Despite fears stoked by a Jeopardy!-conquering computer program named Watson that we're on the brink of a Terminator-esque awakening of artificial intelligence hell-bent on killing all humans, the science behind the program is driving practical innovations in everything from the checkout line to search engines.

A fleet of IBM engineers based out of Westchester County, New York, developed Watson over four years to take on two Jeopardy! champions, Ken Jennings and Brad Rutter, over three matches.

To do so, programmers had to install four terabytes of data — more than 10 million documents, such as encyclopedias, Wikipedia, dictionaries and the Bible — that Watson could instantly pore over once an answer was given in a near-lightspeed game of word association.

(If you've never watched the long-running TV show, Jeopardy! provides answers to which contestants must give a correct question).

But that reservoir of knowledge would be useless if Watson wasn't able to determine just what host Alex Trebek was looking for with each clue. Because Jeopardy! clues often incorporate puns, wordplay and idioms, Watson would have to be able to essentially read between the lines of clues to arrive at an answer.

The IBM team jumped that hurdle by downloading 10,000 old Jeopardy! answers, as well as their correct questions, onto Watson's memory. From that sample set, Watson was able to divine patterns from the clues that he would then recognize in new answers — meaning the programmers found a way for Watson to use old knowledge to arrive at new questions.

Or, how to learn.

This process of "machine learning" is one of a number of fields of artificial intelligence research, but it's one that's increasingly offering solutions to real-world issues.

At its most basic, machine learning is the principle behind such web applications as Amazon.com or Netflix, which analyze users' data to recommend movies or books they might like to watch or read.

According to Steve Baker, a science writer who followed Watson's development for his just-released book, Final Jeopardy, IBM is already working with Columbia University's medical school to put Watson's capacity to learn to work as a digital doctor.

"If you give it enough symptoms, a machine like this could go through hundreds of thousands of research papers, much more than humans could ever digest, and come up with hypotheses on what ails a certain patient, and which treatments might prove effective," said Baker.

An iteration of Watson could make its way to call centres, said Baker, replacing the workers needed to fulfil mundane tasks with computers who don't need coffee breaks or benefits.

Baker also speculated that large corporations could use Watson-like software to monitor communications to make sure everyone in the company is keeping in line with what regulators say they can and can't disclose.

Closer to home, a University of Toronto computer science professor is using similar methods to help computers "see" and predict what humans might be thinking.

Geoff Hinton, who last week was awarded a $1-million research grant for winning one of Canada's top science prizes, the Herzberg Gold Medal, has devoted decades of his life to artificial intelligence.

"People are really surrounded by artificial intelligence," said Hinton. "It's just that once it works, they stop calling it artificial intelligence. So for example, when Google returns just what you're looking for the first time, in the old days, they would've called predicting that artificial intelligence. Nowadays, they just call it run-of-the-mill Google."

With grad student Ilya Sutskever, the first Canadian to be named a Google scholar by the tech giant, Hinton is developing a program that takes the concept of predictive texting found in cellphones to new heights.

The two downloaded all of Wikipedia onto the program, which analyzes millions of words in the online encyclopedia to arrive at a set of possible continuations for a sequence of characters entered into it. The pair started with just individual characters, so that if you entered a 'D,' the program would provide a short list of the most likely letters to come after it, from which the user would select the right one, thus "teaching" the program how to identify patterns in words.

Now, they've developed it to the point where it can respond to a string of words.

"We tried it, for example, by saying 'the meaning of life is . . .' and one time it said 'the meaning of life is . . . literary recognition', and that (entry) is not in Wikipedia," said Hinton. "So it had to understand things like that 'recognition,' for example, is a sensible word to come after 'literary,' and that it's probably a good thing.

"Another time, it said 'the meaning of life is . . . ancient human reproduction.' "

Hinton said the program has already learned to identify the difference between singular and plural words.

"It's figuring out rules of English grammar on its own, just like kids do. Kids can't tell you what the rules are, but they know the rules quite well."

The applications of such developments start with improving spellcheckers and Google's predictive-text search engine. Recognizing text is one thing; recognizing images, movement or sound — senses humans take for granted — poses a challenge that researchers like Hinton are steadily surmounting.

This same process that turned Watson from a repository of pub trivia into a responsive, thinking contestant is being used by Japanese high-tech manufacturer Toshiba to develop scanners for grocery stores that can determine whether a customer is buying, for example, Macintosh, Empire or Red Delicious apples by cross-referencing the object in front of them with a database of images and information stored in their memory.

Hinton pointed to a trend in automaking of including optical sensors to scan for signs of trouble — animal movement beside a road, oncoming traffic — that a driver might miss.

"People have lapses of attention, and you could enhance people's abilities to drive by having computers see what's going on, much as you can make aircraft land more safely by having a landing system that can enhance what the pilot can do," said Hinton.

"It can make up for human error."

Baker said these advances will likely accelerate as the cost of the technology behind it goes down. IBM hasn't disclosed how much it cost to develop Watson, but Baker estimated that to employ 30 people full-time for four years, the company would have spent about $30 million — the cost of 10 Superbowl commercials, and pocket change to IBM's $6-billion research-and-development budget.

"What is going to happen is that kind of intelligence is going to be available to all of us for much cheaper prices," said Baker. "Bits are going to be embedded in the devices we use. Now we usually have to type for machines to understand us. We're going to be in a world where we can talk to machines, using our normal language, and they're going to understand us much better than they ever have.

"Part of it is going to be a sort of democratization of (Watson). We're all going to have access to Watson and its parts whether we like it or not."

That fear and fascination with the unknown side of technology has been a Hollywood staple since at least the computer HAL from 2001: A Space Odyssey decided it wouldn't let its human operators do what wanted they do. Artificial intelligence has almost always been portrayed as a villain in films, from Blade Runner to The Terminator to the Matrix films.

And while Watson bested two of the world's better brains — the affable Jennings quoted the Simpsons in a final answer, joking that, "I, for one, welcome our new computer overlords" — Hinton provided some balance to half-serious claims of artificial intelligence surpassing our own.

"You'll notice Watson doesn't have a real body, and it doesn't know how to move about in the real world. But curiously it can do things we regard as involving a lot of intelligence, but actually, they're intrinsically simpler things, it's just our brains haven't evolved to do them very well.

"The things our brains have evolved to do well, like controlling our bodies, recognizing objects, recognizing sounds, we're still much, much better than computers."

mbarber@postmedia.com

Twitter.com/mbarber86