On reputation economies

Some notes on reputation economies I have been throwing around and thinking about.

The idea that we are living in a reputation economy seems to gain more and more traction. In Daniel J Soloves important book The Future of Reputation (2008) we find a discussion about privacy couched in terms of reputation and just recently Daniel H Kahn published an interesting paper on social intermediaries, entitled “Social Intermediaries: Creating a More Responsible Web Through Portable Identity, Cross-Web Reputation and Code-backed Norms”.*1*

This idea, that we can build a more responsible web and move away from the tendencies towards anti-social behaviour online with norms and persistent reputations seems to be clear enough as well as fairly plausible. After all, we assume that people do not behave badly in real life because their behaviour reflects on them, changes our perceptions of them and allows us to treat them according to our analysis of their actions. When that reflective element disappears and social reciprocity is eliminated by the anonymity of the web, we see weird phenomena like net hate and rampant defamation take over. So let’s reintroduce reputation and all will be well!

I simplify, but somewhere along those lines we find an argument that is becoming more and more prevalent. I believe, however, that it is worthwhile to try to see if there are weaknesses in that scenario. Not, primarily, to try to kill this line of reasoning, but rather to modify it and allow for a bit more complexity. In order to do so I will offer three observations.

The first observation is that reputation and track record are two very different things. The studies on reputation in economics are primarliy about studies of abstract actors and firms, not about humans as social beings. This observation may seem obvious, but it could carry with it some fairly profound challenges to the idea that we are going to build a “reputation economy”. If we, with a “reputation” refer to the sum total of information available about an actors actions and go on to assume that this will have a predictive value for trying to calculate what that individual will do in the future, well, then we are using the notion of reputation in a limited way. In fact, what we are describing here is less of a “reputation” than a “track record”. The difference may not be obvious, but I would hazard to say that a reputation is a far richer and deeper social construct than a track record. To know something about someone’s reputation is to know, in a sense, more about that individual than to know their track record. Now, we may full well decide to delimit the meaning of “reputation” to be synonymous with “track record” and leave it at that, but again I think we would miss out on a couple of interesting observations. A track record allows us to predict with greater accuracy how an individual will act in a specific situation – say how he or she will make a move in a specific position on a chess board. A reputation will tell us something about their style of play; is it defensive, positional, tactical or perhaps modern? A reputation allows us to make far more general predictions about an individual. Take an example that is obvious even to a non-native English speaker: it is obvious to “tarnish a reputation”*2*, but it is fairly rare to say that someone “tarnishes their track record”. One way to understand that subtle difference is to say that a reputation does not change with a single action, where as a track record actually does. There is such a thing as “reputational inertia” and a reputation will shift from good to bad not in a series of actions but at a certain tipping point. Reputation – as a social attribute – moves between punctuated equilibria in discrete steps rather than continously as an expected cost. This is perhaps also reflected in the way we think about reputation being earned, not accrued. (But is this not a contradiction? If we earn a reputation, is it then not obvious that we realize in our language that reputation is like capital? We earn our salary after all! Well, no. I would argue that the word “earn” here is used in the sense “deserve” rather than in the sense “gain”, and even if we applied the second meaning we would find that you earn discrete entities rather than continous sets. But it should be pointed out that I think that it is language like this that leads us to think about reputation as an economic good.)

The second observation is that a reputation is relational and not absolute. When we speak about track records we speak about the sum of an individual’s actions over time and this is quantifiable in a simple way. Everyone has a track record. But not everyone has a reputation. We may say about someone that he has “quite a reputation as a baseball player” and part of what that says is that when we think of baseball players this individual may naturally come to mind. But it would be absurd to say that I have a reputation as a baseball player, and here is an interesting guess: I would venture that to say that someone has no reputation as a baseball player implies more knowledge of baseball than saying that someone has no track record as a baseball player.*3*

Now, this is interesting because one part of the usefulness of reputation lies herein: it allows us to distinguish a small group with exceptional qualities in a given area from a larger group.

This observation can be expanded into an observation on the problems with the envisioned “reputational math” that seems to underpin much of the thinking around the usefulness of reputation in an online environment. Take reputation on online auction sites as an example. How do these work? Specifically, how do we value reputation on these websites? I think we can argue that the evaluation of reputation on these websites is intrinsically hard to do. Let us take an example. Assume that a seller has a reputation R that consists of 100 positive reviews and 3 negative reviews. What does this mean to a seller? To be able to decide that we need to examine not only the reputation as such, but also the reputation relative to all other sellers. Again, let’s assume that you are buying an iPod and that you have three sellers to choose from: the first has 100 positive reviews and 3 bad reviews. The second has three positive reviews and no bad reviews. The third has 70 bad reviews and 20 positive reviews. At a first glance it would seem natural to exclude the third seller without any more consideration, right? But what if that seller has sold 70 boxes of cookies at 10 dollars that have tasted horrible, and has sold 20 electronic goods of different kinds at a total dollar value of ten thousand dollars with shining reviews?

