September 8, 2007
Singularity Summit 2007
The “Singularity” is the predicted event when artificial intelligence expands beyond its programmed 0’s and 1’s and starts to think for itself. As this undoubtedly means a huge change for us measly humans (whether very good, very bad, or somewhere in the middle), an organization called the “The Singularity Institute for Artificial Intelligence” holds a yearly conference to address the progress, roadblocks, and concerns with bringing such life into the world.
What will it mean to have a computer that is smarter than a human? Will society give it rights? Will it need emotions to understand the questions we ask of it? Will groups be founded to “free” the enslaved robots? These questions and many more have been asked, skirted around, addressed, and discussed by many of the foremost researchers, authors, professors, and futurists today at the Palace of Fine Arts Theatre in San Francisco.
Ray Kurzweil, while only attending via video-conference tomorrow, is present in spirit with every speaker as a leading futurist and inventor of many cutting-edge technologies in use today. In his recent book “The Singularity Is Near: When Humans Transcend Biology” he predicts the singularity will occur sometime in the next 10 to 40 years. This is rather optimistic in my opinion but I’d sure love to see the results of such a change within my lifetime.
The conference is focused on artificial “general” intelligence (or AGI), rather than a more task-focused, “specific” AI, such as IBM’s Deep Blue. AGI’s will essentially be frameworks which, once created, require no more programming (in the traditional sense) to learn and apply any other type of intelligence. An example used by several speakers is a high-schooler who can be trained to do many low-skilled jobs. These types of jobs typically involve repetition and an ability to follow simple rules but they are still well outside the grasp of even the most powerful supercomputer running our most advanced algorithms.
I mention algorithms because brute force alone will not solve this problem! Simply having a computer with the processing power of every human brain on the planet will not suddenly make it self-aware. However, just because we don’t have this power, doesn’t mean we can’t get started today…
In the factoring of prime numbers, would you rather run a 1977 algorithm on 2007 hardware or a 2007 algorithm on 1977 hardware? Many people assume the former simply because computers have gotten so much faster. The fastest solution is the 2007 algorithm running on 1977 hardware. It would take about 12 years to factor a 90-digit number using the 1977 algorithm on Blue Gene and only about 3 years to factor a 90-digit number using the 2007 algorithm on an Apple II. Developing the algorithms that drive artificial intelligence today will produce some major results on a smaller scale as we build up computing power.
Today’s talks focused on what advanced AI is, what is being researched and what progress has been made. A fun equation created by Dr. Rodney Brooks calculated that by 2019, we would be able to fit every movie even the slightest bit interesting on an iPod. In only two years, we’ll be able to fit the entire Library of Congress. Barney Pell, while his talk was more of an overview of the future of AI, had some interesting things to say about natural language processing, or NLP, and his company Powerset (which aspires to collect knowledge from the web, rather than data). NLP is fascinating to me because there’s just so much data on the web and so few ways to process it. Google and Ask Jeeves are on the right track with answers to simple questions, and even simple dialogs with phrases like “did you mean…” but there are many more places that search tools could go to become more useful to everyone. The panel discussions were quite interesting, with questions ranging from “What world (real or imagined) would you choose to live on?” (Paul Saffo answered “this one” after relating a story about driving across the golden gate bridge the night before) to “Why are people working to prolong life?” (which drew some puzzled looks and an answer from Peter Voss, “some of us have tried this life thing for a while now, and we kinda like it”) and plenty of on-topic ones as well.
I suppose the theme of the conference is that there are no “correct” answers yet, so feel free to give the wrong ones because no one will be able to prove you wrong for years. At this point, keeping open the lines of communication between everyone in the field will help anticipate any problems and will help everyone avoid them.
Tomorrow, the talks will cover the ethical issues surrounding AI as well as the potential risks of smart computers (queue almost every recent Hollywood movie about AI). Kurzweil will also be making his video-appearance so that should be exciting.
