News
Improving communication between man and machine
Psychiatry professor Werner Ceusters opened his UBThisSummer lecture yesterday with a clip from the TV show “Star Trek: Voyager” featuring The Doctor, a holographic computer program that could diagnose and treat patients, conversing with them to gain an understanding of their ailments.
Such technology is still the stuff of the future.
But developing machines that can understand what physicians and patients say is not an impossible feat, Ceusters told about 80 people who attended his presentation, the second in the annual summer lecture series.
A major problem with today’s technology is that many electronic health records systems are difficult for a computer to understand because the medical information they provide is not precise, said Ceusters, director of the Ontology Research Group in UB’s New York State Center of Excellence in Bioinformatics and Life Sciences.
A records system might use a numeric object code, for instance, to identify types of injuries—fractures of the shaft of a left femur, for instance.
But if this object code is used twice in a patient’s record over a period of two months, a computer would not be able to infer whether the code referred to the same exact fracture, or if the patient suffered the same type of injury on two separate occasions.
The system’s usage of codes to identify types of disorders, rather than particular instances during which disorders occurred, leaves room for confusion.
Ceusters offered another easy-to-understand example of why computers need precise information to understand a problem.
He presented a PowerPoint slideshow that contained the following information: “The surgeon examined Maria. She found a small tumor on the left side of her liver. She had it removed three weeks later.”
A person reading that paragraph would deduce that the surgeon had found a tumor and that Maria had it removed three weeks later.
But for a computer, the meaning of the words would not be so obvious, Ceusters said. A machine would not know whether the tumor or liver was removed, or whether it was Maria or the surgeon who underwent the operation.
The paragraph contains “lots and lots of ambiguities that the machine has to deal with,” he said.
The key to improving communication between man and machine, then, is to find ways to eliminate ambiguities: “I’m trying to build systems in which I can build a digital copy of the world,” he said.
In the medical field, that entails developing electronic health records that assign unique codes to each unique disorder a patient suffers from.
Ceusters’ ideas regarding communication between humans and computers are rooted in his work in ontology, which he called “the study of what exists.”
He described himself as a “philosophical realist,” meaning that he believes that reality exists objectively in itself, independent of the perceptions of beliefs of cognitive beings.
In layman’s terms, “We see cows and tree because they are there. They do not exist because we are seeing them,” he said.
Humans use words to describe reality, but words are difficult to analyze because the same word can conjure different images of reality, Ceusters said.
Search for pictures of moles online, for instance, and you will find photographs of stuffed animal moles, facial moles and mole, a sauce used in Mexican cuisine, he told his audience.
In the same way a word can have many meanings for human beings, an object code in an electronic health records system can have many meanings for computers.
The best system, then, according to Ceusters, is one that focuses on the realist’s reality—one that uses object codes to identify specific problems that exist, leaving no room for confusion.
Reader Comments