Google is testing a system that competes with Wikipedia. There are two key differences in Google’s version. One is that authors of entries are identified. Second is that Google encourages competition among authors and will rank different pages about the same topic using some algorithm that it has developed. Another difference is that authors can put ads on their entries (surprise …!).

The second of these two innovations is the most interesting to me. The problem with Wikipedia is that it’s hard to judge the quality of entries unless you’re an expert on the subject, but if you’re already an expert, you probably don’t need to read about it on Wikipedia. To maintain high quality, Wikipedia depends on self-moderation by its authors — if someone writes some nonsense, another author is supposed to see that and correct it. Google, on the other hand, it not an expert about everything (yet …) but it is very good at ranking things. So instead of relying on a disorganised and decentralised process to rank results, Google generates a ranking using a centralised algorithm. Since most people already know that Google is good at ranking web search results, it can leverage this credibility to get people to trust its rankings of other things.

To me, it still seems to be an open question as to which process — centralised or decentralised — is better at sorting the good from the bad. For a decentralised system to work, it depends on the incentives of the individuals involved being aligned in the right way. This is hard to do, as websites like Digg have discovered. The ‘crowds’ are often quite good at sorting the good from the bad, but there are opportunities for gaming and other effects that let some bads get mixed up with the goods. A centralised process, on the other hand, perhaps lacks the flexibility to respond to changes, and is also prone to gaming once people figure out how the algorithm works, but may be easier to implement compared to setting up the right incentives for everyone.

In any case, it’s going to be a very interesting experiment to see which model performs better. The only problem is that Wikipedia already has a massive head-start, and I don’t know if authors of Wikipedia pages will want to switch to Google, even if they can earn ad revenues. On the other hand, since Wikipedia content is Creative Commons licensed, switching is more or less just a matter of copy and paste …

HT: CoreEcon

Recently there was some controversy on the Flickr photo sharing site about their decision to censor (ie remove from their site) someone’s photo of a child smoking a cigarette. Flickr (which is owned by Yahoo) has a policy of not allowing images of children smoking on its site. This seems sensible — smoking is bad and most people don’t want to encourage children to do it or glamourise it. Except in this case the photo in question was of a photojournalistic nature, showing realistic images of how children are affected by poverty. Should such an image be banned or not? It’s not an easy thing to decide.

That debate is too controversial, so I don’t want to get into it. But it did get me thinking about what “Web 2.0″ sites like Flickr, Facebook and others actually produce, in an economic sense. They don’t regular goods like apples and cars. They produce communities. An essential part of any community is the standards or ‘rules’ by which its members must abide. Therefore, the following two activities are essential features of the production process of a Web 2.0 business:

1. Set the rules
2. Enforce the rules

Both of these things could be somewhat complex tasks, and are likely to significantly affect the costs and revenues (and hence the profits) of such a business. In terms of setting the rules, different potential users of the community (ie its potential customers) will have different preferences about the rules. Some people won’t mind seeing pictures of children smoking, while others will hate it, and others will fall somewhere in between. In some sense this is a similar problem to a regular business choosing the quality of its product — some people prefer high quality even if it means a high price, while others prefer low quality at a low price. However, there is a crucial difference for a business that creates an online community. What people really care about is probably not the rules per se, but the activities of others in the community. Thus the community operator cannot just simply say “no photos of children smoking” unless it also enforces that rule. In other words, the operator can set whatever rules it likes, but the standard of behaviour that actually arises will be determined by how well it enforces those rules (the standard of behaviour is endogenous, if you want to use a fancy word).

Thus the value of a community to its users, and how much they are willing to pay for it if it charges subscriptions like Flickr does, depends not only on the rules that the community operator sets, but also the enforcement of these rules. In terms of rules, we could think of them lying along a continuum from “anything goes” to “very strict”. Under “anything goes”, anything at all would be allowed and no enforcement would be required. This is the policy of the infamous Japanese 2channel Internet forum, where people can post messages anonymously and without fear of censorship. As the rules become more strict, enforcement costs are likely to rise, as the rules need to be monitored by system administrators somehow, and appropriate actions taken. For example, Flickr and eBay have staffs of people dedicated to policing the rules on their sites.

