Hi, robot! – revisited

Articles

In June of 2000 I wrote an article called “Hi, robot!” (which you can still find on my website) in which I said, among other things, that a robot vacuum cleaner for households would be among the first robots available, that a general-purpose robot housekeeper might be available within 7-10 years, and that human-like commercial robots would be generally available within 20 years, and probably sooner. Roomba and Scooba have proven the first prognostication, and Toyota’s work on a household robot, to be brought to market sometime in 2010, looks likely to make the second come true. Accordingly, I thought it was time I revisited the subject, and wrote the following article on a recent plane ride.

A Coming Discontinuity

Our natural tendency is to assume that tomorrow will be very much like today, but that’s a mistake. In particular, there are times of discontinuity when what happens next is very different from what happened before. We are approaching one of those times in the field of computing, artificial intelligence (and related software), and robotics.

Most people are aware of Moore’s Law, a rule of thumb coined (and refined several times) by Gordon Moore, one of the founders of Intel. Although Moore couched his law in terms of the number of transistors on a chip, the financial effect is that computers will double in speed, and halve in price, every 18 months. Ray Kurzweil, whom Inc. magazine described as the 21st century Edison, plotted the increase in cost-effective computing from the beginning of the 20th century, and found that Moore’s Law is too conservative. Kurzweil reckons that today, a $1000 computer is doubling in speed every 11-12 months, and that the period of time is shortening with each successive doubling. At present rates, this would imply that 10 years from now, a $1000 computer will be 1,000 times faster than computers today, which, in turn, implies that computers will be qualitatively different as well as quantitatively different. In every day life, this may be hard to see as most contemporary computers are disguised as something else, such as a BlackBerry, an iPod, or a smartphone. But one area where the discontinuity will be in robots, which are not widespread or commonplace.

Today, robots are viewed with a kind of amused indulgence. The robots we’re aware of are either bolted to a factory floor, live only in science fiction stories and films, or are cute little toys that can sort-of walk, kick a ball, and respond to rudimentary commands. Few people take them seriously.

Yet, more than 1 1/2 million household robots have already been sold at retail in the task-specific forms of Roomba and Scooba, which, respectively, vacuum and wash floors. And Toyota is investing billions of yen in developing a full-fledged household robot, which it currently plans to begin marketing in 2010 (undoubtedly to be sold first in their home market, Japan). So robots are already entering into our lives – and a 1,000-fold increase in computer power over the next 10 years will make them smarter and better at what will seem like an astonishing rate.

AI and robots come of age

The writing for these changes has been on the wall for some time, but the blossoming of this apparently new field is now coming to fruition.

If you look at the field of what is loosely called “artificial intelligence” (which is actually not a very good description), IBM’s Deep Blue computer beat the human chess champion, Gary Kasparavo, in 1997, demonstrating that computers can manage abstract thinking of great complexity. Seattle traffic is directed by an AI system called SmartPhlow. AI is used extensively by hedge funds for trading systems (not that that’s necessarily a recommendation). AI is now managing global corporate supply chains through the Internet to keep JIT processes moving smoothly. In a revealing counterexample, Formula 1 racers have, until very recently, had computer software managing their traction control, so that it was easier to drive faster, and corner at higher speed without spinning out of control. That’s now about to change – this kind of software has just been banned in order to allow the more skillful racers to win instead of the ones with the best software. A self-driving Chevrolet Tahoe with no one at the wheel or in the car, successfully navigated a simulated urban environment with streets, traffic lights, pedestrians, and so on, winning a $2 million prize from DARPA (the Defense Advanced Research Projects Agency of the U.S. Department of Defense, and the agency that created the Internet).

And it’s not just speed that matters – novel approaches and new tools are making a difference as well. John Koza is a pioneer in one such field, called genetic programming (“GP”), with which I have worked and have some familiarity. GP evolves software solutions through a form of machine learning rather than human-designed, hand-crafted coding. On one of his websites, Koza talks about how GP is now producing results that are, as he calls it, “human competitive”: “There are now 36 instances where genetic programming has produced a human-competitive result.…These human-competitive results include 15 instances where genetic programming has created an entity that either infringes or duplicates the functionality of a previously patented 20th-century invention, 6 instances where genetic programming has done the same with respect to a 21st-century invention, and 2 instances where genetic programming has created a patentable new invention. These human-competitive results come from the fields of computational molecular biology, cellular automata, sorting networks, and the synthesis of the design of both the topology and component sizing for complex structures, such as analog electrical circuits, controllers, and antenna.”

GP has already been used in a surprising range of industries, such as to create a crop optimization model for a major seed company, and develop a dynamic controller for weaving paper products for a major consumer products company.

