|
TIME MAGAZINE, JUNE 4, 2001, VOL.157 NO.22
I'm Afraid I Can't Do That
Getting computers to think out-of-the-box is a tough job, but the payoffs will be enormous
By MARVIN MINSKY
Why don't computers have common sense? Why can't they learn from experience? Our programs can soundly beat people at chess yet none can do most of the things young children are capable of. Many assume this is because machines lack some special quality that only a human brain possesses. But the problem is that most present-day programs work only one way. If something goes wrong, they get totally stuck.
Our programmers still use an outdated strategy: find the best way to solve a problem and write a program accordingly. Although this might seem sensible, each program can only solve some particular kind of problem. That means a total loss of resourcefulness. Here's what I think our brains do instead: consider several different methods, each of which might have a good chance to succeed. When one method fails, the brain quickly switches to another.
In physics, researchers have made great progress by searching for logical, unified theories. But in real life, things are never that simple. Practical thinking will always be based on messy collections of incomplete theories, which means each person must learn to switch between different styles and methods of thinking. Since each of our commonsense rules of thumb is likely to have many exceptions, we learn to depend on analogies instead of logical certainties.
A second fault is that we have misconceived how we ought to aim our programs: instead of telling computers what to do, we should tell them about our intentions and goals. If they could understand what we mean—and if they were more able to think for themselves—we would not always have to reprogram them for every slightly different task. To make our computers more useful to us, they will need to understand what we mean when we interact with them. And the secret of knowing what something means lies in how we interconnect our ideas. If you understand something in several ways, you can turn ideas around in your mind until you find one that works for you. That's what we mean by thinking.
So instead of blaming computers for their faults, we must make our machines more resourceful. First, we must give them more common sense by supplying them with more knowledge. Then we must keep them from getting stuck by giving them more ways to think: some ways that are like what we call our emotional states.
If you were to mention a string, a typical child would know these things about it: a string can be used to pull an object but not to push it; it isn't good to eat a string; you cannot make a balloon with string; before you put a string in a box, you must first open that box; if you steal someone's string, its owner will be annoyed.
How large are our networks of commonsense knowledge? Most people know thousands of words, and each word has some links to other words, as well as to other, nonverbal structures. Each of those in turn leads to multiple links. But we still know almost nothing about how our millions of fragments of knowledge and skills are organized inside our minds. Making good theories of how that might work will be, I think, our principal challenge in solving the problem of making machines that use common sense.
When we imagine, predict and decide, we use masses of factual knowledge, but we also use knowledge about how to think for supervising those processes. We need ways to describe what we're trying to do and for thinking about those descriptions. (This must be an important part of what people attribute to consciousness.)
To make computers more pleasant to use, some researchers have tried to make them behave in more natural or lifelike ways. But each friendly interface has ended up making the same mistakes because computers simply don't know enough. Once our computers have more common sense, then everything else in our world will change. When our future machines are as brainy as we are, they could replace all our present work (except for the jobs we wish to retain). Mass production could disappear: every product could be tailored to fit. Everything in our world would change—our work, our leisure and all our ambitions.
Marvin Minsky is Toshiba Professor at M.I.T. His upcoming book, The Emotion Machine, probes computer intelligence
Related Sites
Marvin Minsky
IBM's Almaden Research Center
Note: Pages will open in a new browser window
External sites are not endorsed by Time Inc.
More The Future Stories
The Write Stuff
Still hanging on to your Bic? This new technology will let you use pen and ink and be digital, too
What's Next?
Draft: Proposed United Nations Treaty on Human to Smart Object Interrelations
I'm Afraid I Can't Do That
Getting computers to think out-of-the-box is a tough job, but the payoffs will be enormous
A Day in My Life, 2025
My wife's a hologram, my boss is buggy and my virtual analyst doesn't understand me
|
advertisement
|
|
Libraries
Full Contents: all of the stories in one simple list
Multimedia:  the home of our video, audio and interactive features
Video: CNN circles the globe for how technology is changing our lives
Toolbox: software you may need for this site
|
|
|
Subscribe to TIME
Magazine
Stories from this week's issue
Ethics
Big Brother is watching the Net. Do you care?
Living
Talk to your thermostat, surf from the toilet, phone your fridge
Entertainment
Music mixing as easy as logging on to a website and typing on a keyboard
|
|
Specials
CNN's hour-long special program on Our Interactive World, hosted by Michael Holmes and Tumi Makgabo, featuring luminaries from the world of information technology
Brian Bennett, reporter for TIME magazine, interviews MTV Asia's LiLi, a virtual veejay
Lili on her life and work: chat transcript from May 31, 2001
|
|
|
U.S. | Asia | Europe | Current Issue
U.S. | Asia | Europe | Sci-Tech
[an error occurred while processing this directive] | You are