Now, obviously the nature of the good that produced the review seems to be relevant, and perhaps even the dollar value – right? Assume that the second seller has sold three used books at a total dollar value of three dollars, and is now selling an iPod – will you rely on the all-positive review? And what if the first seller has had a 100 great reviews for everything but the three iPods he sold previously? Reputational math is not all the easy, and this may seem obvious, of course. You could argue that this is the way reputation works in the real world as well, and that there is nothing strange about this. But then you would be wrong, I think. In the real world reputation is embedded in a form of life or a social context, that is horribly hard to reproduce in mediated forms. When we talk to a friend who recommends someone*4*, or we hear about someone who has acquired a great reputation, we are not calculating dollar values and evaluating the structure of their reputation. We are partaking in a social practice where reputation allows us to lower the transaction costs involved in finding someone to interact commercially or otherwise with.

One possible consequence of this is of course that only completely clean reputations are of any value. If the cost of evaluating a complex reputation is high, the use of reputations will lapse into a simple binary signal analysis: all good reviews signals that there are no obvious reasons to think that a seller is fraudulent. That would mean that the entire signalling semantic in complex reputational offline structures would be reduced to a green/red-light signal in the online environment (some research shows that the impact of the first negative review is much higher than would be predicted by a simple expected cost-model, other research studies indicate that the reputation system is being simplified by users – many more studies are available, but a quick analysis seems to suggest that they all support the hypothesis that online reputation behaves differently from the more complex social function of a reputation in the offline world). Hardly something worthy of all the hope and expectations being invested in the coming “reputation economies.”

The third observation I would like to make is that reputation is possible only in a community. It is not possible, I would argue, to think about reputation without thinking about it as situated in a community of people. Global networks cannot carry reputation across the world. In fact, this is probably the consequence of another problematic insight: global networks are not strong norm environments. We have seen in a series of recent writings the hope and call for new norms on the Internet as a solution to problems that we now think that legislation cannot address. The problem with this is of course that norms also need social contexts and communities. I think you could even say that norms are possible only in networks of a certain size.

This, certainly, maybe a provocative proposition to some, but I think it presents the almost trivial insight that norms are produced and reproduced by acting subjects in contexts that involve repetitive interaction and dialogue. This requires time and ultimatelty attention to eachother, social reciprocity, and since this is (following Herbert Simon’s observation) the ultimate scarcity we should not expect normsystems to be terribly large. Perhaps they are only possible within social organizations of the approximate size of the villages Ferdinand Tönnies discusses in his distinction between Gemeinschaft and Gesellschaft. For any hope of a reputation economy to grow this would mean that we would need to find some way to replicate social reciprocity in complex and mediated networks. That does not seem likely, right now.

Why spend time on attacking a sweet idea like reputation economies? Is it not a beacon of hope in what is sometimes felt to be an increasingly anti-social online environment? Well, no. I would argue that it presents a weak response to the real challenge that we may face in creating open, global networks: that these networks very may well not bed open to norms, to reputation or any other socially mediating or regulating factor.

Of the four Lessigian regulators, norms may actually be out of the question for the net, globally, and only work in specific contexts where online and offline are meaningless anyway as the social fabric becomes integrated across social practices and technologies.

We may have islands of norms in a sea of normlessness.

If so we may actually need to ask ourselves to refrain from trying to regulate this normlessness, and realize that the anti-social and sometimes hateful behaviour of human beings is a constant of the human experience, and that the challenge becomes to find ways to handle a huge, global and norm-resistant communication network, where net hate, abuse and malicious anonymity may – in a worst case scenario –  be the prize we pay for economic growth, creativity, freedom of speech and innovation.

Now, this in no way means I advocate giving up on thinking about reputations, norms and other social mechanisms in the online context. I just want to moderate the expectations of those writers who think this is a medium-hard problem.

*1*See Kahn, Daniel H., Social Intermediaries: Creating a More Responsible Web Through Portable Identity, Reputation Building, and Code-Backed Norms (December 26, 2009). Available at SSRN: http://ssrn.com/abstract=1528482
*2*The same phenomenon occurs in Swedish, where it is possible to “fläcka ett rykte”.
*3*This would at least be true in Swedish, I would argue.
*4*I am aware that recommendations and reputations are slightly different. We will soon return to this.