TR: Where else might your work help?
JC: I think that recommendation systems are going to be as important as search algorithms [see “Recommendation Nation”]. In a recent piece of work, we came up with a list of desired properties for a recommendation system, and what we ended up doing was proving mathematically that there is no possible recommendation system that has all these desired properties. So I would have to choose which properties I am willing to give up and design recommendation systems that preserve the properties I want most.
TR: What kinds of properties?
JC: There’s transitivity. If I trust the recommendation of person B, and person B trusts the recommendation of person C, then I should trust the recommendation of person C.
TR: What about privacy? Can recommendation systems still let users keep control?
JC: Those are exactly the kinds of questions that we’re asking. We didn’t consider privacy in our work, but one could definitely add privacy to the list of properties, and then it might be possible to come up with a theorem saying, for example, you can’t have a recommendation system that will deliver all the information you want and have all the privacy.
TR: How might this research change the way we use these applications?
JC: It could be that at some point somebody could go onto a social network and say, “Here are the properties that I want for my recommendation system,” and a different person could go in and say, “Here are the properties that I want,” and they could get two different recommendation systems. In a similar way, search engines have been around for a while, but I think they’re still very far from exactly what we want, and over time we’ll be able to come up with search engines which are much more personalized. Then we would also have to figure it out on the back end. Can we accommodate all these different algorithms? I hope that at some point computations would be done differently for your search engine and recommendation system and for mine.