Stan Franklin

Nearly 100 people who will change your life forever gathered in Memphis recently to discuss the future of artificial intelligence, and to revive the goal of building machines capable of abstract thought.

The FedEx Institute of Technology and the University of Memphis hosted the first Conference on Artificial General Intelligence March 1-3.

Artificial General Intelligence (AGI) is a concept that dates to 1955, but only now is there technology that makes it feasible. People encounter artificial intelligence every day and don't realize it. Artificial intelligence is the reason Google searches seem intuitive; Google remembers your previous searches and learns how you think.

Also, many online hosts of forums and chat rooms are also robots that learn. Medical diagnostic systems gain accuracy with experience. The Air Force uses artificial intelligence in robot drones such as the Predator to analyze battlefield targets.

If your 401(k) is doing well, it's because your brokerage uses AI to study markets, patterns and trends to make stock picks.

But those are all very specific and narrow applications of artificial intelligence, and a long way from AGI.

"The first killer AI application was expert systems, which are widely used today to capture human expertise in a software system for diagnosis, maintenance and other very practical, but narrow, tasks," said Conference Chair Stan Franklin. "Artificial intelligence got away from its initial goal of AGI primarily because AGI rapidly proved too hard a problem to solve. AI researchers concentrated on narrow goals, building smart machines in narrow domains. Narrow AI has played, and continues to play, an important role in our lives and our businesses."

Of machines and men

Franklin is one reason the University of Memphis is considered a global leader in artificial intelligence. He is the W. Harry Feinstone  interdisciplinary research professor in the Department of Computer Science at the U of M and co-director of the university's Institute for Intelligent Systems.

One of Franklin's accomplishments is Ida, a personnel system used by the 800,000 sailors and Marines scattered around the globe. By phone or e-mail, Ida coordinates transfers from one unit to another, offering job alternatives and suggesting additional training for career advancement.

Ida is a "cognitive agent" that thinks and learns; with each successful job transfer, Ida gets better at something that previously was done by humans shuffling paper.

AGI is a realistic goal today, Franklin said, with a convergence of computer science, cognitive science and neuroscience.

Most people can grasp the utility of AI in specific applications, he said. AGI is, however, closer to pure research. Just as a microbiologist might unlock the secrets of a living cell and leave others to do something useful with the discovery, AGI is a goal
unto itself that in a few decades will take root in thousands of ways nobody has dreamed of.

"No one would have predicted that the microelectronics industry would grow out of space flight," Franklin said. "My interest in AGI is from the science, rather than the engineering, side. AGI will help me understand how minds work, perhaps the single most interesting problem there is."

Search for funding

AGI is analogous to the Manhattan Project, which built the first atomic bomb during World War II, said Ben Goertzel, CEO of San Francisco software developer Novamente LLC. He was co-chair of the conference and one of the top draws for the event.

"Think of it like the Gemini and Apollo missions, or the human genome project," Goertzel said. "It's a huge undertaking. Unfortunately government funding is not oriented toward long-term research."

It's been in the narrow applications that companies like Novamente are able to generate revenue to pursue AGI. Goertzel developed a proprietary cognition engine and has built applications for defense contractors, the National Institutes of Health and a host of others.

Right now the company is working on virtual pets for computer games and cell phones that learn from their masters. He started with synthetic dogs - because they don't talk - but hopes by next year to integrate speech and create an animated parrot.

"The novelty is that these virtual dogs can learn new behaviors; they're not limited to a set of group behaviors," he said. "They integrate what you've taught them into their behavior."

Virtual pets only exist for entertainment, but developing them contributes to research while being commercially viable.

Toward collaboration

The conference is designed to generate enthusiasm for AGI in the Ph.D. scientists of university and corporate labs who spend their days now working mostly on applications. Franklin would like to see an AGI professional journal emerge from the conference, perhaps a book series and an online community where experts can share ideas in real time.

The event also draws serious attention to the U of M and the Institute for Intelligent Systems. There's no specific recruiting going on right now, Franklin said, but impressing Memphis onto the brightest minds in the world only can be beneficial for the future.

Along with Franklin, some of the other key people in AI include:

Cognitive psychologist Art Graesser, who created AutoTutor to train Navy aircraft mechanics. An interactive avatar teaches the subject, learning and adapting to each student. AutoTutor is also being adapted to elementary education by researchers at the U of M College of Education.

Computer scientist Max Garzon, whose work focuses on advanced computing methods.

Dipanker Dasgupta, a professor of computer science and director of the Center for Information Assurance. Dasgupta specializes in making information systems more robust, including self-diagnostic, distributed repair systems modeled after the human immune system.

"It feels to me like the time is right for the AI research community to reorient itself back to the original dreams," Goertzel said. "My end goal is to create machines that think like you and me and, ultimately, better."