However, as rules become very strict, enforcement costs might decline again. The reason is that relatively ambiguous rules leave more room for interpretation, and are more likely to lead to disputes. In the case of the Flickr picture of the kid smoking, after people protested on its site, Flickr reversed its decision to ban the photo in question. However, leaving interpretation of the rules open to judgment like this is more likely to lead to controversial decisions, which may be time-consuming and costly to resolve. Strict rules may be easier to enforce — just delete all pictures of children smoking with no further discussion entered into. So the costs of enforcement as a function of the strictness of the rules might look something like this:

In terms of implementing enforcement, online communities have a range of options. An interesting question is how much of the enforcement should be ‘centralised’ and done by the system operator’s staff, and how much should be ‘decentralised’ to the community members themselves. For example, in addition to its in-house ‘police force’, eBay relies on user reports via its feedback mechanism to help prevent fraud. As long as members care about their reputation on the site, this kind of reputation system allows the operator to ‘outsource’ some of its enforcement of the rules, and probably reduce the costs involved. The downside is that reputation systems may be susceptible to gaming or manipulation by the members themselves.

To sum up, Web 2.0 sites produce communities. Part of this process (part of the “production function” in economic jargon) is to decide the rules for standards of behaviour by community members, and to enforce the rules. The rules that are set will affect the costs of enforcement, and the level of enforcement that the operator chooses will affect the actual behaviour that occurs, which will in turn affect users’ willingness to pay to use the site, or the willingness of advertisers to pay to advertise on the site. In addition, some or all of the enforcement function can be “outsourced” (or “crowdsourced”) to the community members by using a reputation or feedback mechanism. Since the production of communities departs from production of ordinary goods and services in significant ways, I think it will be an interesting area for economic research.

It really helps to have influential friends. Apparently, Steve Fossett’s buddy Richard Branson got in touch with his buddies at Google and asked them if their satellite images could be used to help in the hunt for Steve who has been missing for almost two weeks after his light plane disappeared in Nevada. Subsequently, a set of fresh high-resolution satellite images were taken of the relevant area after he disappeared and uploaded to Amazon’s Mechanical Turk service.

Mechanical what? The Mechanical Turk is a website that allows people to collaborate on tasks that only humans are good at (at present). They call it ‘artificial artifical intelligence’. For example, looking for Steve’s plane in thousands of satellite photos. I tried it myself and found that I can check an image in less than 10 seconds. A computer may be able to do it faster, but computerised image analysis is probably not as accurate as human eyes. The Mechanical Turk site facilitates projects like this, where many people can contribute a small part of a larger data-processing project that would be difficult or impossible to program into a computer algorithm.

The Mechanical Turk includes a payment system where you can get paid for each little bit of work that you do. When searching for Steve you don’t get any reward except the warm fuzzy feeling of doing a good deed. But for other projects on the site you can get paid small amounts (usually less than 10 US cents) per small task that you do. Still, at say 5 US cents per task you can make US$10 per hour if you can complete each task in around 18 seconds. For people in developing countries that’s a pretty good wage, and that kind of speed seems feasible for relatively small well-defined tasks.

I have a couple of comments. The first is quality control. Since people get paid by the quantity of work that they do, they have an incentive to do it as quickly as possible. In the search for Steve, they’ve set each image to be examined by multiple people before it’s eliminated from the search set. That’s one way of doing quality control — get multiple people to do the same task and compare the results. However that could be expensive when you have to pay people to work. Alternatively, I’m not sure if Amazon does this already, it might be useful to incorporate some kind of reputation system for workers. If people can establish a reputation for high quality work, and if this could be linked to how much they are paid, then maybe people would have an incentive to work accurately in the first place. It’s a matter of setting the right incentives for speed versus quality.

My second comment is on the design of the site, which I think is bad. As with Amazon’s main website, it’s not very well laid out, and most importantly it’s not optimised for speed. I could review Steve images much faster if the site design were changed a little. For example I had to scroll down to see each image and mark whether it contained anything interesting or not, and then scroll up again to submit the result. The important buttons are just too far apart, and this slows down the work speed. Also the entire page has to reload between images, which is slower than it should be.

Overall, I think the Mechanical Turk is a great idea, provided that the quality control problem can be solved and the site is optimised a bit further. It’ll be interesting to see if this kind of outsourcing (’crowdsourcing’) can be economically viable, or whether the costs and difficulties associated with generating quality work are too great.

If you want to help look for Steve, click here.