Wednesday, October 22, 2008

We, Robot--Part 2: Artificial Intelligence Today

Artificial Intelligence: the ability of computers to perform functions that normally require human intelligenceEncarta World English Dictionary, 1999.


Dr. Mark Humphrys of Dublin City University defines artificial intelligence as "engineering inspired by biology.” Today's robots and AI systems are no smarter than insects. Despite this current limitation, there are many reasons to sit back and enjoy the myriad of services technology has created for humanity through AI systems. AIs now play chess, checkers, bridge, and backgammon at world-class levels (e.g., IBM's chess computer, Deep Blue, beat Garry Kasparov, the world champion, in 1997). They compose music, prove mathematical theorems, synthesize stock option and derivative prices on Wall Street, make decisions about credit applications, diagnose motor pumps, and act as remedial reading tutors for elementary school children. Robots mow your lawn; conduct complex scientific research, surveillance and planetary exploration; track people; play table soccer; and act as pets. But they can't "think" like you and me. And they don't possess common sense . . . yet.

Today, AI systems are still nothing more than glorified adding machines or "idiot savants," capable of manipulating vast amounts of data a million times faster than humans. AIs can't understand what they are doing and have no independent thought. They also can't program themselves. Today's most complex robots use a simple feedback mechanism to move and act (e.g., Attila, MIT's "insectoid" robot), based on paradigms used in nature to simulate intelligence. Like real insects, these automatons are capable of making their own decisions (e.g., symbolic AI like Shakey, the first mobile robot built in 1969, which contained an internal model of its micro-world), as opposed to the industrial robots on assembly lines, which are pre-programmed. In the final analysis AIs are still an oxymoron.

Artificial intelligences still have a long way to go before attaining anything remotely close to a human's thought process and achieving what we call "common sense" (e.g., nothing can be in two places at the same time). But the new approach to AI known as "nouvelle AI," pioneered at MIT in the late 1980s by Rodney Brooks, appears more likely to attain complex reasoning than previous approaches. The previous top-down approach championed by Douglas Lenat and others attempted to endow AI with an encyclopedia of "common sense" (e.g., Cyc).

Instead, Brooks' team uses bottom-up biology-based models of intelligence by implementing a long history of interaction with the world and other biology-based intelligent systems, rather than force-feeding abstract reasoning and logical deduction. This is called "situated AI," the building of embodied intelligences situated in the real world and following the process of "the normal teaching of a child." As Kaku said of this philosophy of AI: "learning is everything; logic and programming are nothing." According to Dr. Kaku, the most successful AI systems, like Brooks' biology-based models, are those that learn like we do, through trial and error (e.g., Terry Sejnowski's neural network, NETalk, that learned the English language heuristically).

In the meantime, AI components are getting smaller and more affordable. The first nanochip was produced by the semiconductor industry in 2000, not only packing more transistors per cubic centimeter but also lowering the cost per transistor, increasing the speed of microprocessing, and permitting a whole new array of uses for and within humans. Which brings us to the two major areas of AI development: 1) robots and AI systems external from humans; and 2) interactive implants inside or on human bodies.

External Systems

Regarding the first, Dr. Michio Kaku, cofounder of string field theory and author of Hyperspace, describes a new branch of AI research called heuristics, which would codify logic and intelligence with a series of rules and would permit us by 2030 to speak to a computerized doctor, lawyer, or technician who could answer detailed questions about diagnostics or treatment. These "intelligent agents" may act as butlers, perform car tune-ups, perhaps even cook gourmet meals.
However, despite their many human-like characteristics, such systems remain a far cry from achieving what we call "real intelligence." They would still be glorified automatons, albeit sophisticated diagnostic tools, taking on the form of a human figure on a screen or a humanoid robotic form. Although they would give the appearance of human intelligence and likely pass the Turing Test, these essentially pre-programmed systems would not "think," be "self-aware," or have "common sense" as we know it. According to Dr. Kaku, this level of consciousness, which is the ability to set one's own goals, may only be achieved after 2050, when he predicts the top-down and bottom-up approaches to AI development will meet, giving us the best of both.

Internal Systems

Regarding the second area of AI development, research labs are already developing a vast array of "intelligent clothes," which can interface with us and enhance memory, awareness, and cognition. Along with these exterior enhancements, microchip implants, such as radio frequency identification devices (RFIDs) inserted in humans, are gaining momentum. On May 2, 2002, the first human was "chipped" for security reasons; the idea was that if he became ill or impaired, professionals could access his medical history by scanning his microchip implant. The next step in the evolution of this technology is the ability to track people using GPS and connect to additional personal information of importance such as medical data. Science fiction writer Robert J. Sawyer calls such devices "companions," as used by an alternative society in his Neanderthal Parallax trilogy. Since 9/11 the idea of national identification has gained much approval by US citizens.

