Don't Just Talk to Intelligent Agents -- Be One

Back to Contents of Issue: October 2001

AI entrepreneur Hideto Tomabechi works to integrate humans and machines in a "hyperself" architecture.

by Chiaki Kitada

IT COMES AS NO surprise to Hideto Tomabechi's boyhood friends that the one-time teenage prize winner of a Harajuku computer game competition has grown up to be an AI programmer and head of a cutting-edge software firm. But few expected that he'd also turn into a pony-tailed, monk habit-wearing religious thinker who'd gain media attention as a deprogrammer of Aum Shinrikyo cultists.

In 1995, Aum gained lasting infamy for launching a nerve gas attack in the Tokyo subway system that killed 12 and injured 3,800. Tomabechi was called in to help deal with the cultists, and last year published a book (Brain Washing Theory, Shunjusha, 2000) to report on his experiences with "exit counseling." His engagement and marriage to a former Aum cult follower became a hot topic among media watchers at the time, as did his release to a TV station of the videotape of a counseling session with a former cult follower suspected in the attempted murder of the National Police Agency chief.

Earlier, Tomabechi had left Japan to study artificial intelligence under AI guru Roger Schank at Yale University. He gained his PhD in computational linguistics from Carnegie Mellon University in 1993. Back home, after four years of teaching and research into functional brain science at Tokushima University, he headed up the fundamental research lab at Tokushima-based Just System, one of Japan's leading software developers. It was during this time that he was called in to help deal with Aum. He was keen to assist, particularly after seeing the familiar face of a former college debate team member he had taught back in the early 1980s appearing on TV, representing the cult.

After leaving Just System in 1998, Tomabechi revived a company called Cognitive Research Labs (CRL) that he had founded during his Carnegie Mellon days, to work on government-sponsored projects, including next-gen advertising delivery software in 2000, dynamic home servers, and an integrated mobile online supermarket system in 1999. Today, CRL has grown to 50 staff members and produces software that's conceptually based on an AI theory called "hyperself," which Tomabechi has come to espouse after more than 15 years of research work as an AI scholar and functional brain scientist.

The intense, eclectic Tomabechi met with J@pan Inc's Chiaki Kitada and Daniel Scuka at CRL's Roppongi offices, where he spoke at length on artificial intelligence, the hyperself architecture, Buddhism, and running a startup.

How do you find working in Japan after spending many years in the States as a researcher?

Everything is harder here.

First, US culture is laissez-faire. They are, in a sense, open. Also, fairness doesn't exist in Japan, at least in my industry. I'm lucky, because I have been getting government funding. I've [received] almost $5 million in funding from the former MITI [Ministry of International Trade and Industry] alone in the past three years. I was a member of a committee advising MITI when I first got initial funding -- about a half million dollars. Usually, it's impossible for tiny companies to receive such funding because the money is [allocated] as a government version of venture capital to promote the information or high-tech industries. Because it's tax money, the government is reluctant to take risks [by] giving money to companies too tiny to have a brand name or history. So it [goes] only to major companies in reality. [But] cutting-edge technology only exists in small companies staffed with intelligent people -- that's where the funding should go. So basically, it is not very fair [here]. But if it's unfair, why not take advantage of it?

Second, the reality is if you're 20 percent better than other people, they envy you and try to kill you with nails and hammers. But if you're 200 percent or 300 percent better, they're just going to leave you alone. That's us.

What is your work style? Do you just work all the time?

Well, work and pleasure are exactly the same things for me. I don't think I would like to only be a CEO, but [I] enjoy being CEO. I still do programming. Like this i-Appli thing, this [mobile online shopping] mall; everything is either directly programmed [by me] or [I] designed [it] and asked engineers to program it. I design 100 percent of our company products. And as CEO I have the goal of controlling net income. That's the only way that we can be independent and make what I think is right -- [to achieve] the hyperself architecture. Persuading investors in Japan is a lot tougher than those in the States.

Could you explain what hyperself architecture is about?

Well, many famous researchers -- like Roger Schank, my first adviser -- have said that 90 to 95 percent of intelligence is of the animal kind. Only 2 to 5 percent is of the human kind. We have not yet [been able to create even] the animal [level] of intelligence. Sony's Aibo is not as smart as your dog. Maybe that could take five to 10 years. We don't call that AI; the rest -- the 2 or 3 percent -- that's what we call AI. Hard-core AI researchers, like me, have never given up on it and are still trying to tackle this 2 or 3 percent. That's what I [believe] is different from agent technology.

Agent-based [systems] promise to do 100 percent on the machine. We say that the 2- to 5-percent fraction [representing] human intelligence [is not replicable by machines]. The rest, 95 percent or more, can be done by programs. So I call [the combination of the two] the hyperself architecture. What [this theory means] is extending your bodily functioning into cyberspace but your intelligence stays within your body. So basically, it's the opposite of [how] agent [technology works].

With agent technology, you have your secretary [in] your machine. This secretary is [supposedly] smart [enough that] it does everything for you, if you talk to it in natural language. Well, that's not going to ever happen, because it will never be able to understand what you really mean by just talking to it.

