The next best thing to a brain?
Will computers ever be as clever as us? A DCU researcher believes so, reports Dick Ahlstrom
Television and films are littered with clever and articulate computers, but these systems are nowhere to be found in the real world. A thinking, reasoning, even emotion-feeling computer might become a reality, however, if artificial intelligence researchers join forces.
Such is the goal of Dr Mark Humphrys, a lecturer at Dublin City University's school of computer applications. He is setting up an Internet-like network of artificial intelligence research data that he calls the World Wide Minds.
"The World Wide Minds project is central to what I have been doing for the last few years," he says. It revolves around a single issue: how artificial intelligence (AI) is going to scale up.
Dr Humphrys has a team of three researchers at DCU and another at Dublin Institute of Technology. Together they are beginning to build their network. The idea is to put individual AI research elements online, making them as readily accessible as the billions of pages of data and software currently residing on the World Wide Web.
"One of the things about artificial intelligence is its long history of false dawns and great ideas that didn't pan out," says Dr Humphrys. Two reasons why progress has been slow are the complexity of achieving this goal and the way the AI community operates.
"It is simply impossible to make a machine that has intelligence," says Humphrys. How can we get a computer to imitate thought when we barely understand how thought works in the mind, he asks. "We don't really know what the internal dynamics of the mind looks like."
The AI community also slows progress, he believes. A great deal of work has gone into, say, developing machine vision, speech recognition or machine learning as individual elements. Yet these elements are never brought together into a single computer entity. The research stays separate, as do the software and systems that run the projects, as they use incompatible languages.
Yet specialisation is essential for progress to be made, he believes. "In the future of artificial intelligence, researchers are going to have to specialise in particular areas - and they are going to have to link all of their work."
This is the purpose of the WWM. It will allow researchers to gain access to other researchers' endeavours in the same way that you can access remote computers and data over the Web. He and his team are defining ways for people, computers, software and data to speak the same language - in effect, developing an Esperanto for AI researchers.
Another impediment that a common language could help iron out is the long-standing "tension" between AI researchers and the engineering community. The former try to build something autonomous while the latter delivers solutions to particular problems, such as speech recognition. Both aspects are necessary to advance AI technology, however.
Humphrys admits to placing himself among the former group. They produce "animats", short for animal-like robots. They are very simple but entirely autonomous computer entities that have "lives" of their own and respond to their environments in a rudimentary but coherent way. "My background is with the artificial-life people and models based on numbers, models of the mind that involve numbers."
This camp holds that if you learn how to build successful but very simple animats, then improvements and more research will lead in time to cleverer animats. Complexity of behaviour would be increased.
The WWM would merge the disparate research groups, allowing their work to combine into AI complexity. Humphrys likens it to building a city such as London or New York. Creating a city involves "many thousands of people working over time, none of which sees the big picture, yet it is there".
The Internet grew in the same way. Many people know a few pages or a few sites intimately, but nobody knows the entire Web. Yet the Web exists and it works.
"Humanity collectively can build bigger things than individuals working separately," says Dr Humphrys.
He and his team have begun to assemble "worlds", or individual problems to be surmounted, and "minds", or solutions to problems.
The goal is for his WWM to allow remote researchers to make use of other people's minds and worlds to create new minds that can tackle tougher worlds. "The idea is you can construct larger minds," says Humphrys.
The approach should allow faster progress, as researchers won't have to reinvent components as they attempt to solve a problem. They will be able to build on the work of others and, finally, begin to take artificial intelligence to a new level.