With all these talk in recent years about crowd sourcing and big data, I wanted to delve a little bit deeper on the costs of crowd wisdom. It is easy to see how crowd wisdom can be beneficial to society as a whole; for example, there are many crowd volunteering websites that aim to aggregate crowd decisions for social good. One of them is Tomnod which rose to fame when they used satellite imagery to crowd source data on where the missing MH370 plane is. However, in subtle ways, crowd wisdom might not be all that good.
In one of the more telling problems of crowd wisdom (herd mentality in particular), Sunil Tripathi was implicated as a suspect in the Boston marathon bombings because a Reddit user posted in the forums that they looked alike. Soon after, more people started agreeing with the first poster, and spread like wildfire over Twitter. It went into mainstream notoriety when mainstream media reported it over TV and newspaper. The lack of fact-checking by the latter was disturbing. However, this shows one thing – that crowd wisdom can turn mad. Critics of this point might point out that crowd wisdom works best if people are independent – they have no knowledge of other’s ideas. While this can be replicated in a laboratory setting, the fact is that it is difficult to ensure this online. People invariably will use search engines or share it on Facebook, which leads to a decoupling between reality and ‘reelity’. It is what makes crowd wisdom a dangerous idea to wholeheartedly believe in.
Secondly, the issue of privacy comes in. In recent years, due to the proliferation of smartphones with embedded sensors, app developers have been using the sensors to predict weather or traffic. Dark Sky, a weather app on the Apple App Store, uses pressure sensors on the latest iPhones to crowdsource data on whether it is going to rain in a particular region. Waze, a traffic directions app, uses your phone’s speed and location to crowdsource data on traffic conditions. If you are moving slower along a particular road, the app may intelligently tell other road users that a traffic jam might be imminent, and to reroute road users accordingly. It seems beneficial until you realized the same app can be used for nefarious purposes. Previously, Tinder, a dating app, uses one’s locations to tell potential dates where you are. Whilst other people see you as “3 miles away” in a vague sense of distance, some tech-savvy people figured out a way to perform deep-packet inspection of the traffic to find out your exact coordinates. It turns out that the actual coordinates are actually being sent to your phone, which then processes the data in conjunction with your location to provide you with the distance. Hence, it is possible to know where you potential ‘dates’ live, and possibly stalk them. This is a design flaw more than anything else though, but it still shows the flaw of crowd wisdom – when companies attempt to collect everything that it is possible to gather from a particular user, the possibility for irresponsible disclosure is heightened.
Thirdly, it might not be even accurate at all. As seen in class, the more variables there are in a particular prediction, the less accurate it becomes. For example, predicting how many leagues in the depth of the Marianas Trench requires people to have some notion of a league, in addition to predicting the depth of the trench. A more popular example which deals in big data was Google Flu Trends, which was recently shut down. Although it wowed researchers by predicting flu trends ahead of a full-scale epidemic, it constantly overstated the severity of it. People were too prepared, it seems. Whilst this might not be a bad thing, I can see the flaws of trying to extract information from crowd data – it might be fairly accurate now, but might not be a few years down the road.
Finally, I would like to end off with some food for thought. How do we ensure that crowd wisdom is indeed accurate? We might take a look at the intricate system Wikipedia has in place for its editing – talk pages, a system of editorial processes and its ilk. Is it possible to generalize for online communities and spaces?