Medical implants are not new; they are used in every organ of the human body. More than 1,800 types of medical devices are currently in use. These run the gamut from heart valves, pacemakers, and cochlear implants, to drug infusion devices and neuro-stimulating devices for pain relief or to combat certain disorders like Parkinson's.

On October 14, 2003, the Associated Press announced that monkeys with brain implants could consciously move a robot arm with their thoughts, representing a key advance by researchers at Duke University, who were hoping to permit paralyzed people to perform similar tasks. Paul Root Wolpe, a bioethicist at the University of Pennsylvania, declared, "We're on the verge of profound changes in our ability to manipulate the brain." New developments in neuroscience promise to improve memory, boost intellectual acumen, and fine-tune emotional responses through brain implants.

This excites transhumanists, who seek to expand technological opportunities for people to live longer and healthier lives and enhance their intellectual, physical, and emotional capacities through the use of genetic, cybernetic, and nanotechnologies. From the transhuman perspective, "in time the line between machines and living beings will blur and eventually vanish, making us part of a bionic ecology."

The US National Science Foundation and the Department of Commerce initiated a program that "wires together biotechnology, IT, and cognitive neuroscience (under the acronym of NBIC) into one megatechnology by mastering nano-scale engineering." In a detailed report that projected twenty years into the future, the authors declared that: "understanding of the mind and brain will enable the creation of a new species of intelligent machine systems." The report envisioned technological achievements that would seize control of the molecular world through nanotechnology including the re-engineering of neurons "so that our minds could talk directly to computers or to artificial limbs." Brain-to-brain interaction, direct brain control devices via neuromorphic engineering, and retarding of the aging process would then be feasible.

Future Systems

When might all this be possible? Some of it is already occurring (e.g., the recent work of Duke University mentioned above). Dr. Kaku asserted that "after years of stagnation in the field of artificial intelligence, the biomolecular revolution and the quantum revolution are beginning to provide a flood of rich, new models for research." Drawing on the insight of AI pioneers like Hans Moravec of Carnegie Mellon University, Dr. Kaku suggested that this may happen only once the opposing schools of AI research amalgamate, combining all the ways humans think and learn: heuristically, by "bumping into the world;" by absorbing certain data through sheer memorization; and by having certain circuits "hard-wired" into our brains. He predicted that this would occur sometime beyond 2050, at which time AIs would acquire consciousness and self-awareness. MIT artificial intelligence guru and transhumanist, Ray Kurzweil, agreed in his 1999 book The Age of Spiritual Machines, that sentient robots were indeed a near-term possibility: "The emergence of machine intelligence that exceeds human intelligence in all of its broad diversity is inevitable." Kurzweil asserted that the most basic vital characteristics of organisms such as self-replication, morphing, self-regeneration, self-assembly, and the holistic nature of biological design can eventually be achieved by machines. Examples include self-maintaining solar cells that replace messy fossil fuels and body-cleaning and organ-fixing nanobots.

When you mention AI and robotics, we tend to polarize. Some of us are truly excited by all this and others of us are truly frightened (see this previous post of mine on artificial intelligence). Then there are those who are both excited and frightened! When I sat on several panels dealing with AI, robotics and science at Vcon (Vancouver's science fiction and gaming convention), I found this to be the case. This was not so much determined by intelligence, knowledge or insider's information; I think it was more a result of our own world-view and faith in humanity: are we optimists or pessimists?

This is an excerpt of an article I wrote in Strange Horizons several years ago, entitled "AI: Changing Us, Changing Them" (for more of the article and comments, go here). Go to the right sidebar for more articles from my series on artificial intelligence under my favorite science posts. Ciao!

Recommended Reading:

Humphrys, Mark. "Reaching out to over 15 million people" in Robotbooks.com, The Future of Artificial Intelligence.
Kaku, Michio. Visions: how science will revolutionize the 21st century, Anchor Books Doubleday, New York. 1997. 403pp.
Kurzweil, Ray. The Age of Spiritual Machines, Penguin Books, New York. 1999. 253pp.
Ford, Kenneth & Patrick J. Hayes. "On Computational Wings: rethinking the goals of artificial intelligence", in Scientific American Presents: Exploring Intelligence 9 (4) Winter. 1998.
Copeland, Jack. "What is Artificial Intelligence?" in www.alanturing.net. May 2000.
Hutcheson, G. Dan. "The First Nanochips," in Scientific American 290 (4): 76-81. April, 2004.
Pentland, Alex P. "Wearable Intelligence", in Scientific American Presents: Exploring Intelligence 9 (4) Winter. 1998.
CBC News: www.cbc.ca/stories/2002/11/05/consumers/microchip_021105, May, 2002; Julie Foster in WorldNetDaily.com, 2001.
Gaitherburg, MD. Medical Implant Information Performance, and Policies Workshop, Sept. 19-20, 2002. Final Report.
Sawyer, Robert J. The Neanderthal Parallax Trilogy, TOR.
Alex Dominguez (Associated Press). "Monkeys move robotic arms with their minds," in: The Vancouver Sun, October 14, 2003.
Center of Cognitive Liberty & Ethics: www.cognitiveliberty.org/neuro/bailey.html.
WTA World Transhuman Association. transhumanism.org.
Thomas, Jim. "Future Perfect?" in The Ecologist, May 22, 2003.
National Science Foundation and Department of Commerce. "Converging Technologies for Improving Human Performance: Nanotechnology, Biotechnology, Information Technology, and Cognitive Science." 2002. 402pp.




