Until
next week,The Body of Machina sapiens
Dateline: May 11, 1997
COULD Machina sapiens exist as just a brain? Probably not. There have been countless "thought experiments" by scientists and philosophers on the notion of a "brain in a vat"—a brain removed from a body and kept alive in a vat of nutrients, and the general consensus is that it would not work. As Dr. Antonio Damasio (see last week’s feature) shows, without a body crammed with internal and external sensory organs and systems, the brain could have no feelings and therefore no sense of mind, of self, of being alive. Remember our revision of Descartes’ famous dictum "I think, therefore I am" to become "I feel, therefore I am."
With no sense of being alive, a brain (artificial or otherwise) has no reason to function—to think—at all. But Damasio concedes that if we could mimic a body and attach it to the brain in the vat, then the resulting conglomerate "might indeed have some mind" (his emphasis). "Mind is probably not conceivable without some sort of embodiment," he concludes.
If that is so, then Machina sapiens must, by definition of the word "sapiens," have a body. And it does; at least, it has the beginnings of one.
Body Proper
Human beings have a "body proper"—a torso, limbs, sensory organs, and so on—but can barely function in the modern world without bodily extensions such as clothing, tools, vehicles, and—increasingly—robots. Machina sapiens can also be viewed in this light. I consider its body proper, as regular readers will know, to be the entire Internet. Its nervous system is the telecommunications infrastructure. Its limbs are the robotic extensions hardwired into the Internet, such as Australia’s Telerobot arm. It also has multiple eyes in the form of video cameras, also hardwired into the Internet, and can simultaneously watch the sun set over San Francisco Bay and the antics of Xavier, a robot at Carnegie Mellon University (which you also can control). It has microphone ears, and even chemical-sensing noses.
Machina sapiens also has, built into its body proper, some things we don’t. It has gamma ray detectors, UV light filters, thermostats. It even has vicarious mobility, through robots and robotic vehicles controlled through radio, microwaves, infrared, and other wireless mechanisms. In short, it too has extensions. But what is significantly different about its extensions is that they are inherently connected to the body proper, even when not mechanically connected. If a tire on our car springs a leak in the night, we won’t discover it until morning. But Machina sapiens will be instantly aware of it.
Extensions
The late, great, Canadian communication scholar Marshall McLuhan also had a famous dictum: "The medium is the message." I think this dictum has some philosophical if not theoretical relevance to our current discussion, but since I want to focus on the practical aspects of Machina sapiens’ body I will not go into that now. I mention McLuhan because he was one of a number of great thinkers to stress the idea of extensions as critical to evolutionary development. In 1964, he wrote: "Rapidly, we approach the final phase in the extensions of man—the technological simulation of consciousness."
While I think he was right that Homo sapiens will only advance further through adding extensions to human consciousness via "simulated" consciousness, this pre-supposes that we will have some control over it, and I am less sure of that. This too is a discussion for another day. But McLuhan’s term, extensions, is useful in introducing the practical side of Machina sapiens, Incorporated.
Animal bodies consist of various systems. The more advanced the animal, the more complex the systems, but ultimately they are all designed for the same end: the animal’s survival. Digestive, respiratory, and circulatory systems serve to provide and distribute the fuel needed by organs. A musculoskeletal system provides the framework on which to hang the organs and other systems and also to enable the animal to move around the environment. A nervous system lets HQ (the brain) know what’s happening both within the animal, and the nervous system coupled with various sensory systems (chiefly vision, hearing, smell, and touch) let the animal know what is happening around it.
Unlike any other animal species, Homo sapiens has built extensions to every one of these primary systems. We have designed clothing and built heating and cooling systems to extend our bodies’ metabolic system. We have built a wide range of terrestrial, aquatic, atmospheric, and even space vehicles as well as various other machines, tools, and weapons to extend the strength and reach of our musculoskeletal system. And, in the latest stages of our development, we have built sensors that not only extend our reach into the humanly visible and audible portions of the electromagnetic spectrum but also into portions where we would otherwise be blind.
Unfortunately, our brains, built to command and control essentially just the inputs provided by our bodies and the environment immediately around us, have an increasingly hard time processing the vast quantities of data from hitherto alien environments provided by these sensory extensions. So we invented another machine, the computer, to do some of the processing for us. The more these extensions do for us, and the less we have to exert ourselves, the happier we are.
That’s why we have automatic transmissions in automobiles, autofocus lenses in cameras, and thermostats in heating/cooling systems. It’s why we have robot arms in manufacturing assembly plants. And it’s why we’re working toward automobiles that drive themselves, so we won’t even need to steer them to our destination.
Autonomous Body Parts
The ultimate in labor-saving extensions are autonomous robots. They do all the work (that they are designed to do), without any instruction beyond their initial, built-in programs. One of the first things we want them to do is to stop pestering us! A robot that can do the work of ten people at a tenth of the price is not such a big deal if it takes twenty people to maintain it.
