Robin Burgener wants to turn child's play into rocket science.
When he speaks to a room full of NASA scientists, programmers and technicians later this month, he'll explain how a simple parlour game he first adapted into a computer program 20 years ago might just be the answer to some of the agency's most pressing issues.
The game is 20 Questions, once a social icebreaker in which participants had that many queries to determine what their opponent was thinking about, spawning the classic opening line: "Is it animal, vegetable or mineral?"
The computer version created by Mr. Burgener, "chief thinkologist" of 20Q.net Inc., the Ottawa company he runs with his wife, Tanis Stoliar, is similar, but the questions are posed by software driven by an algorithm and an ever growing database -- what Mr. Burgener describes as "common knowledge" gleaned from players themselves.
The program, which recently morphed into a best-selling toy, gets it right 80 per cent of the time at 20 questions, and 98 per cent of the time with 25 questions.
Leveraging that neural network for NASA isn't pie in the sky, as he'll explain at the Goddard Space Flight Centre in Maryland when he discusses artificial intelligence (AI), spacecraft and customer service.
"What's intriguing to NASA is the ability of the neural network and the algorithm to formulate an answer even if the data isn't accurate," he says. "So if a sensor fails, you're able to see past it."
More importantly, based on the data the AI is fed, it might then predict problems with systems before they happen -- an invaluable tool in space.
Like any system, it must first be loaded with answers and configured. The algorithm weights the questions and answers according to probability based on past experience and the database, which is ever expanding.
"You can play the same game and it will ask different questions in a different order," he says, stressing the neural network is not simply a series of yes-no gates that drive inevitably to a conclusion.
The game began on a 5.25-inch floppy disk. Over the years Mr. Burgener tinkered with it as a distraction from his real job, first as a game designer and later at such high tech companies as Corel Corp.
By about 1998 he'd created a program that not only anticipated questions but learned from the answers it was given, even if the player was lying or obfuscating. It then morphed into a popular website -- 20Q.net -- and more recently into a source of income with 15 million units of a handheld version being sold around the world at about $10 each. It was lauded as toy of the year in 2004 by the Toy Industry Association, and a downloadable mobile phone version is now available.
What's fascinating is that the more people who play the on-line version -- which has been played 35 million times so far and acts as an incubator for the handheld version -- the more data it captures and the more it learns.
As such, that database of common knowledge isn't infallible.
"For a long time it thought a dolphin was a fish," he says, "because many people make that mistake and that's what it learned."
It is also available in 14 languages, with more to be added soon. It also comes in American, Canadian and British versions, which has proved a learning experience for the AI as it determines cultural differences in languages and meanings. It is also adaptable, scalable, modular and embeddable, able to be configured to any platform, from handheld to supercomputer.
"The on-line version is a neural network with 10 million synaptic connections," Mr. Burgener says. "The handheld version has about 250,000 -- about as many as a simple insect like a fly."
Game fans have plenty of trivia contests to choose from, but Mr. Burgener is aiming for a new horizon, one that takes him to places like NASA more often. It would be a welcome change from the toy industry.
"The toy business is the most cutthroat, low-margin business around," he says. "Everyone's looking for the next big thing."
Already he's looking at adapting the system to identify hazardous materials in the field.
"It might not be able to immediately identify exactly what it is, but in 20 questions it should be able to say whether it is dangerous or not and what actions should be taken," Mr. Burgener says.
Similarly, technical-support call centres could use the system to pre-screen and solve callers' problems more efficiently.
"Observing call centres, you notice that the technicians get bored asking the same questions over and over again," Mr. Burgener says, adding his software could eliminate the majority of inquiries and direct more unique challenges to human technicians.
And though one of the most obvious applications would be in the health sector, Mr. Burgener is reticent. "The thing about machines is that they don't sue, people do," he says. "While it would be great to configure a neural network for medicine, there's a liability issue. There would have to be warning labels all over the screen."
Still, he says, an artificial intelligence in the hands of a lay person might be invaluable after a massive disaster such as a tsunami, when qualified medical personnel are swamped and in short supply. Those in remote communities could triage cases more effectively, too, he says.
How the game works
20Q is best described as a "simple algorithm that works on a massive scale," creator Robin Burgener says. Here's how it works, in simple terms, when a player thinks of an object, such as a truck:
1. The question "animal, vegetable or mineral" is designed to eliminate possibilities and narrow the field. But even if the answer is "other," Mr. Burgener says, that's okay, because a non-answer is still an answer to the algorithm, and "all answers at that stage are equally valued."
2. At each question the list of possible responses -- yes, no, unknown, irrelevant, sometimes, maybe, probably, doubtful, usually, depends, rarely and partly -- shifts the weight of probability to a specific group of answers within the database.
3. While the correct response for truck would be mineral, many players would say "other" or "don't know." Like chess software calculating outcomes, the neural net reviews its database for outcomes from other games and asks the next question: "Is it smaller than a loaf of bread?"
4. The answer is no. Now it eliminates all things within its database that are smaller than a loaf of bread. It suggests an object of some kind, and so on, until question five: "Is it mechanical?" The affirmative answer causes the system to review other game outcomes in which a mechanical object has been the solution.
5. The system then offers several solutions after more questions until it has enough information to provide the correct answer. It also can challenge the player's responses. In the case of the truck, for example, a negative response to the question "does it bring joy to people" should have been affirmative, the program says.