Mar 18, 2008 12:33 PM
8,896
Enter your username and password.
San Francisco, 5:41 AM
Wed Feb 10
25 posts in the last 24 hours
One of the most popular futurist hobbyhorses is the idea that artificially intelligent machines will soon become ubiquitous and change the world forever. This is an old dream, which may have started with Isaac Asimov's idea that superintelligent computers would take over the geo-political management of Earth (see the final story in I, Robot) and create a more rational world. Early computer geeks like Alan Turing imagined that AI would simply be a perfected human brain, sentient but far more powerful and capable of solving problems humans can't. Most scientists and futurists agree that true AI has the potential to create a better world, but what can you put on your to-do list today that will help make AI a reality in fifty years? Actually, there's quite a lot.
To-Do List for Futurists: Creating A.I.
1. Today: Tag everything you can on the Web. Many A.I. theorists believe that the first steps to creating a sentient computer involve teaching it to recognize information in the same way humans recognize it. So, for example, if you tag images on photo-sharing site Flickr, you are building up a database for a future A.I. who can look at a picture of a car and say to itself, "90 percent of people called this a car, so it's most likely a car."
2. Today: Along the lines of the "tag everthing" task, you can also teach future A.I.s how to evaluate what they're seeing in a subjective way too. For instance, you can start generating data that will teach A.I.s to recognize the difference between science fiction and science by using services like StumbleUpon or Del.icio.us, where you have a chance to categorize and rate any Web page. Find an excerpt from a novel about computers by Neal Stephenson? Categorize it as "science fiction." Find a book about computers by journalist Steven Levy? Categorize it as "science." The richer our metadata is, the closer we come to creating machines that can evaluate images, text, and objects in a human-like way — simply because the machine will have so much data about how humans have already evaluated them.
3. This month: Tutor a kid in math or computer science. You may not be the next big genius who is going to invent the nice A.I. who does an anti-Skynet and stops all war through rationality. But the kid who lives in your neighborhood who doesn't have the cash to buy her own laptop? She might be. So help out by tutoring — you can often find opportunities via services like VolunteerMatch.
4. This month: Help make statistical machine translation of human languages as natural as possible. A few hours' worth of work with MIT's open source MOSES software project, and you can help A.I.s of the future gain a nuanced understanding of how to do idiomatic translations from one language to another. This will, of course, also help A.I.s to gain a feel for speaking in natural languages themselves. Basically, you upload "training data" to MOSES — usually two texts, one an original and one a translation — and then you give MOSES feedback on whether the translated phrases it has now learned work in all situations.
5. This year: Many experts now believe that A.I.s will only evolve if we can place them inside robotic bodies, because sentience is so bound up with being able to move around in the world. (So say goodbye to the idea of an A.I. that just sits in a giant box.) Get educated about robotic intelligence by visiting a robot show (Robogames is a good one, and you can look for others like it in your local area). If you can't make it out to a robot show, try reading up on the future of robotics in a great book by MIT AI lab researcher Rodney Brooks called Flesh and Machines. It was written a couple of years ago, but it's still up-to-date in terms of what the most cutting-edge research is.
Yesterday's to-do list: How to Build an Ecotopian Society
Tomorrow's to-do list: How to Colonize the Moon
Send an email to Annalee Newitz, the author of this post, at annalee@io9.com.
Original material is licensed under a Creative Commons License permitting non-commercial sharing with attribution.
Start a new discussion