To work autonomously does not necessarily mean that a robot must be detached from the rest of the world, but in many cases they are. Suppose you wanted (and don’t worry, they’re already working on it) a vacuum cleaner that roamed the house on its own. It could be connected to its energy source (the electrical supply) with a cord, as is usually the case today, but it would be more convenient if it were cordless. That means it must run on batteries, and that means the batteries will need charging—the robot will need to eat.
Scott Jantz and Keith L. Doty of the Machine Intelligence Laboratory at the University of Florida describe their experiments in getting a robot to learn to eat—to find a battery charger and plug in when their batteries run low. They endowed their robots with a "learning algorithm" (a program) that would let the robots figure out their own "eating habits." The robots, the researchers report, "learned to stay in the most efficient region of the batteries’ operation." In other words, they learned to eat wisely. Not so little that they would run out of steam in the middle of a job, and not so gluttonously that they would waste energy.
They also gave them a realistic environment in the sense that their resources (battery chargers) were not unlimited, and they had to compete for their dinners. "Those which [did] not compete successfully, [died] off."
But the ultimate goal of the research was not a feeding robot. They note that during the 1950s a vacuum tube robot "turtle" built by W. Grey Walter could recharge itself. Rather, their main goal was "to develop a robust robotic platform capable of surviving for an indefinite period of time without human intervention. Learning algorithms are implemented in the agents to facilitate the longevity of the robots and to imbue the robots with a primitive instinct for survival" (emphasis added).
Feeding Frenzy
If you find a little unsettling the notion of robots able to get long just fine without human help, thank you, then wait, there’s more. Professor Doty and colleague Ronald E. Van Aken in 1993 created a software simulation of a swarm of robots whose emergent behavior successfully performed the required activities without central planning. Swarm robots are "groups of autonomous mobile robots whose sensory-driven state behavior produces emergent group functionality not characteristic of the individual robots" (emphasis added).
The swarm robots had sensors to detect collisions with one another, with walls, and with other elements in the environment. "The sensors served as inputs to robot state machine controls and essentially allow the robot to adapt and interact with its environment without learning" (emphasis added). There was, however, a level of command and control, and we’ll get to that in a moment. Meanwhile, at the Johns Hopkins University Robotics Lab, they’re busy building robot swarms (though they don’t call them swarms) that can join together in various configurations, rather like "constructor" toys that can change shape and function—with a little help from the kids.
They are called metamorphic robots. "A Metamorphic Robotic System," says JHU’s Web site, "is a collection of independently controlled mechatronic modules, each of which has the ability to connect, disconnect and climb over adjacent modules. Each module allows power and information to flow through itself and to its neighbors. A change in manipulator morphology results from the locomotion of each module over its neighbors. Thus a metamorphic system has the ability to dynamically reconfigure."
Command and Control of Autonomous Body Parts
What we have so far, then, if we put it all together (and believe me, someone will) is a bunch of creepy but lovable little monsters able to do their thing, feed themselves, fight for survival, and metamorphose if needed. Their "thing" is to some extent preprogrammed (into the individual robots), but at a higher level various behaviors just emerge when they get together. This is not so far removed, in principle, from a description of an ant colony. Some ant species are known to clump together and "metamorphose" into a bridge to enable their colleagues to cross an ant ravine. Thankfully, neither ants nor our metamorphic swarm robots are very big, powerful, armed, or smart. Yet.
But as for smart, re-enter Professor Doty, this time with IBM researcher Akram Bou-Ghannam in tow. Together, they have worked on the control architecture for an intelligent, fully autonomous mobile robot.
Noting that "Animals live in a dynamic environment and tailor their actions based on their internal state and their perception of the external environment," and that "Animal interaction with the environment becomes more complex as one ascends the hierarchy of organisms," they conclude that "Animals lower in the hierarchy behave reactively to stimuli where those at the top end of the hierarchy employ learning and intelligence."
This is much in line with our discussion last week about emotion and reasoning, and Dr. Damasio’s categories of emotional decisions. Lower-order animals rely more on "basic" (instinctual, hard-wired, preprogrammed) decisions than on "acquired" decisions (basic decisions refined in response to experience), and even less (if at all) on "intellectual" decisions.
The question for Doty and Bou-Ghannam, then, was: Should robots be modeled on human behavior and intelligence that relies more on intellect, or on insect-like intelligence that relies on basic, instinctual behaviors? Their answer was to combine both, but since humans (as Damasio shows) already combine intellect with basic, instinctual intelligence, then to my mind they really opted for the human model.
If I have understood their research correctly, the three principle components of the Doty/Bou-Ghannam robot are: A behavior component, which is essentially a set of pre-programmed responses to stimuli—the "hard-wiring"; a perceptual component, which consists of sensors and programs that perform initial processing of sensory input and pass it to the cognitive component for analysis, integration with previous knowledge, and if necessary instructions to the behavioral component that controls the robot’s actions; and a cognitive component, which "manipulates perceptual knowledge representations" and "performs higher machine intelligence functions such as planning."