Nina Munteanu is an ecologist and internationally published author of novels, short stories and essays. She coaches writers and teaches writing at George Brown College and the University of Toronto. For more about Nina’s coaching & workshops visit www.ninamunteanu.me. Visit www.ninamunteanu.ca for more about her writing.

8 comments:

Anonymous said...

I would like to add one commet:
Broooks has an European counterpart, Rolf Pfeifer. The webpage of his lab is www.ailab.ch and he has two books on new AI
"Understanding Intelligence" 2001 and ""How the body shapes the way we think" 2005

Good article!

Nina Munteanu said...

Cool! I'll have to check what Rolf Pfeifer is doing. Many thanks for the comment and sharing the link!

Anonymous said...

I heard a company that creates robots and technology calls itself 'Cyberdyne' after the group that started the terminators!

Nina Munteanu said...

EEK! Science fiction or fact?... Doesn't exactly promote confidence does it? Of course, we know how that all ended...

graywave said...

Hmmm. I think you're being a bit unfair to insects there. I can't think of a single robot around today that wouldn't be blown out of the water by your average insect.
And 'heuristics' is definitely not a new approach in AI - the idea was well established when I started in the field in the early 1980s.
But the general oint you make is a good one - AIs are really dumb right now but they can still be useful. However, don't hold your breath waiting for the 'singularity' to happen!

Anonymous said...

Pretty Good Article Nina!

A couple of thoughts -
Computers do not yet "compose" music. Computers do create some musically valid pieces based on a set of rules that a programmer/musician creates. A company out of Victoria created program called "Band in a Box" where the user (me) can enter a set of chords, the key for the piece, tempo and choose a "style" from a list that comes with the software and then create a song with a musically valid Bass, drums, piano, guitar parts. I can then tweak the results to tell the program more about the song structure and where to place drum fills. It even includes a limited amount of randomness (depending on the style) such that every time you play the song it comes out a little different based on the parameters set up for the style. Works pretty good for a guitar player that does not play drums and bass. But in the end, the composition based on my input and the creator of the program and the computer is just the tool.

Similar to mandelbrot set generators. The generators can create very interesting pictures but it is just a stylized rendering of a mathematical formula.

BTW: With the GPS embedded in most smart phones the real time tracking of people is already here. There is an application for Apple's iPhone that allows a group of friends to share their locations with each other - Bob tweets that the music is hot at the club he's at and Sue, Chuck and Barb can see on a Google style map where Bob is and how to get there and then converge (Unless Bob is boring then they know where stay away from :)

Never confuse game playing machines with intelligence. They follow very specific rules. Regarding Kasporov vs Deep Blue and Chess: Sucess in chess is proportional to how many possible next moves you can "look ahead" to then choose the move that will take you on the path with the highest probabilty of winning. Rookie chess players will only evaluate 1 or 2 levels ahead and only for a limited number of pieces in play. The better players move out to 4 moves or more and look at more pieces. Deep Blue was set up to be able to look much farther out than any human could. But the weighting of what determines a better outcome was based on programmers creating models bsaed on advice from chess experts. The other important factor, chess has a big emotional context - Intimidation, bravado, deceptions, key elements that Kasporov could not use against Deep Blue. - Note: if you want an edge against a computer - Throw it a move that no chess player in his right mind would do (might not trip up Deep Blue but I have used it against some computers to win).

Time to drop to minimal power and enter sleep mode....zzzz

Brie's Guy

Nina Munteanu said...

Cool thoughts, Brie's guy! Very good points about the emotional aspects of chess... Lots of human-related tactics just wouldn't work on Deep Blue, who might be construed to be as boring as our friend Bob with the cool phone... :)

Nina Munteanu said...

LOL, Graywave... you're probably right about the insects. I think they're certainly FASTER than most AIs! And, yes, heuristics is something even we ecologists were playing around with back in the late 70s (ohoh... this ages me!)like in the heuristic models of reservoir zonal stratification and movement, for instance. Actually, a lot of ecological investigations use heuristic models because the environments we are studying are so complex and dynamic that determining paradigms through "learning by doing" is sometimes the best way to start. Chaos theory and non-linear relationships are another approach to studying ecology.