Google Internet Summit 2009: The Future

The Winner of a Twenty-Year Bet

"How many people have in their lives a 2 to 10 percent chance of dramatically affecting the way the world works? When one of those chances comes along, you should take it."

- Douglas Lenat

From a very early age, science has been an outlet for curiosity for Douglas B. Lenat, born in Philadelphia in 1950, and fortunately for Lenat, science is a theme that resonated throughout his life. Although he did not receive very good primary education, he had a natural talent. In 1967, for example, he became a finalist in an International Science Fair for his work on the closed form definition of the nth prime number which was judged by a company of scientists, researchers, and engineers. In 1968, Lenat entered the University of Pennsylvania initially pursuing a degree in physics and mathematics. He quickly changed his mind however after taking a course in 1971 with John W. Carr III, a computer science professor at the University of Pennsylvania who had introduced Lenat to artificial intelligence. The research in the field was just starting out, so Lenat decided to pursue it. In 1972, he attended CalTech for a PhD program in applied mathematics and computer science but promptly realized that a lot of work on artificial intelligence at the time was done at Stanford, so he transferred in order to begin his work with McCarthy. Unfortunately, McCarthy left on sabbatical for MIT the same fall that Lenat arrived at Stanford, so instead, Lenat ended up working with Cordell Green on the theory of automatic programming that tested falsifiable hypotheses. Falsifiable hypotheses were an extremely important part of artificial intelligence because AI has to "interact directly with the external world" (228). Therefore, purely mathematical or logic predictions about these expert systems have to be tested. Lenat also applied these theories, and all the heuristics theories, to his doctoral thesis - a modest Lisp program that worked out various mathematical concepts named Automated Mathematician (AM); to this day, this thesis remains "one of the most original AI programs ever written" (229). After he had received his PhD in 1976, Lenat developed on AM with a project named Eurisko which brought a lot of interesting applications - including circuit design and minimization, solutions to tactical game problems, and searches for missed loopholes in game situations - to the table but never took off in the markets because it could not be extended to other various useful domains. In 1984, Lenat left the academia to pursue a business opportunity embodies by the project Cyc, a piece of artificial intelligence which possesses a large part of the entire existing body of knowledge. A lot of research, design, and implementation have been put into the project, but its success have yet to be seen. The project is currently considered to be one of the most controversial in the discipline, yet Lenat still "believes that AI projects like Cyc can become 'knowledge utilities'" in the future (242). Social reception of the technology cannot yet be considered; a much better question is - will Cyc actually work? According to Shasha and Lazere, yes, but only in part, due to various flaws in the current approach to expert systems and lack of understanding of the inner workings of the human mind. With all this evidence in mind, it is clear why Lenat remains "the boldest kind" of explorer of artificial intelligence leading the way to a bright future in the world of computing (242).

The Keeper of the Power of Knowledge

"There are three important things that go into building a knowledge-based system: knowledge, knowledge, knowledge. The competence of a system is primarily a function of what the system knows as opposed to how well it reasons."

- Edward Feigenbaum

Edward A. Feigenbaum's story begins with a tragedy. He was born in Weehawken, NJ in 1936, and just before his first birthday, his father tragically passed away. Feigenbaum's stepfather, an accountant of a small local bakery, was in turn charged with the job to ignite his interest in science and technology, and so he did by taking young Feigenbaum to the Hayden Planetarium in New York City once a month to all the new exhibits. In 1952, Feigenbaum started his college career at the Carnegie Institute of Technology (now known as Carnegie Mellon University) majoring in electrical engineering, per his parents' request. Computer science did not yet exist for the average undergraduate at Carnegie, so Feigenbaum "began taking courses at Carnegie's then new Graduate School of Industrial Administration" (210). These courses, via professor James March, were the first to introduce him to the ideas of game theory and a lot of other work done by a Hungarian mathematician John von Neumann. Feigenbaum also had a rare opportunity to attend a course read by Herbert Simon, a professor at Carnegie in the fields of political science, sociology, and economics, as well as a former federal  administrator for the Marshall Plan, on mathematical models in social sciences. One day,
Simon, and his co-lecturer Allen Newell, announced that they had invented "a thinking machine" called "the Logic Theorist" and handed out user manuals for the IBM 701 (211). Feigenbaum took the manual home, read it, and finally realized what he wanted to do. The idea of the Logic Theorists was interesting: the "...program attempted to discover proofs in propositional logic" based on some other logic that is already known to the program using educated guessing problem-solving technique formally called a heuristic by a Hungarian mathematician George Polya (212). Fascinated by these ideas, Feigenbaum stayed at Carnegie with the School of Industrial Administration until 1956 when he graduated with his PhD in electrical engineering. His doctoral thesis involved more work with the Logic Theorist as he attempted to further model human problem-solving abilities, such that he could draw some conclusions about human problem solving. It turned out to be a very hard problem, but it was completed under the name Elementary Perceiver and Memorizer (EPAM), and it is still used today at Carnegie Mellon. More specifically, the program modeled how humans are able to memorize pairs of unrelated, nonsense words in a stimulus-response setting. The process included a training portion and a testing portion, and from a psychology standpoint, provided a lot of insights into the working and abilities of short-term memory. This research lead Feigenbaum first to the University of California at Berkeley and then, eventually, to Stanford where John McCarthy was doing his work with artificial intelligence in 1965. At Stanford, Feigenbaum began to formulate his thoughts about expert systems. In a collection of papers Computers and Thoughts that he co-edited with a colleague Julian Feldman, he first began advocating for further exploration of computer-based processes of induction. In 1964, Feigenbaum, Joshua Lederberg, the chairman of the Stanford genetics department, and a Stanford chemist Carl Djerassi, began their work on a joint project Dendral which attempted to develop a "Mars probe that would land on the surface of the red planet and explore for life or precursor molecules" (216). The project, which a year later had been declared successful, is considered to be the the world's first true expert system capable of determining chemical structure of molecules even better than most humans could. This project also laid out the framework for expert systems in general: "a set of data, a set of hypotheses, and a set of rules to choose among the hypotheses" (218). Soon, the company developed all kinds of other expert systems including Mycin, which was meant to help doctors diagnose infectious diseases and recommend treatment to numerous patients, and airline management systems, which supported airport traffic controllers. In the end, the idea of standardized knowledge turned out to be key to the expert system structure. The more knowledge exists in the system, the better, more efficient, and simply smarter the expert system can be. Despite the extensive work that Feigenbaum did in this area, expert systems are in their developing stages in the world of computer science, but even Feigenbaum himself believes that "the expert system will gain its rightful place as an intelligent agent that can cooperate with people to solve some of the world's more challenging problems" (222).