NASA had software based on a set of differential equations to model the chemical kinetics of the combustion of jet engine fuel. The problem was that this model was very slow. They’d spent about $10 million trying to speed it up and had achieved a 3x increase in speed, which allowed them to produce one complete modeling run every three months. GP software “modeled the model” using genetic programming to evolve a solution in a period of weeks on a desktop computer, and came up with a simple function that described the mathematics of happens when you light-up a jet engine with a given fuel. This functional description was 2700x faster, 98% accurate, and cost $80,000. And, of course, you know the end of the story: because this was a solution from left field, and it looked too simple and didn’t cost enough, and because it rendered NASA’s $10 million investment virtually worthless, it was rejected.

This is not unexpected, because truly novel solutions are so far from the experience or background – so “outside of the box” – for most technical people that they don’t understand the new technologies, they don’t believe such technologies can produce superior results, and they therefore don’t seriously consider them. But GP is now a functioning reality, not just an interesting pipe dream, and it presages a day when computers can solve problems independently, including problems that humans have not solved or cannot solve.

And what does this mean? A few applications

Now imagine what will happen over the next 10 years as computers do, indeed, become 1,000 times faster, and our experience in applying and developing more sophisticated software continues to grow. Artificial intelligence will become widespread, and rapidly more sophisticated. Robot helpers will become more capable – although perhaps not to the level of the general purpose, Isaac Asimov-style positronic robot.

Computers will help you drive your car more safely, integrating a wide range of data, both from the traffic network, and computer-enhanced vision, radar, sonar, and laser range-finding. AI will assess the driving of people near you, looking for danger from others. Traction control will allow you to drive more safely – and may prevent you from doing something stupid, even if you want to.

Household robots will do housework, including windows. They will become an entirely new consumer product category, and will become the second most expensive piece of household equipment (after your automobile). They will, finally, become true labor-saving devices (if we can figure out the instruction manual…).

Transportation companies will employ robots and/or computers to supplement or replace human drivers in trains, trucks, and automobiles. (It won’t be necessary to replace pilots in planes – most contemporary planes can already takeoff, fly to a given destination, and land without a human touching the controls.)

Hospitals and health care clinics will employ robots in many forms. There are already robot surgical extensions, that act like extra hands for surgeons in procedure areas like neurosurgery and prostate cancer, but this will extend to many more areas of medical procedure. This is especially true as, with the baby boomers retiring, there won’t be enough human health care professionals to do the work. In most cases, robots and computers will suggest, refer, and act under the control or supervision of a human, but they will also serve as independent monitors, watching the health of at-risk individuals. Dementia patients will be watched to make sure they don’t wander off, or cause harm to themselves or others. People at risk for heart-attack or stroke will be monitored, heartbeat-by-heartbeat, and if they experience an attack, help will be called for them immediately, and in a way that is sure to elicit a response.

AI will design drugs targeted to specific groups of people, based on their genetic code and health threat, and do so faster, and much more cheaply than the massive, trial-and-error methods currently used by pharmaceutical companies. They will ferret out the causes of disease, and suggest diagnostics, cures, and treatments. And they will help doctors identify the specific threats from, say, cancer in an individual by identifying which cancer is involved, specifically where in the body it is located, what stage it is at, and how it can best be treated – and all of this far faster than doctors can do on their own.

If they could read your mind…

Public places that are currently monitored by closed-circuit cameras, will now be monitored by computers that will automatically seek to match faces and other individual biometric characteristics with the identities of specific individuals. In airports, this will mean that ordinary travelers will be identified, pegged as being non-threatening based on past behaviors, and allowed to proceed rapidly to their flights. Travelers identified as being potential threats, or whose behavior patterns are not known, will be tagged as being security risks, and screened in-depth by human security officers. Moreover, faces will be read by computer to determine whether someone is hiding something, lying, or unduly nervous.

The potential may also extend to schools. An individual student’s computer will be able to judge whether the student is comprehending the material she is studying, whether she’s bored, whether her attention is wandering, and, indeed, whether she is actually focusing on the material or not. This will allow computers to tailor material to the individual, seeking out those means of explaining and presenting material that are most effective for that individual, and altering the methods from minute-to-minute, if necessary, to help the student do her best and stay engaged with the material.

Does this sound Big Brother-ish? It should, because it is also leading to a world where computers, acting under human orders, will be capable of routinely violating our privacy, civil rights, and even the sanctity of the very thoughts within our heads, without our consent, or even awareness.

As with any technology, robots, computers, and AI over the next 20 years will be two-edged swords. But because they will be such incisive swords, and cut so deeply, they will have the potential to do great things for us – and to us. Watch carefully for what’s to come, for it will be dramatically different from what we are experiencing today.