SM: Based on the way you are approaching the problem, it seems much easier to do relevance in a constrained domain than in an unconstrained domain.
TC: Absolutely. Relevance is the idea of constraining things. There are huge things you can do with information. It is amazing when you look at search, especially the type of search Google came out with where you type in words and get back results, and as a whole the field is amazingly blunt. It is a brute force tool.
If you go back to the 70s, when people had ideas that were completely computationally impractical, those ideas are probably extremely fast to compute now. The vision people had back then in regards to what computers should be able to do was much more ambitious.
Computers have become more present in our lives since the Internet, but people’s ambitions for what they can do has slipped tremendously.
SM: People have tolerated much less than they should expect to.
TC: Which has resulted in an entrenchment of people not trying to make computers do more. One of the things is to get to a place where we can create and do different things. I learned from my time at IBM that the place to create is not at a big organization. Even Google has gotten to a point where innovation is slowing down.
SM: They have a business to run. They cannot do something radically different that disrupts proven success.
TC: You won’t see them really start innovating again until they have a few bad quarters. Don’t fix what is not broken.
SM: With Web 2.0, search changed from horizontal to vertical, which constrains the domain and increases relevancy. My thought is that with Web 3.0 we will take that further and get into real personalization.
TC: Personalization is really hard. There are two kinds of personalization. You can find personalized items, where you can have your initials monogrammed onto a particular item. A lot of personalization I have seen is very much like that and it is cheap.
If you get handmade shirts, that is another level of personalization. That is something that happens after everything else is right. Personalization to that level only makes sense once you have covered the basics and all of the other pieces are in place. In restricted domains you can get there.
SM: It will be very hard to get there in unrestricted domains.
TC: Especially for things like search, where it is only luck that search works. In some ways it really is luck. Computers have no idea what they are doing. They have amazingly little understanding of what is going on. People are willing to put up with a lot of randomness in that space.
Compare that with a domain like ecommerce, where there is an enormous amount of understanding as to what the products are and who buys them. In search we are still amazingly early on.
SM: Large-scale ecommerce is a search problem.
TC: Part of it is a search problem. One of the big problems with ecommerce in Germany is that people do not have credit cards. Germans do not have cards. They have a tremendous aversion to hyperinflation, and avoiding credit is a cultural issue.
Cross-border ecommerce in Europe is also facing problems because of cultural issues. There is a lack of trust of people in other countries. It is bizarre to me that I had to explain to a European commissioner how eBay works. The commissioner could not grasp the idea that random Americans trust other random Americans.