If these walls could talk

FLUENT COMMUNICATIONS

Winograd and his students are now moving on to specific applications. For example, they are trying out the interactive table on people with Asperger's syndrome, a type of autism. Such people often are whizzes at using computers but poor at working with other people; the hope is that the table might help them improve their social skills.

Winograd says a key impediment to users' acceptance of rich, multimedia spaces is "disfluency," something IROS was designed to minimize by making the technology as seamless and transparent as possible.

"Whenever you have to stop focusing on what you care about to focus on how the machine is doing it, you lose fluency," Winograd says. "In every design of interaction, that's a concern you want to bring to it."

BUILDING AN E-OFFICE

Several years ago, IBM teamed up with Steelcase in Michigan, U.S., to develop BlueSpace, a prototype office that integrates sensors, actuators, displays and wireless networks into an office space. Guided by user commands via touch screens, these gadgets can dynamically optimize heating, ventilation and lighting, and control interruptions depending on the perceived behaviour of the occupant.

The prototype room even has a "steerable" projector that can project images onto any surface in the office. Aided by a sensor in his desk chair and other detectors, the occupant's incoming e-mail is sent by the projector to the surface most likely to be visible to him at a given moment.

The IBM/Steelcase prototype is intended to prove the viability and value of the concept, says Stefan Hild, senior manager for responsive enterprise solutions at IBM Research. The prototype serves those two purposes, he says, but because it requires specialized hardware and furniture and a high level of integration, the system is not practical for most companies.

"The investment of taking an office building and enabling it that way is fairly high," Hild says. "But you can get 80 percent of the benefit with 20 percent of the cost." A follow-up project at IBM called mySpace aims to give office workers greater control over their work spaces, mostly using technology already in many office buildings.

"So if you have a wireless LAN in your office, it implicitly also gives you a location system because you can get the 802.11 endpoints, such as your [laptops], so I get the location of my co-workers," Hild says. "And you may know voice over IP as a telephone system, but VoIP in your office building also gives you a level of presence information. And if it's integrated with my calendar, you can use the system to figure out when I'm available."

DATA OVERBOARD

But widespread deployment of smart spaces faces two tough challenges, Hild says. First, such projects tend to incorporate scores, if not hundreds or thousands, of devices, each capable of generating torrents of data. Scaling up networks, central processors and storage systems to handle that is an obvious challenge, he says.

"But a much deeper issue," Hild says, "and one that will be with us for many years to come, is that our IT systems today are optimized to process data extremely reliably but not necessarily very fast." In contrast, smart environments generally must process sensor input in milliseconds. A collision-avoidance system in an airplane must react immediately. A package bearing an RFID tag moving across a receiving dock must tell a worker within a second or so where to put it, not five minutes later.

"Marrying this real-time world with the transactional world is the real challenge we are facing," Hild says.

THE DARK SIDE

Hild echoes the concerns of many researchers about the reliability of increasingly complex, distributed and heterogeneous systems. "One of my worries is we are not building circuit breakers into integrated business processes," he says. "What will happen eventually is that a faulty sensor reading somewhere will have a ripple effect through the network and cause untold damage somewhere else. We need something that knows what normal is and, if it's widely outside the norm, raises an alarm."

"Reliability is a major, major issue that's not being recognized at this point," agrees Leonard Kleinrock, a computer science professor at the University of California, Los Angeles, and one of several engineers credited with creating the Internet. "The complexity is essentially out of hand. We used to think that the systems we built we could understand, but that's no longer the case."

And things will get even more complex as we move to smart homes, offices, factories, stores and automobiles, Kleinrock says. "We are deploying on the order of 1 billion microcontrollers a year, in the car, in refrigerators and so on. But they haven't yet fully engaged the end user, and they are not networked. But I see that coming. It's going to surprise people when they begin to interact with these devices."

Scott Hudson's studies of human behaviour are aimed at making smart environments more acceptable to their occupants. Hudson, a professor of human-computer interaction at Carnegie Mellon University, says people have traditionally thought that the interruptibility of an office worker depends primarily on two things: how deeply the worker is engaged in an important task, and whether the worker is socially engaged with another person, as in a phone call.

"But it turns out in our studies that it's much more about social engagement," Hudson says. He says his experiments with voice detectors in workers' offices show that a system can predict the workers' willingness to be interrupted at any given moment with 85 percent accuracy.

Asked about machine learning and statistical models that might eventually tune smart offices to the habits of each unique occupant, Hudson says he has been surprised by how well certain techniques work across groups. "But in the end," he says, "I think we'll want to go to individual models for that extra 5 percent or 10 percent of accuracy."

But isn't 90 percent good enough? "Well, you can say it's pretty good, or you can say no, it's wrong one time in 10," says Hudson. Many users find that kind of experience frustrating and unacceptable, he says.

GROUNDED

You enter an airport terminal and your airline's computer detects your presence because your cell phone is location-aware. And since you're carrying the airline's RFID- enabled frequent-flier card, you are checked in electronically and issued an electronic boarding pass. The system sets a flag at passport control so that when you approach, your passport image is electronically displayed. You board the plane without ever touching a piece of paper or a screen.

An airport circa 2010? No, Swissair had just such a system in operation at the Zurich airport 10 years ago. Then came 9/11 and a downturn in air travel. Swissair went out of business.

But there were other obstacles to wide deployment of smart air terminals, says IBM, which built the system. "It was limited to one airport and one terminal for a single operator, and getting it deployed in other settings was difficult from a coordination point of view," says Stefan Hild, a researcher at IBM. Airlines were forced to cut costs after 9/11, he says, and there was "no appetite" for that kind of system.