Computers that mimic intelligence

Do computers need 'brains' to function intelligently?

In the 1950s AI researchers figured that if a toddler can learn to speak and understand, it should take only about five years to teach a computer the same skills. Fifty years on, there’s been little progress, so I won’t even try to guess when, if ever, software will emulate or embody the linguistic talent at the core of human intelligence. Of course we don’t want rivals that can ace the Turing test. We want agents that can help us manage communication, organize tasks, analyze data, and perform transactions -- the very functions Apple famously imagined in its 1987 concept video, Knowledge Navigator.

In the same year Apple produced that video, I interviewed Georgetown University’s Michael Zarechnak, one of the early pioneers of machine translation. Progress, he said then, had been “more a matter of perspiration than inspiration.” If there’s going to be a quantum leap, such as the “singularity” that maverick inventor Ray Kurzweil expects around 2020 -- when Kurzweil suggests we’ll have the raw computational power to scan and emulate brains -- then all bets are off. Otherwise, we should plan conservatively for a long series of three-yard gains ending in clouds of dust.

This year’s remarkable success of Bayesian spam filters hints at near-term future developments. There’s nothing resembling intelligence in my spam filter, only a database of word frequencies. It works magically well because the filter observes, adapts to, and assists my own intelligent behavior. In five years I expect such technology will monitor and assist with more kinds of personal information management chores, as well as with collaborative efforts and transactions on Web services network. In 10 years, I expect voice and video to be among the data types available to my assistant. In 15 years. I hope to enjoy something similar to the scenario Apple envisioned back in 1987. We don’t have to endow computers with intelligence in order to make that vision real. I’ll settle for a solution that structures computer-assisted work so that computers can seem intelligent by patterning their behavior on ours.

Click for larger view.