-
A lot of people ask the question “can computers ever be as smart as humans?” The answer, is “it depends on what you mean by ‘smart’”. Computers can do some things faster and more accurately than humans; they can accomplish mind-numbingly boring tasks faster than a human can pick up a pencil. But the human brain is always going to be much more adaptable than a computer’s.
Is a computer that can translate words from English to Japanese intelligent? Or is it just following a series of complex rules, no different from a fancy calculator that knows how to add or multiply? Is the ability to follow a series of complex rules a sufficient condition for intelligence?
Alan Turing, an early computer programming pioneer, proposed a test to determine machine intelligence. Briefly, if a person is interacting with a computer, but is convinced he is actually interacting with another person, that computer can be considered intelligent. (Turing was a brilliant man who helped win World War II for the allies, and his life and accomplishments helped build the foundation of computer science as we know it.)
Filling In The Blanks
We’ve all seen the poster that says “F Y CN RD THS Y R SMRT” It’s the sentence “If you can read this, you are smart” without any vowels. Kids who learn to read are usually impressed with themselves for being able to understand that sentence. There is a part of our brains that is able to bridge gaps in information that it receives. Computers, however, have a rigid processing system, and need to be told rule-by-rule what to do. Would a computer be able to "read" this sentence and make sense of it?
You could program a computer to try out different vowels at each position in the words above, and see if those words match a word in the dictionary. “Y” is an interesting case – it is the only consonant in the word. You could program the computer to know what “Y” means (possible values would be eye, you, yea, aye, etc.) A better way to do it would be to program the computer to look in the dictionary for every word whose only consonant is a “Y”.
But how would the computer know which of the words to choose? It could apply some rules of grammar, like Microsoft Word’s grammar check does, and attempt to piece together a sentence whose words make the most sense together. I hope it wouldn’t come up with this:
FEE EYE CONE ROAD THESE EYE ORE SMIRTA!
I made that last word up. But there are a lot of possibilities.
(A funny note: Before I posted this, I ran the spell checker. When the spellchecker got to the quote “F Y CN RD THS Y R SMRT”, it flagged 'CN', 'THS', and 'SMRT' as misspelled. The rest of the "words" in that quote are apparently spelled correctly.)
Relationships and Assumptions
It is difficult to teach a computer how to derive meaning from a sentence, especially one with shortcuts like pronouns.
“The teacher sent him to the principal’s office because he was misbehaving.”
Who is “he”? “Him?” The teacher or the principal, or another, unnamed person? How do you know? How would you help a computer know the answers to those questions?
“The teacher sent him to the principal’s office because she wanted to see him.”
Is the principal male or female? How about the teacher? Who is “him”?
“The teacher sent her to the principal’s office because she wanted to see her.”
Is the principal male or female? How about the teacher? Who is being sent to the principal’s office? The mother of a student? A female student teacher? The classroom's pet snake? Who wanted to see whom? Context is everything in these sentences, and in some cases we don’t have enough information to answer these questions. It would be important, then, to teach a computer to keep track of the preceding and following sentences to get information to help put together the full meaning, as well as teach it the possible relationships between different people at a school. But doing this would be very difficult.
Context, Implied Information, And Jumping To Conclusions
Context often goes beyond resolution of pronouns or relationships between actors in a sentence, and goes to meaning, assumptions, language variations like sarcasm, and so on.
Person 1: “I’m hungry.”
Person 2: “The sandwich shop is open until 9:00.”
Person 1: “I’ll get the keys.”
Before you read further, picture the above conversation in your mind - who is speaking, are the two people male or female, where are they when they have this conversation? What assumptions are you making about the people who are speaking?
An astounding amount of information passed between these two people, including several implicit agreements and statements. Person 2 seems to be hungry like Person 1 – or does he? They’re going to drive to the sandwich shop, right? We assume it is not yet 9:00, and that they will have enough time to get there before 9:00. We assume Person 1 is driving the vehicle - is that a fair assumption? We assume they’re both going – or maybe not? Maybe Person 2 is not going because he just had Wendy's, and is stuffed to the gills, and Person 1 will have to buy sandwiches on his own. Perhaps Person 1 is a 12-year old boy studying for a test, and Person 2 is his parent, offering (implicitly) to fetch a sandwich for his hard-working son.
Perhaps there is a running joke between the two that they will never visit that sandwich shop again, and Person 2’s response is actually a joke. We would have to know the two people involved and their history before we could say with certainty exactly what the meaning of the above conversation was and what agreements, if any, were made.
In this scenario, how would you introduce the facts of the matter to the computer? Does a computer know that 12 year-olds don’t usually drive a car? Does the computer know what it means to be busy studying and have someone else make a sandwich run on your behalf? Does a computer understand sarcasm? Does a computer know that “I’ll get the keys” implies a car or other motorized vehicle is involved?
It's easy to jump to conclusions based on our own histories. For example, did you assume the keys in question were keys for a car? What if the people are on an island and the keys are for a boat? What if they’re stranded on an island, and the second and third lines of that discussion are a joke between the two speakers? We don’t know for sure unless we know a lot more about the speakers and their situations.
Unspoken Communication
It's common in sports such as football and basketball that players on the same team will use hand signals or other non-verbal cues to communicate. A quarterback who reads the defense and is sure the receiver to his left is going to have one-on-one coverage can give a slight nod to the receiver, and that is all that's needed to let the receiver know what the quarterback is thinking. A quick glance between two basketball players can be an implicit agreement between them that the one with the ball is about to pass it to the one without the ball.
And of course we all use non-verbal communication in our daily lives, like the puzzled or exhausted looks we give our kids when they make the error of acting their age. This non-verbal communication that regularly happens in sports is the kind of communication that humans use all the time, but (and I’m making a strong statement here) computers will never have.
There are computer systems that do can things like use a camera to follow the eyes of its user, moving the mouse pointer or performing some other action when their eyes move; in fact, the graduate student who worked with one of my professors at MSU developed a system that did just that. But in this case, the computer was following a set of rules about what to do when an image captured by a camera changed, not what the user was thinking.
I'm sure there are some exceptions to these statements, and the industry is always inching closer to more intelligent machines that can respond to the user based on that user's history, but as far as true human communication, it will be a long haul before computers can keep up with humans. And once it does, there will always be the next challenge waiting: opposable thumbs.
No comments:
Post a Comment