over their data, it will be only natural that the end user should make money from that. Based on market caps, we should expect that to be worth quite a lot. The question is how exactly that will come about.
Bulletin board example: A Troll Spotter or Sybil Spotter in arranarr will have the option of giving his ratings away or selling a subscription for a microfee in bitcoin lasting a preset amount of time, meaning that any user with an active subscription is allowed to access the latest, up-to-the-minute ratings. (If frequent respawns of Trolls or Sybils are an issue, then access to the latest ratings may be worth paying for.) Suppose a bulletin board has 20 million unique users per month (which is what Reddit reports), and a Spotter charges 10 cents for a one week subscription. Assuming that only 10 percent of users subscribe, this market could support 1000 Spotters working full time to ferret out bad actors, each one making on average 2000 dollars per week, 100k per year.
Self-explanatory.
5. Portability: Identity and reputation must be readily portable between one social network and the next.
Many pieces of data generated on one social network will be useful on other social networks. For example, a troll on another social network will probably be a troll on arranarr. A useful script will be one that amalgamates multiple ratings of a given user across multiple networks and come up with a single composite score. However, a major barrier is that there is a lack of uniformity between networks. One network may rate products on a 0-5 scale called "Quality"; another, on a 1-5 scale called something else. The labels may be similarly defined, but not quite the same. For ratings across networks to be put into meaningful composite scores, it will simplify things greatly if different networks could employ labels that were defined in exactly the same way.
Bulletin board example: When a concept such as Troll, Troll Spotter, etc is generated on arranarr, a complete specification of each concept will be constructed (eg in json format) and it will be given a cryptographic identifier, will be signed by its author, and (if desired) will be posted to a public data repository. This will allow the concept to be imported into another network (by the creator of that network or by a user within that network), so that there can be uniformity across networks wrt how the relevant labels are defined. Other pieces of data, including statements of individual ratings ("Bob says Alice is not a Troll") will also be given cryptographic identifiers and posted (if desired) to a public repository, so that a good reputation built on one network can be automatically imported and used in the context of a different network.