Autism is a heart-rending mental illness that has become entwined with a contentious social issue. While neither is conducive to easy solutions, some Yale University researchers are hopeful that the clever use of technology might prove helpful to both.

Though there is a range of manifestations of autism, the disease is generally regarded as involving impaired social skills. Autistic persons, for example, usually can't engage in normal conversations. The cause isn't known; most researchers believe it involves brain biochemistry, and the only current treatment, often only minimally effective, is behavioral.

The recent sharp increase in reported cases of autism has created a controversy. Is there actually now more of the disease? Or is it simply better understood, and thus more commonly diagnosed? Are the numbers further swelled by parents demanding an autism diagnosis for troubled children, bringing with it a chance at better access to social services?

Brian Scassellati is a robotics researcher in Yale's computer-science department, and is part of an interdisciplinary group on campus that includes doctors and others. Part of his contribution has been to build very simple robotic heads -- more like smart toys -- then to watch how different children, autistic and nonautistic, respond to them.

These devices can be programmed to monitor where the child is, or whether the child has said anything, and then to say something appropriate. In other cases, the robot head will spout things randomly. Prof. Scassellati said that with three year olds, nonautistic children will continue to interact with a robot that is responding appropriately, but will quickly tire of one that isn't. Autistic children, however, show no such preference, and will be equally fascinated by each.

It turns out, he says, that there are a number of other ways that autistic children respond differently around technology. One of the most striking, and potentially most useful in terms of diagnostics, involves eye gaze, which can be tracked with special machines.

Nonautistic children shown a movie of two people talking will usually go back and forth, looking at the eyes of each speaker. Autistic children, though, consistently look at other things, like necks and hands. Using a film showing a child playing, nonautistic children fix their gaze on the child, while autistic children commonly stare at the empty basketball court in the background.

The group at Yale, as well as groups at other autism research centers using technology, are attempting to quantify those differences, with the goal of developing objective diagnostic tests. Currently, diagnoses are made by doctors after interacting with children; Prof. Scassellati says different doctors often will come to different conclusions, if only because the child is in different moods when examined by each physician.

It's not known how early autism can be detected. Prof. Scassellati says one of his goals is to develop a gaze-based diagnostic approach that would work with children as early as one. Current autism tests involve social skills, and so can't be given until a child is two or three and capable of some basic interactions.

Yet, as with most diseases, the earlier the diagnosis, the better the chance for effective treatment. A standardized set of diagnostic tests might shed light on the question of whether the disease really is on the rise.

Prof. Scassellati says his group's work so far is devoted entirely to diagnosing autism -- not to treating it. (Even with that more modest goal, the results aren't in yet.) But he and others can't help but think of the ways that robots might be useful one day in actually treating the disease. That would mean taking on some of the behavior-based conditioning that, in the absence of a solid medical explanation for autism, is now state of the art.

Working with autistic children can be exhausting, notes Prof. Scassellati, but machines don't tire. "It is hard to focus on eye contact if the kid is standing a centimeter away from you," as autistic children do, he says. "But that would be very easy for a robot to do."

Researchers at Yale, and many other places, are designing robots and tools such as videogames to teach socializing skills to autistic children. The robot maker's standard tool kit by now has all the technologies needed for a computer-controlled machine to interact with a child for a period of time.

The machines are no panacea. With human teachers, autistic children often don't apply something learned in one situation to a related situation. Prof. Scassellati says it's not yet known if things will be any different with robots.

Considering the pressure some parents place on doctors to get an autism diagnosis, one wonders if parents could somehow "coach" a child into acting a certain way during a automated test. Prof. Scassellati says no.

"These are very basic social abilities that we are looking at. If you have them, it's hard to hide them. And if you don't have them, it's easy to notice they are missing."