While these projects are important steps in the right direction, functional results come slowly. Like the human infants they model, contemporary humanoids are inefficient at most tasks and require intensive training. One of the implications of this research is that to create humanlike adaptability and versatility, it may be necessary to introduce an element of human frailty and inconsistency. We have already manufactured machines with the ability to perform a task in exactly the same way, time after time. While such behavior can be useful for some tasks, it is brittle and will fail as soon as the wind changes. The future will bring humanoids designed to take part in the drama of chaos, inconsistency and error we know fondly as the real world. Such humanoids will not be hindered by complexity and complication, but will embrace it and thrive on it.
Lifelike 'martial arts expert' head made by YFX.
Many of the traits we consider uniquely human stem not from our strength, reliability or the precision with which we execute tasks. In fact, we do quite poorly in these arenas. This is not coincidence, but a vestige of our adaptability and ingenuity. Optimality brings stasis and hinders versatility. It is a concept that has little to do with the flux of change in our real world. Unlike classroom computer science, the algorithms of human intelligence are neither provable nor constant. When we move from algorithms and virtual worlds into the real-world arena of noisy sensors, humidity and slippery floors, the notion of one perfect solution will give way to the more compelling possibilities of flexibility and adaptation.
Perhaps the most "human" thing of all is our amazing ability to be error-prone and inconsistent and yet cope. This ability derives, at least in part, from our ability to recognize imperfection and even exploit it, using the arbitrary fluctuations in ourselves and our environment to drive learning, creativity, humor and inventiveness. Humanoids will not encroach on these human attributes, but rather bolster them, allowing us to move further toward the heart of what it means to be human.
Interior mechatronics of an animatronic head made by Fumio Hara at the University of Tokyo.
As humanoids begin to tap the random variation of the external world and transmute entropy into creativity, philosophers will argue indefinitely about whether they have 'true' intelligence, will, or emotion. Regardless, the majority of us will project these traits into them anyway. These humanoids will not necessarily look like Robby or C3PO. YFX studios has designed an amazingly lifelike animatronic head. We will watch as humanoids do things our flesh bodies do not allow and will observe their power run out and see their bodies malfunction. We will know perfectly well that they are not human and yet this fact will not preclude emotional involvement. Even those who consider humanoids to be mere machines, unworthy of emotional interchange, will find themselves responding differently to one humanoid than another. Compare the YFX martial arts master with the picture of the interior mechatronics of a robot head made by Fumio Hara at the Science University of Tokyo. Regardless of whether either has an explicit emotional effect on the viewer, the viewer will invariably be affected by the visual context of face. Humans are so well versed in the art of lopsided communication, we can pour affection on a virtual pet key chain, much less a humanoid that makes us laugh, and listens to us when no one else will.
Humans have always been eager to project emotion into machines. Our imaginations seem to have little trouble rising to the task. Recently, Hasbro has enlisted the help of top roboticists at iRobot to develop robot infants that can be sold as toys. Complete with numerous servo motors and sensors, these inexpensive robots can simulate body functions such as feeding and digestion, interact with humans and even simulate real infant language learning, progressing from proto-language babble to the use of real words. The dolls are able to move their limbs and facial degrees of freedom in realistic ways to communicate a wide spectrum of desires and emotions in response to human attention. Fortunately for the designers, modeling realistic infant behavior is a decidedly easier task than adult capabilities and already includes an element of awkward imprecision. While many attempts to create adult humanoids have floundered, infant behavior may not be such a long shot. It seems that the first kind of humanoids to enter our homes will be animatronic infants.
'Robota' dolls developed by Aude Billard at USC.
As the number of humanoids increases, the collective population of humanoids will learn, develop and perhaps eventually reproduce themselves more effectively. Unlike cars or televisions that improve along a linear, highly controlled trajectory, humanoids will be the ultimate in self-accelerating technology. Moore's law holds at least in part because we can use new chips to design and engineer their successors. Likewise, robotics is a self-enabling technology. Robotic tools will make the humanoids we ourselves could never make. Once we have a large population of self-motivated agents attending to separate tasks, these agents will negotiate, exchanging tasks and resources in mutually beneficial ways. Humanoids will comprise a new distributed infrastructure comprised not only of information, but real-world action. As a given task arises, humanoids will place bids, often partnering with other humanoids to get the job done. Humanoids will not only share workload and resources, but will also evolve by passing host-independent, modular code. Already, researchers have achieved surprising success with co-evolution, a form of online, collective learning that distributes strategies across a population of robots and then allows individual knowledge and skill to be shared according to a measure of each individual's "fitness."
