Machine learning coupled with 'rich interaction' may make computers a partner, not a product
Researchers at Oregon State University are hoping to improve artificial intelligence with a project the uses "rich interaction" to teach machines when they make mistakes.
Their work would allow for ordinary users who spot a computer's errors to be able to step in and explain directly to the machine the logic it should be using.
The scientists claim that the project is based on an idea that is one of the latest advances in machine learning and artificial intelligence-- A computer that not only learns from its own experiences, but also listens to the user, tries to combine what it "hears" with its internal reasoning, and automatically updates its code in order to avoid making the same mistakes again.
The result is a computer that wants to "communicate with, learn from, and get to know you better as a person," say the OSU scientists.
"We want to develop algorithms that will allow the end user to ask the computer why it did something, read its response, and then explain why that was a mistake," says Weng-Keen Wong, a computer science professor on the project. "Ideally, the computer will consider the response and change its programming to perform better in the future. It's like debugging a program."
The researchers say that many advanced learning systems begin learning the moment they are delivered to an end user's desktop in an effort to customize themselves to the end user. Such systems are the basis of spam filters on personal computers, e-mail sorting, product recommendation. And most are based on word statistics, set rules, and similarities, and other fuzzy logic. But the researchers point out that even the most advanced systems only allow a user to tell the computer something is right or wrong, and not to explain why.
They say that a big challenge is to design interactive systems that are easy enough for everyday computer users to operate, and not just programmers.
"In the future we believe the computer should be like your partner," said Margaret Burnett, an associate professor of computer science at OSU. "You help teach it, it gets to know you, you learn from each other, and it becomes more useful."
It's an intriguing concept and a worthwhile research direction for AI, but the connection between the algorithms and the intelligent user interfaces that will give computers the ability to self-customize to the needs of the end user, isn't entirely clear (voice recognition?). And some examples would be helpful.
Machine learning systems may become more proactive once user input is added, and if coupled with data from locally networked sensors it would be able to factor in your behavior in a given environment into the equation to anticipate your next steps and needs.