Artificial Intelligence

The Biologist of Computing

"Clearly, the organizing principle of the brain is parallelism. It's using massive parallelism. The information is in the connection between a lot of very simple parallel unit working together. So, if we built a computer that was more along that system of organization, it would likely be able to do the same kinds of things the brain does."

- Daniel Hillis

W. Daniel Hillis's interests in science and technology came from his parents. His father was an epidemiologist while his mother was very interested in mathematics, and both his parents went to great lengths to instill curiosity in these subjects. Hillis's curiosity and craftiness enhanced his experience with technology during his youth. His first real exposure to the world of digital computing was in the late 1960s when he had the chance to look at George Boole's An Investigation of the Laws of Thought (1854) which outlined the principles of elementary Boolean algebra. After toying with these ideas, he eventually learned to program. In 1974, however, he entered MIT "determined to find out how the brain worked" planning to major in neurophysiology (192). At MIT, Hillis met Marvin Minsky and John McCarthy and fully discovered the vast world of computing and started working at MIT's Artificial Intelligence lab in the LOGO group on a project involving computer technologies that followed the evolutionary principle of emergence which states that "interacting agents will adapt, through a process of selection, a mechanism for survival" (193). Soon, Hillis' idea for the Connection Machine was born. In attempt to mimic the massive parallelism used by the brain, the Connection Machine was designed to be a computer made up of thousands of processors all linked together each with its own control and its own memory. The machine was connected, initiated, and set "free" to run in hopes to discover the emergence of new smarter technology from the pre-existing one. A perfect project for a lover of both biology and computing, the Connection Machine was the perfect project for Hillis, but it is clear that the Machine is only the beginning to our understanding of evolutionary trends in computing.

The Driver of the Digital Fast Lane

"Speed is exchangeable for almost anything. 
Any computer can emulate any other at some speed."
- Burton Smith

Burton J. Smith's career had a very rough start. Smith, born in 1941 in Chapel Hill, NC, moved with his family to New Mexico when his father, a professor of chemistry, was offered at job as the head of the University of New Mexico's chemistry department. Smith was constantly fascinated by technology, but only after he came back from the military did he know what exactly he wanted to do - design of electronic devices. Therefore, he graduated in 1968 with a B.S. in electrical engineering from the University of New Mexico and went on to MIT where he completed his doctorate in 1972. A lot of Smith's work from then on was focused on optimizing the hardware structures that support the newly implemented pipelining process - both pipeline parallelism and multiprocessor parallelism - used in computation. Smith and his colleagues at Denelcor, a small computing company based in Denver, Colorado, strove to create a supercomputer which would employ high-efficient parallel processing. In other words, they wanted "to design a machine that would perform and operation as soon as its inputs were ready" (180). They called this approach, which was first developed by Jack Dennis of MIT in the 1970s, "dataflow architecture" (180). After some time in development, it was clear that this approach has a very significant impact on the discipline, as dataflow architecture is also applied to digital memories and networks of all kinds, just to name a few. Smith's success eventually followed him to the Tera Computer Company in Seattle, WA where he had another revelation that boosted performance of the pipelining process - "different operations within a task may sometimes be executed out of order" (186). This idea significantly sped up data processing in computers and ultimately led us to the modern-day process.

Computer History Museum - The IBM System/360 Revolution