As robots become more pervasive, they will, like automobiles, become increasingly complex. Already, some robots are comprised of millions of parts. Those skeptical of humanoid research often point to the high price tags of today's humanoids. If fast, cheap, rapid manufacture of robots is to occur, it will be necessary to remove humans from the design and manufacturing process. To do so, Brandeis is using commercial CADCAM simulators together with a genetic algorithm to evolve the body and brains of simple robots. Their daring project has created an almost fully automated system that can design, successively improve and then actually make a real robot all with only trivial human intervention. The initial design space includes linear actuators, bars, ball joints and neurons that comprise a rudimentary nervous system. Through mutation and recombination, the genetic algorithm might modify bar length, split bars, or connect neurons to various components as it propels generations of increasingly fit robots. Finally, the robots are fabricated automatically by a machine that prints the robots, layer by layer, out of plastic.
So far, work has focused on creating robot life forms capable of locomotion. Eventually, this learning approach will allow near-perfect solutions to be evolved even for specialized tasks requiring unintuitive approaches. Since the genetic algorithm can search the so-called "irrational search space" of possible designs, most of the Genetically Organized Lifelike Electro-Mechanics (GOLEMs) produced are nothing like their human-designed counterparts. Remarkably, however, recognizable shapes have emerged including creatures affectionately named "crab" and "snake."
'Baby' robot developed at iRobot.
The ability for a machine to design and manufacture another machine is no joke. In regards to the notion of artificial life, we are faced with the prospect of something that seems to echo (albeit a distant echo) the phenomenon of biological reproduction. Even if humanoids do not reproduce themselves, imagine the utility of a humanoid that can evolve and fabricate its own tools on the fly to be perfectly suited for the task at hand. On the other hand, there are currently severe limitations to the strategy employed at Brandeis. A fabrication machine that can handle larger, more complex robots may be, for a time, prohibitively expensive. Also, more complex behavior will require more accurate real-world modeling and richer simulation. The increased demand in computational power and time may limit the complexity of robots that can be evolved.
Although the application of Moore's law to silicon microchips will taper off some time around 2020, there are possibilities on the horizon for optical computers that switch photons instead of electrons; DNA computers that use sequences of bases to encode information; molecular computers that use molecules as logic gates and quantum computers that encode information by manipulating the rotation of individual atoms. Some who champion the potential of nanobot technology are convinced that within a century we will use molecule-sized robots to infiltrate and understand every pertinent function of the brain well enough to replicate each in an artificial medium.7 It is quite possible that such predictions overestimate our skill while at the same time underestimating the organ that produces it. On the other hand, the computational power of computers is increasing exponentially while the human brain is fixed at around 100 trillion connections. The somewhat shocking realization is that if someday AI can reach a level comparable to human intelligence, there is no reason why it will not continue to sail past it.
An automatically generated robot designed and fabricated without human intervention by a machine at Brandeis University
While we cannot ignore this theoretical possibility, such predictions fail to realize that humans and humanoids may evolve along separate trajectories such that it may never make sense to equate human and humanoid intelligence. Most likely, humans and humanoids will continue to be good at different things. Digital computers can already operate more than 10 million times faster than the electrochemical processes in our brains. The structure of the brain, however, more than makes up for this ostensible shortcoming. The power of the human brain arises not from the sheer number of bytes it can store or from the speed with which its electrochemical processes operate. Unlike the linear, digital computer, the brain is a massively distributed, parallel processor where information is stored implicitly as associations between neurons. Thus, the fact that computers are moving quickly toward faster and larger computation does not necessarily mean they are moving closer toward human intelligence. The brain's capacity to not only learn, but also be aware of and able to direct its own learning, may result from recursive mappings between local collectives of neurons. Recent research indicates that this phenomenon of re-entry provides the brain with optimal complexity a perfect balance between the tractability of a highly integrated, well-ordered structure and the infinite potential of desultory variation.
As computing does become faster and more pervasive, it remains to be seen whether humanoids can become crucial arbiters of the emerging New World. Some have argued that distributed computing will sweep away the need for humanoid robots. According to this reasoning, computing will not need to be centralized in a single sophisticated agent, but will rather reside throughout the environment in every object around us. No doubt the role of humanoids will evolve alongside the changing lives of the humans they serve. Most likely, these changes will accentuate, rather than remove the need for intelligent agents that can mediate between humans and the increasingly complex, technological world we are creating.
- David Bruemmer, Send E-mail