A learning computer that evolves has long been the goal of many scientists. Phil O'Connell reports on pioneering research that may in the future lead to intelligent machines making some very important decisions
For years researchers have been looking at ways to make computers more personal, to give them a "thinking and evolving mind".
A UCD research scientist has taken an important step on this road by creating a number of programmes that can think and adapt in a way similar to a two-year-old when learning certain simple tasks.
The work by Dr Nicholas Kushmerick of the Department of Computer Science represents the beginnings of practical artificial intelligence (AI), "replicating human level expertise" through computer machine code.
We haven't progressed in certain applications of AI - for example, in our use of the Web. If you log on to a search engine the computer assumes you are an "average" person with average interests, explains Dr Kuskmerick. "We are designing learning programmes that will eventually understand the user's interests and adapt to them over a period of time," he says.
The two main parts of the research concern information extraction and personalised information retrieval (PIR). The goal is to improve the response given by the computer to your search through the Web.
The team is working on the information extraction to enable the computer to respond to shorthand requests. For example, using this technology you could ask the Web search engine: "I am going to Malaysia from Dublin. How do I get there?" The programme will come up with flight details, accommodation options and so forth, all relevant to the request with no excess information.
The second area, PIR, is of major international interest, says Dr Kushmerick. The aim is to cut out the tedious side of retrieving information from the web.
A financial analyst might know, for example, that IBM bought a smaller company for €100 million. By asking the computer a direct question one should be able to find out what other companies are buying around the globe, says Dr Kushmerick.
Another planned project concerns a computer-based training programme designed to adjust itself to the needs of the individual rather than the "one size fits all" approach that exists today. Computer tutoring is a multi-billion euro industry and offers great opportunities for the right kind of software.
Dr Kushmerick's work - aided by international collaboration with other computer scientists - involves assessment of a student's progress, looking for any area a student might find challenging. The software will then alter the course material as appropriate.
"This tool is designed to help instructors in a factory setting - for example, if an employee is having problems operating certain machines. It will assess why and perhaps propose moving this individual to a new area more suited to that person's strengths," he explains.
Another element of the research under way at UCD certain to interest many professional people is e-mail management. "E-mail used to be about connecting people. It is not any more, as high-level management companies don't read their e-mails anymore," says Dr Kushmerick. "Part of our project concerns developing a programme that will filter out excess junk mail and other stuff."
The UCD research team has received a five-year €1 million grant from Science Foundation Ireland to complete work on these next generation "intelligent" information services. The work undertaken at UCD represents only a small piece of the puzzle to build computer programmes that can think for themselves. "The level of learning capacity we are developing here would only be about 2 per cent of the learning capacity of a two-year-old child, but at the same time upmarket people in the US hire people to do their shopping for them," states Dr Kushmerick.
"I would estimate that, through the suite of techniques developed here, we could perform 80 to 90 per cent of the tasks that those personal shoppers do (if based in a Web scenario)."