Making Reputation Reputable

The world is full of opinion. The internet has made this abundantly clear. We sometimes treat this as a revelation: “who does this obanana think he is, the president of America?” and other stupid comments are somehow treated as top news stories, even though they have always been there. The difference now is that the likes of twitter and other social media help spread them faster and further than ever. Treating such exclamations as if they were new or newsworthy is as stupid as the comments themselves, but people seem not to get bored by them appearing in headlines, so “stupidity voyeurism” will continue, I have no doubt. In cases such as these, it is easy to dismiss the sentiment as worthless, but in general, deciding whether someone’s statement of fact or confident prediction carries merit is difficult. Is this hotel review a report of a real customer experience or just an employee partaking in some extra marketing?

Even when we can somehow be sure that a DVD review is a bone fide opinion, how do we judge whether the reviewer’s tastes match our own? Perhaps the person writing about their shopping experience was just having a bad day and they are overly prone to being subliminally influenced in their comments. The answer, of course, is reputation. People with higher reputations can be trusted more in what they say. However, ‘reputation’ is a very complex idea. It may not feel like it at first, but given the range of situations where we want to know how much weight we should put on what someone is saying, on what advice they are giving and on the information they are presenting as fact, it is not only a very important idea, but also a very difficult one to work out.

The internet is making information flow across the earth faster and more accessible, which opens up opportunities for cooperation for the betterment of everyone like never before, but at the same time lays down more pitfalls for fraud than we can currently cope with, so a scaleable solution to ‘the reputation problem’ is sorely needed.

An oft-returned to solution is ‘appeal to authority‘, that is, pick a person or set up an institution to act as the authoritive expert to fall back on. Then we can ask, “Well, what does the President think of this research outcome?”. This approach has some desireable attributes such as practicality and the ability to draw upon costly resources (for example, it is rather expensive to test cars’ ability to protect their occupants), but it is inadequate for many reasons, including

  • a) that single points of authority tend to be overly prone to drift, arrogance and corruption
  • b) they cannot cope with the exponential rise in the rate of reviews, products, facts, etc that are being produced year on year
  • c) it is hard for individuals to incorporate their personal experiences, whatever they are worth, into that which the authority takes into account
  • d) when they do fail, there is nowhere else to go.

Another common solution that we see is the star rating. This stands at the opposite end of the scale to relying on a single authority by drawing on the wisdom of the crowd, peer-to-peer style.

Most of us have personally given online sellers and products a rating from 1 to 5 that is saved and aggregated. Then it can be given to those that haven’t yet submitted a rating an idea of what to expect. This approach has the advantage of being domain specific (although sometimes too much so) in that star ratings are not transferrable to the other products or services (e.g. a reliable retailer of bath toys is not necessarily going to be good at evaluating the latest developments in cancer research), and also the advantage of being scalable, despite people most often being lazy and not giving reviews themselves. However, it too leaves a lot to be desired: its simplistic design means that it can easily be gamed, with owners faking reviews and ratings with ease; there is also no mechanism for you to influence how other people’s ratings are aggregated, even if you know certain people’s opinion is better or worse than others. Many ideas have been suggested and tried to aleaviate this problem, but none really get to the heart of the problem and more often than not they make the system worse.

 

coinscrum_AB_reputation

BEFORE TORRENTING THERE WAS BOOK LENDING

Let’s explore a bit more what the problem actually is. Reputation appears in many contexts. Big brands build reputation through marketting and repeated experiences. Publishers, such as newspapers, acquire reputations for giving (un)helpful opinions and (un)factual reporting. Professional critics gain a reputation for giving consistent advice about films and plays. Personal friends get a reputation for being knowledgeable and authoritative on particular topics for pub quiz success.

The first thing to note here is the number of levels on which the concept of ‘reputation’ works: from the directly personal all the way to the most wide-spread impersonal levels society can handle.

The second aspect to keep in mind is the range of domains that ‘reputation’ is important in: everything from services, such as shopping, through products, like films, to information, which itself can range from opinion to hardcore scientific facts. The breadth and depth of these dimensions shows how important the idea of ‘reputation’ is, and also that if anyone insists that a single system can manage it, that system is going to have to be special, and probably complicated. Keep in mind that in addition to the range of domains and scales that reputation works on, almost all of it has a subjective element. While some electricians are objectively better than others, sometimes a good one comes along that just asks for too many cups of tea while they are working.

