Ask the Expert

Last year, I took a sabbatical at what was then considered a small social networking company, Twitter. I designed the initial algorithms for many of their algorithmic products and learned more than I would have thought possible in the span of a year. I am now on their technical advisory board.

There are several trends in social networking that have real staying power. The first trend is personalization. Facebook, LinkedIn, and Twitter all have explicit products which recommend friends and connections based on your interests. These products are immensely valuable in driving growth, and are poised to become even more so: Humans are diverse creatures, and if we follow the online activities of a large set of friends, we are likely to find that much of the information we receive is not useful or interesting. In other words, we lose “precision.” If on the other hand, if we are very selective about our online friends, we will miss out on important information in our extended friend network, as well as national and global events of interest to us. In this case, we have low “recall.” This is where personalization products will play an increasing role. They observe and learn what we find interesting and useful and triage information from our social network before it gets to us, ensuring both high precision and high recall.

The second trend is new models of collaboration and information exchange. On Amazon’s Mechanical Turk, a system for crowd-sourced micro-work, many employers issue a task to two workers, and automatically check their answers against each other. Groupon, a deal aggregator, creates economies of scale where there was no natural mechanism of getting people together. Twitter allows the flow of information upstream, from regular users towards corporations and celebrities. On Quora and Stack Exchange, experts band together to answer thorny questions. Much innovation has happened here, and more is on its way.

The third trend is innovative market design. Groupon is an example of networked commerce. Prediction markets such as Intrade and social commerce sites such as eBay and craigslist represent novel peer-to-peer markets. In my research group, we are trying to develop an algorithmic and socio-economic understanding of these emerging models.

But this does not answer the question, “Is this all a fad, or even worse, an end to rich social relationships which will now be measured out in superficial 140 character coffee spoons?” My belief is that social networks will indeed have a transformative positive impact on human society. They remove friction from relationships by eliminating the need for physical and temporal proximity from communicatoin. They allow us to discover relevant information. And they allow us to collaborate in novel ways. It is the latter which I believe represents the richest medium to long term direction for academics and budding entrepreneurs. Two examples are politics and education.

In politics and policy making, social networks could allow a large population to form consensus opinions on issues such as health care policy, social security reform, and taxation. In other words, we could benefit from a social network X where X is to Washington what YouTube is to Hollywood or Twitter is to print journalism, an emergent distributed network that augments and rivals the more mainstream centralized institutions. Most online discussion boards today devolve into name calling and flame wars when discussing contentious issues. We are doing research on social networking primitives that will promote consensus building, but more effort is needed in this direction from more groups.

In education, the story is very much the same. We need a system of peer learning and tutoring that leverages recent advances in social networks to deliver quality college education at unprecedented scales at a fraction of the current cost. This would be a major boost to much of the developing world.

The above questions may sound philosophical, but as soon as you dig a little deeper, they involve rich mathematical, algorithmic, and empirical work. Consider a peer-to-peer market such as craigslist, and imagine that we have the Facebook or Twitter id of each participant. How can we automatically infer trust relationships between a buyer and a seller who don't know each other directly? This calls for economic models of assigning trust and algorithmic approaches to calculating it efficiently. Designing web-sites for collaborative policy making will require incentive mechanisms for rewarding consensus and preventing gaming, as well as algorithms that can identify extreme opinions and prevent them from escalating into a back-and-forth frenzy of insults and name calling. We are excited about these research directions; if you would like to learn more, visit my Web site at www.stanford.edu/~ashishg or feel free to contact me at [email protected].