Computers and language are about to merge in a way never seen before and the results are likely to be interesting, and I suspect, profound.

One aspect of the near future for computers has been clear for many years and it’s the “Start Trek TNG” vision of computing. In the TNG vision of the future, people don’t have computers, people have microphones that communicate with a computer. Where or what that computer is doesn’t matter as long as it answers back when you speak to it or shows you what you need to see on whatever display is handy.

Apple’s Siri voice assistant is a clear step in this direction and shows us how the “talking computer” is closer than we would have thought possible just a few years ago.

One of the beginning steps was tackling the problem of voice recognition but the early work at Google made those guys realize that figuring out what people are saying is a piece of cake compared to figuring out what they mean. Figuring out the difference between “whirled peas” and “world peace” is one thing; understanding what “Arise, fair sun, and kill the envious moon” means is much harder. You can’t just give a computer a dictionary and expect that you’re done.

This problem of meaning is, of course, Google’s stock in trade and they are deadly serious about being the best at getting it right, because if they are, then they will be the computer you’re talking to and who is answering you back. The computer of the future may very well be Google and the only thing you own will be a microphone, a speaker, a TV and a mobile display.

Hardware, of course, is the least of the changes. This version of computing is also incredibly personal in ways most never imagine. Computers will remember every interaction you’ve ever had with them. The computer of the near future won’t just know what Shakespeare means, it will know which passages you like the best and which you might like if you’re looking for something new. Computers will know where you are, where your friends are and which friend you are most likely to call on a Thursday afternoon. They will know what you like to eat and what that means in terms of recommending a vacation spot.

As for language, Google won’t care if you speak every word of a sentence in a different language or if you make up your own words. As long as you are consistent, they’ll figure it out.

The point here is that until now, language has come from people talking to people. In the not too distant future, the “person” you talk to the most may very well not be a person at all and when one person in a conversation isn’t a person, the nature of conversation will no doubt change and it’s easy to see how that change could be profound… or not.

I think you’ll find that, as with flying cars, the not too distant future will turn out to be the distant future after all.

(Flying cars is a stupid idea anyway ... I hardly trust most people with regular old “driving cars” ... but I digress)

I’m amazed at how far things have come in just a few years. I just got an android phone and the speech recognition is far superior to speech recognition from two or three years ago. Back in the day, you had to train the program by reading their pre-assigned text. Now anyone can speak into the phone and it will hear and type the message. There are problems like “whirled peas”, but I’m just amazed at how far things have come. I have little doubt that the “not too distant future” is really not too distant.

There’s a huge difference between speech recognition software that allows you to input your search terms vocally and actual artificial intelligence that would permit a genuine conversation, which I presume is what we’re talking about (OP is not entirely clear).  The former is impressive but not earthshaking, and is already here.  The latter is definitely far off and very likely impossible (in my opinion).

I usually think in terms of practicality when it comes to this topic.  Yes, speech recognition is pretty good these days.  Yes, being able to carry on a conversation with a computer may not be too far distant.  My question is, do we really want that?  I can’t imagine an entire office full of people interacting verbally with information tools.

I remember several years ago when there was practically a war going on concerning cell phone etiquette, between people like me who were tired of hearing other people’s private conversations, and those who thought they had a right to talk about everyone and everything in public.  Thankfully, texting came along and saved us from a lot of that.

I think interacting verbally with information tools will be possible, but will have limited use in the future.

FWIW, I think we’ll get conversing computers before we get (practical, widespread) flying cars.

I think we’ll get computer “conversations” that emulate human conversations fairly soon, but they’ll be limited in the range of topics they can deal with, say an answering service that can direct you to a particular department. Siri is actually pretty close to this for certain types of questions.

But I think languagehat is right. A general, all-purpose system that can converse on any topic will require a revolution in computing technology far beyond what we’ve got not.

As for flying cars, the technology is practical. The application is not. That’s why we don’t have them.

If you say to your computer (phone), “I’m hungry for a burrito.” and the phone says back, “Anita’s Mexican Food is just three minutes away and I’ve heard the California Burrito is excellent” then you’ve just had a human style conversation with a computer and the fact that the computer doesn’t have “actual artificial intelligence” doesn’t matter one bit.

It doesn’t take human style intelligence to combine GPS data with reviews from Yelp and then present them in a conversational way. These sorts of simple conversations are available today, so I don’t think it’s “flying car” territory to believe that more complex conversations will soon be possible.

I wholeheartedly agree that human style artificial intelligence will probably never happen, but I think we’ve learned from Deep Blue and Watson and Siri that computers don’t need human style intelligence to simulate human performance. Computer intelligence won’t be “real” intelligence, but it will be good enough to get the job done.

"Arise, fair sun, and kill the envious moon”

We probably don’t need computers to analyse R+J just yet but if they can handle simple instructions, that would be good.

A lot of the problem will be not just in interpreting the instruction but in having the abilities to do it. Language is only a little fraction of AI.

I wholeheartedly agree that human style artificial intelligence will probably never happen

In some ways it will be better than human intelligence, which can be somewhat flaky and biased. Your computer for diagnosing cardiac disorders will be a fricken computer for diagnosing cardiac disorders, not a computer for spending some time diagnosing cardiac disorders while worrying about his mortgage and checking out a patient’s rack and being able to make friendly conversation.

Watson is an excellent example of the limitations of these systems. Jeopardy questions follow specific formulas that are unique to the show. They also cover a limited knowledge base—one that emphasizes shallow knowledge about a limited, but diverse, set of subjects. It took months of instruction for Watson to be trained in how to answer those formulas—a task that human can master in minutes. Were Watson and a human pitted against one another in a free-form trivia contest, or one that required expert knowledge of a subject, I have no doubt the human would mop the floor with the computer. (The other advantage the computer had was in timing the buzzer. Watson’s biggest advantage had nothing to do with AI.)

Yes, a computer that’s trained at diagnosing cardiac disorders would probably do a better job at diagnosing cardiac disorders than a human. But what about those cases where the problem isn’t in the heart but has similar symptoms, or where the heart problem is complicated by another, or multiple, unrelated problem(s)? Or where the individual has natural variations that take them outside the parameters the computer is equipped to handle? People are not made from cookie-cutters; one of the things that keeps clinical medicine from being a true science is the individual variation. Good medicine is a science-based art. Human physicians will do a better job for a long time to come at the really challenging diagnosis and treatment problems. Computers may, and probably should, be used to supplement the human physician, but I don’t see them replacing them any time in our lifetimes.

[“not" inserted—dw]

Yes, a computer that’s trained at diagnosing cardiac disorders would probably do a better job at diagnosing cardiac disorders than a human. But what about those cases where the problem isn’t in the heart but has similar symptoms, or where the heart problem is complicated by another, or multiple, unrelated problem(s)? Or where the individual has natural variations that take them outside the parameters the computer is equipped to handle? People are made from cookie-cutters; one of the things that keeps clinical medicine from being a true science is the individual variation. Good medicine is a science-based art. Human physicians will do a better job for a long time to come at the really challenging diagnosis and treatment problems. Computers may, and probably should, be used to supplement the human physician, but I don’t see them replacing them any time in our lifetimes.

I concur.

People are made from cookie-cutters;

Should there be a “not” in there?

I hope the computer will continue to be our plastic pal whose fun to be with.

OP Tipping - 10 September 2012 07:27 PM

Should there be a “not” in there?

I hope the computer will continue to be our plastic pal whose fun to be with.

Muphry’s Law strikes again.