Now, each of us build up personal experience of interacting with those we interact with—by buying CocaCola, reading The Economist and inviting Sue into your pub quiz team—and so we know how much you can trust them in those contexts. We might even get an idea of how much we can trust their judgement on how much to trust people we don’t yet know—the coveted word-of-mouth reccommendation. Of course, because there is a first time for everything, personal experience will only get us so far. We need to rely on others at some point and perhaps call on that recommendation ourselves. This applies to pretty much everything, from scientific knowledge to buying pairs of shoes. The alternative is trial and error, which can be very costly. Human society has succeeded because of the efficiences that come from trust and cooperation. However, how much should you, as an individual, trust a plumber you have never met or a brand you have never purchased before. So, the problem of ‘reputation’ appears deceptively simple: a measure of how much you can trust someone, but as we have explored above, doing something as naïve as assigning a single universal ‘rating’ to everyone or eveything in order to represent how much everyone should rely on what they say or do is not actually that helpful in practice, even leaving aside the problem of cheats. Furthermore, when you have several options options present themselves, comparitive trust naturally becomes a consideration too. These aspects to the problem are likely very familiar to you, especially if you are reading this online, where adverts constantly go beyond being informative and try and convince you of reputations that are unlikely to be deserved. What we would idealy like is a personal recommendation for every product and service we come across from people we already know and trust. We all have have, consciously or subconsciously, a recommendation network; people whose recommendations we weight strongly. It can be friends, friends of friends, famous celebrities or review websites. However, there are more things out there than the people we trust could ever give opinions on. It is in these gaps when we fall back on star ratings or pot luck. Wouldn’t it be good if we could instead fall back on something better? We need a new system for reputation.

A worthwhile reputation system needs to have several attributes. Firstly, it needs to be personal. This alone has several knock-on implications: it needs to be personally aggregated so that it takes into account an individual’s judgement on the weight that should be given to other people’s opinions or recommendations about things of importance and interest; and it also needs to be adapt and update when an individual’s preferences change. Such an intertwind model, where someone reputation relies on the reputation of those whose own reputation is affected by the first person’s as yet undetermined reputation, might seem like an impossible ‘chicken-and-egg’ problem, but mathematics comes to the rescue with matrix inversion algorithms and the like to sort us out can provide definite answers. Phew. Second of the attributes a worthwhile reputation system must have is the ability to acknowledge the range of domains in which reputation exists, in particular that being reputed for being a good judge of vacuum cleaners does not mean that judgement on other people’s ability to pick out good vacuum cleaners must also be high. Thirdly, it must be robust to fraudulent exaggeration or supression of reputation. These together almost make the holy grail sound unachievable. I know what you are wondering at this point: surely the computation required to determine how reputable everyone is in everyone else’s eyes on any given topic is emmense, and probably far beyond what the energy from the sun can ever allow us to compute. Fortunately, that level of computation will never be needed, as in reality, any individual only needs to get a trust score for a very small fraction of the total number of possible things available in total. Also, no one asks about everything at once, so in practice reputation calculations can be delivered in a just-in-time or on-demand style. Again, phew.

One particular domain where reputation is important to help the system operate is academia. Science is a well designed system and using it the truth will always win out, eventually, even when human nature (from cognitive biases to outright fraud) act to slow it down. However, we humans are an impatient bunch, living for the short time that we do, so speeding up scientific advancement with an efficient academic research ecosystem is desirable. Peer review was introduced to help towards this, but due to a variety of factors (e.g. the exponential increase of research output, the growth in the number of fields of research, the narrowing of research fields, the presure on academics to constantly produce high impact results, despite the presure from funding bodies to only persue directions already known to be valid and therefore necessarily unsurprising) the ideal of peer review is all but lost. Peers now rarely actually repeat the experiments that are being reviewed and the media often take as gospel academic papers or articles which have not been peer reveiwed at all. The impact-rated journal system, where the almost random process of being selected for publication in Nature or Science is necessary for continued funding, has become a self-reinforcing spiral of distortion, resulting in publications whose main motivation is itself a self-sustaining impact factor which may or may not support actual science and even then only as a bonus side-effect.

The example of academia brings attention to some extra attributes that our ideal reputation system should have: democratic independence (it should not be possible for a person or group to gain sway over the system to inflate their influence even more); openness (anyone should be able to join and everyone should be able to see how it is being administered); and interoperability (it should not be the case that reputation for plant science can not be influencial on cancer research becauase sometimes what have become narrow research fields do acknowledge others and benefit from cooperation).

At the present time, there is only one system technology that has all the attributes that we have seen to be necessary for the holy grail of reputation systems: immutable identities, democratic independence, openness, interoperability, robustness, computation ability and untamperable data storage. That technology is smart blockchain technology, of which Ethereum is currently the cutting edge. Reputation is a fundmentally important ingredient for a successful society, but all existing attempts to develop a system to manage reputation in today’s expontially expanding world of global trade are inadequate. I am excited that with smart blockchain technology, we finally have a technology that is capable of shouldering reputation systems worthy of our modern world, and a good way into the future too.

blog_header3

Dr Aeron Buchanan

Aeron Buchanan is a researcher for the Ethereum Project with degrees in Computer Science, Engineering, Systematics and Computer Vision.