make those records available to a research group, or you want to make a subset of them available to a pharmacist. You maybe willing to share information about physical attributes, but you don’t to share information about mental health history. How do you give different groups differential access to information? That’s a policy question people would like answered, but it requires a technical implementation to deliver it.
Or in terms of telecommunications—I was reading your interview with Bill Mark at SRI. Many of the “intelligent agent” technologies they’re working on bring requirements for quality of service—a certain bandwidth that the world of devices will depend on for anytime, anyplace access to communications. With prototypes, we run directly into those quality-of-service issues, which leads to the policy side. If you don’t think about managing the spectrum available for those devices, it’s potentially going to make these kinds of experiences difficult or impossible to realize.
X: The wireless spectrum issue is interesting, because it’s pretty easy to predict that it is still going to be important and contentious 10 years from now.
DR: Yes. We’re going to have to embrace cognitive radio technologies. We can’t make more spectrum, but we can use the existing spectrum more efficiently, in a couple of ways. Obviously we can increase the density of cells and slice things up spatially. But there isn’t going to be any alternative but to do more nimble negotiation.
Right now, [buying a mobile device] is like getting assigned to one lane on the highway at the time you get your license, and that’s the only lane you can drive in, whether the highway is busy or not. That just doesn’t make sense in a world with an exploding number of wireless devices. There needs to be a much more nimble way to negotiate real-time access to that finite resource.
That one is definitely going to be a 10-year trend, both because the technologies are rapidly evolving, and because that’s a place where the policy side will take a decade or more to resolve. How we license spectrum, which parts are unlicensed, how one adjudicates priorities, the economies associated with that, and then all the issues around public safety and national security, balanced against consumer and business desires.
X: Last question. Microsoft has a big team of people looking at artificial intelligence. Do you think we’re any closer today to convincing AI than we were, say, in 1968 when Kubrick and Clarke dreamed up HAL?
DR: I’m sure you’ve heard the saying that once you figure out how to do something using computers, the rest of the world says, “That’s not AI,” so by definition AI is the stuff you don’t know how to do yet. They’re right at some level. We are not on the verge of the grand AI of science fiction. That is the holy grail, and we’re still looking for that. I think what has changed is the amount of data we have. This is one of the places where statistics becomes your friend. That has led to substantial amounts of progress in areas like natural language processing.
It would also be a mistake to dismiss the power of local sensors. One of the historical challenges we’ve had is inferring things from a scarcity of data. But this cloud of data is one of the consequences of the sensor explosion [including mobile devices]. We are just in the early days of what could be done at the point of sale in retail trades, for example. A lot of these near-magical “I know what you want” things occur simply because you are not the only one who wants that thing—there are millions of people who want that thing and are asking the same questions.