Through feedback and feedforward loops similar to those in the human body, the three components are constantly letting one another know what is going on in their neck of the woods. A key aspect of the processing is that it occurs in parallel. Parallel processing is well-known in the computing world as a means of gaining quantum jumps in processing speed over traditional serial processing, and it is the way the human body’s neural network operates.
Have They Created a Monster?
The Doty robots are not the most beautiful of creations, and they currently don’t do much at all beyond mooch around a lab. They are a bunch of wires, motors, computer chips, etc. sitting atop circular metal plates with wheels on the bottom. Philosophically, they may or may not turn out to be monsters when they grow up, but the chances are they won’t look like monsters. Chances are they may look remarkably like you and me, if they are designed as "general purpose" robots as opposed to special function robots.
Humanoid robots (or "androids," to use the science fiction term) as human-like in physical appearance as Commander Data in the Star Trek TV series may be closer than you think. Skin made up of fabric with thousands of tiny, touch-sensitive transducers woven into each square inch, will cover their limbs. Such fabric is already available, used for data gloves and data suits in virtual reality environments requiring tactile sensations. The limbs themselves will not be the mess of solenoids, rods, motors, nuts and bolts of today’s robots, but be made of materials much like human tissues. Take muscle, for example: "The Artificial Muscle Project at the MIT Artificial Intelligence Laboratory is an effort [to produce] linear actuators based on polymer hydrogel. These actuators have characteristics similar to human muscle, in that they are linear compliant elements which undergo reversible length changes in response to chemical stimuli." They’ve actually made this stuff work.
Also trying to make things work are the boffins at Los Alamos National Lab, birthplace of the atom bomb. Their BEAM (Biology, Electronics, Aesthetics, and Mechanics) Robotics project aims "to improve robo-genetic stock through stratified competition and have an interesting time in the process. The science behind the idea stems from current concepts in artificial intelligence (AI), artificial life (ALife), evolutionary biology, and genetic algorithms. It seems that building large complex robots hasn't worked well, so why not try to evolve them from a lesser to a greater ability as mother nature has done with biologics? The problem is that such a concept requires self-reproducing robots which won't be possible to build (if at all) for years to come. A solution, however, is to view a human being as a robot's way of making another robot, to have an annual venue where experimenters can let their creations interact in real situations, and then watch as machine evolution occurs.
In other words, robogenetics through robobiologics."
It seems to me a good concept, as an interim measure until folks like Professor Doty can improve the scale and sophistication of their progeny, and one that can only contribute to the eventual development of a fully evolved, intelligent, autonomous mobile robot.
What Next?
I hope I have not given the impression that Commander Data robots will be on the store shelves by Christmas. Hundreds of obstacles remain to be overcome, a handful of which (such as automatic speech recognition) we’ve already discussed. But there are tens of thousands of the world’s brightest men and women working to overcome the problems and not only are they getting smarter as they learn from their research, but so too are the tools—the computers—they are using in their research.
It would be wrong also to imply that the University of Florida alone has a handle on robots. Far from it—check out the robotics section in the resource area. But I do tend to think that the Doty et al. approach, while complex in its detail, is elegant in its form. And I don’t agree with everything they say, such as this quote from the Walking Robots Group "Robot Philosophy" page (http://www.mil.ufl.edu/groups/walking/res5.htm):
"In a few years, autonomous agent evolution will yield machines which can reproduce themselves. At that point, true "robot evolution" can start and the resulting explosion of agents will be unthinkable. Robots will become predators and prey and the weak and inefficient will be scrapped for raw materials.
Until that point, we are the robot gods. We play with them like the Greek gods of Mount Olympus, helping them along and "retiring" those that break down too often to fix or are just plain useless. We evolve them by making more of those that function well and scrapping those that don't. Let us not suffer the fate of the Greek gods who are said to have disappeared as their human subjects no longer needed their guidance." The unidentified megalomanic author of this passage adds: "Obsolescence should be a fate reserved for technology."
Spoken like a true evolutionary bigot! I wouldn’t want to be in your shoes when Machina sapiens wakes up and reads this. Particularly considering that you are surrounded by robots!
Until
next week,
NEXT WEEK: A Robot to Call Your Own. Commercially available robots. What they can do, where you can get them, how much they cost. More or less than you might think.
P.S. One of two especially nice things about the University of Florida Machine Intelligence Lab Web site is that its publications are posted in .doc format, readable by all major word processors. Most academic sites post documents in an infuriating Postscript (.ps) format, often compressed in an arcane .gz packet. Unix machines (and maybe Macs, I don’t know) can handle these formats with ease, but Windows users cannot. I found a shareware program that sometimes manages to display the files, but the display is almost illegible and very difficult to manipulate.
(Even .doc format could be improved upon: Why not post documents in html? There would then be no question that everyone could handle them.)
The other nice thing is that their publications are neatly formatted and well written; easy enough for the determined layperson to understand. That the folks in Florida should be taking such care and displaying such commonsense in their publications lends confidence that their research is similarly endowed.