COMPUTER INTERFACES: James Lileks mentions computer interfaces in his Bleat today, and I wanted to comment briefly since this touches on one my my major interests. In his final three paragraphs, Mr. Lileks says:
If you opened your desk drawer and saw a floating maelstrom of recent bank statements and insurance bills, it wouldn’t help you find a receipt from 2001. You’d want to say the name of the paper you requested and have it pop out of whatever folder you’d put it in.
So either we breed supersmart hamsters that can throw up the relevant piece of paper on demand, or we work on voice recognition, or we learn to parse marching index cards.
I’ll chose the middle option. I love my computer. I already talk to it. How sweet will be the day when it listens, and replies.
He may not realize that the type of performance he desires is not dependent on mere voice recognition, but will also require the development of supersmart hamsters -- useful AI. Voice recognition is reletively mature, but the AI required behind the scenes to yield such spectacular results does not yet exist.
In the process of earning my PhD in artificial intelligence, I've refined my expectations for the future of AI. Sure, machines like HAL or Johnny 5 would be fascinating, but I don't think the true promise of AI lies in self-aware machines. Consider instead HAL's initial job description: it was designed to add a layer of abstraction to the otherwise complicated task of operating the spacecraft. The task would otherwise be impossible for two astronauts; the International Space Station requires a crew of three just to stay in low-earth orbit. With HAL's assistance the task of maintaining the ship was vastly simplified by leaving the details to the computer.
This is my hope for the future of AI in the real world. Forget about sentient robots that are indistinguiable from humans... what's the point? Instead, AI research should focus on assisting humans with their daily activities. One of the only areas in which computers surpass humans is in data management, and this strength should be harnessed to abstract-away many of the trivial details that consume our time. Consider the simple task of buying groceries. A sufficient articifially intelligent agent could handle the job more capably than a human can, by comparing food prices, keeping track of household inventory, observing consumption trends among household members, and then ordering the necessary food.
If it's done right, the homeowner may never have to intervene other than to tweak preferences or make special requests.