[With the hyperself architecture], a decision is made by you. So you actually exist in cyberspace, instead of talking to [some sort of] agent in cyberspace. That's hyperself architecture, and that's what we develop here. So basically, what we are trying to do is directly connect your unconscious [self] to intelligence that is not [human, but] is artificial. Your program [in the machine] is becoming part of you, as an unconscious part of your brain.


Freedom Mail: The first email software, CRL claims, to work in conjunction with NTT DoCoMo's i-Appli Java system. Freedom Mail is secure and considerably reduces the number of (expensive) packets required to provide enterprise mail functionality on mobile phones. CRL also produces e-applinet, a Java application download site for i-mode where individuals and companies can sell their own i-Applis, and a digital content delivery system.

Next Generation Home Server System: A government-sponsored generic distributed network middleware system for facilitating connection and remote control of AV devices, as well as safe distribution of media that is legally protected by intellectual property rights. The Home Server provides "smart" control of AV devices with the use of a case-based reasoning engine. Users can leave details of control up to the server or get feedback on operations that are currently available with connected devices. Communication between server and client software is done via XML over the IEEE1394 network.

But at the same time, we probably shouldn't discount agent, because there's a lot of work going on with agent, especially in mobile. Many people think the character/agent-based networking, social networking or e-commerce is going to be a big thing.

Sure. I'm not denying the agent research. What I'm doing could even be categorized as agent research. What I'm saying is their underlying philosophy is different from hyperself.

It's like [riding] a bicycle. Once you learn how to do it, you never have to think about it, right? Now it's unconscious. So most of this type of muscular activity can become unconscious. And that's still industrial revolution technology. But if you can do information processing unconsciously, that's a different technology. That's hyperself. [For example,] your robot will act in a way that you want it to. You just think, oh, this room has to be cleaned, even at the unconscious level. And your robot starts sweeping [the floor]. This [hyperself function still] requires the robot technology that is currently carried out by agent technology.

It must be hard to find Japanese investors, especially with a term like hyperself architecture.

Right. Japanese investors are [even] more short-term--oriented than US investors. US venture capital, of course, will intervene [in your company] and sue you and everything, but they know you have to take one year to develop [a product] to sell. In Japan, if you get the funding in June, they want the sales in August. And by March fiscal year-end, you have to be in the black. We're not -- so that's why our shareholders don't like me. By every definition, being a high-tech venture company like us, AI is a lot tougher to do in Japan than in the US.

How do you deprogram cult followers?

What I do is to identify logical discrepancies and explain to them that their mysterious experiences or perceived miracles of God are caused by [normal] brain functioning and, therefore, [can] happen to anybody. You can deconstruct cult theory, but not [a cult member's] personal experience with God. One key for deprogramming is to weaken the impact of the actual experience. What I do during the session is not clinical counseling, but [rather] a debate with the client.

You're an ardent student of Buddhism, practicing Zen every week and studying Buddhist sutras. Why did you get absorbed in Buddhism?

Neither of my parents followed any religion, although my grandfather practiced Zen Buddhism -- [which] could have been Christianity if I were born in the States. But I naturally followed Buddhism, as I was born in Japan. I studied Catholicism at Sophia University (in Japan), although my major was linguistics. The logic of Christianity is based on the real world. All the miracles described in the Bible are real scenes. What Buddhism portrays as its world view is beyond reality. That's why I'm attracted to Buddhism. The intricate logical consistency observed in Buddhist sutras resembles cyberspace. Human beings are born with an inner "cyberspace." Ever since I was a child, I have had a great interest in [this] space.

Your interest in AI must stem from your curiosity about the inner space.

I always wanted to be a rocket scientist when I was a little kid. Well, at one point, I learned rockets are only for the physical world. And I wanted to create a rocket for cyberspace. And that's when I started programming, writing games. I went to Boston to Applefest to sell my game in 1981.

Once you uncover the mechanisms of human psychology, you will want to write it down in some form, like letters and figures. That's how AI research was started -- describing human mechanisms in figures. AI's premise [in] programming human psychology or intelligence is that human intelligence is describable in an abstract manner, specifically with [numbers] or mathematics. Philosophy and mathematics are based on the same logic. For me, mathematics is a tool of philosophy.

What is your ultimate goal?

I'd like to found an authentic liberal arts college where students study history, literature, mathematics, and philosophy for four years. I once taught at a Japanese university, but would like to create an ideal school. SFC [Shonan Fujisawa campus] at Keio University is the kind of school I dislike most because its emphasis on teaching is how to make money. With my definition, universities are the institutions where you study academic subjects, not practice business. If students are looking for work experience, they should do it outside of school like they do in the States. Some people might think liberal arts are not useful. But that's what liberal arts are about: laying the basic intellectual foundation. After finishing these courses, students could go to professional schools, [but] they need to study liberal arts in order to survive in cyberspace. You can't create anything new or authentic unless you are capable of comprehending knowledge. @

Note: The function "email this page" is currently not supported for this page.