This book has been brewing for nearly fifty years, since preschool adventures with a mechanical construction set implanted the consuming notion that inanimate parts could be assembled into animate beings. The brew bubbled over in an article in 1978, a book in 1988, and this work in 1998.
In 1978 I was at the Stanford Artificial Intelligence Laboratory finishing a thesis on a seeing mobile robot, and disagreeing with SAIL's pioneering founder John McCarthy, whose focus was computer reasoning. Reasoning programs had bested humans in specialized areas and John held that existing computers lacked only the right programs to be fully intelligent. Computer vision convinced me otherwise. Each robot's-eye glimpse results in a million-point mosaic. Touching every point took our computer seconds, finding a few extended patterns consumed minutes, and full stereoscopic matching of the view from two eyes needed hours. Human vision does vastly more every tenth of a second.
No machine in 1978 even approached everyday human sensory, motor, or reasoning skills. Worse, progress toward those goals seemed to have almost stalled. Reacting against a pervasive pessimism, I argued in an evolving tract that humanlike performance needed millionfold greater computer power, but might be attained in a crash program in as little as ten years by interconnecting millions of then-new microprocessors. Analog science fiction magazine bought a version of this polemic, titled "Today's Computers, Intelligent Machines and Our Future," in 1978.
The Analog article implausibly called for someone to invest billions of dollars in computer hardware to possibly produce one humanlike machine. Updates of the article were more sedate, offering human competence in multimillion-dollar computers in twenty years. But in AI research, power per computer was falling! Mainframe computers owned by institutions were replaced by minicomputers owned by groups, then by workstations and personal computers. Still, the price/performance ratio was improving, and the decline in individual machine power had to bottom eventually. I resolved to redo my estimates more carefully, and expand the essay into a book. In 1988's Mind Children: the future of robot and human intelligence, humanlike performance was rescheduled for ten-thousand-dollar personal computers in forty years.
By 1990 personal computers outpowered 1978 mainframes, and formerly intractable problems began to find solutions. Home computers soon recognized printed and spoken words, and experimental robots cruised hallways and highways. With a firmer launch ramp than Mind Children, this book projects humanlike competence in thousand-dollar computers in forty years. A slight rise in the estimated difficulty has been partially offset by faster growth in computer power.
Mind Children's near-term projections have held up pretty well, with one big exception. There are still no mobile utility robots to help us around the house. In 1988 several small companies had developed security, cleaning, and transport robots that navigated prearranged paths by sensing walls and special markers, somewhat like insects. I had hoped those would lead to more advanced (and more marketable) robots that perceived their surroundings and roamed freely. Alas, the first machines did not attract customers, and the companies fizzled. There are perceiving robots now, but only in research settings. But, this time for sure! In line with refurbished predictions for utility robots set forth in Chapter 4, I am pursuing a business plan to equip existing factory vehicles with enough spatial awareness to work in unmapped areas by about 2000, expanding to mass-market machines like small home vacuum robots by 2005, then more capable multipurpose "universal" robots by 2010. Its is becoming a race: many other robot ventures are afoot.
Chapter 4 anticipates four generations of universal robot, each spanning a decade. The first has a lizardlike spatial sense, the second adds mouselike adaptability, the third monkeylike imagination, and the fourth humanlike reasoning. The uniform schedule comes from matching each prototype animal's brain to steadily rising computer power. But it took three hundred million years for our ancestors to evolve from tiny ancestral chordates (of insect-like complexity, like robots now) into something lizardlike, an additional one hundred and fifty million to become mouselike, only seventy million more to be monkeylike, and just another fifty million to become human. My time scale is probably too bold in the short run and too timid in the long run, a typical problem with technological forecasts. A future book (dated, say, 2008!) will surely contain a course-corrected projection.
Until then, look for new developments, commentary, references, and color illustrations on my world-wide-web page containing the search word "hpm98book," found, for now, at the location: