“Although I do firmly believe that the brain is a machine, whether this machine is a computer is another question” —Rodney Brooks 70 NA • iEEE Sp iEEE • NA e ctrum • ju • ctrum ne 2008 www.spectrum.ieee.org

photo credit www.spectrum.ieee.org ju ne 2008 • • 2008 I EEE Sp EEE e ctrum • NA NA • ctrum 71 70 NA • iEEE Sp iEEE • NA e ctrum • ju • ctrum ne 2008 www.spectrum.ieee.org

photo credit

mages I etty G resnick/ seth page: this design; christie bryan spread: previous www.spectrum.ieee.org Am aR I B pow A from from it’ I then in principle there’s no reason why we shouldn’t shouldn’t we why reason no there’s principle in then biga is this intelligence, emotions, and even consciousness. remarkable possibility. If we really are machines and if— feelings, our intellect, our to rise give heads our inside notcompletely known, derivingrules fromphysics and molecules that interact according to well-defined, though I be able to replicate those rules in, say, steel.in, and silicon rules ablereplicateto those be our sense of self. place taking interactions biomolecular The chemistry.

believe our creation would exhibit genuine human-level y y , R s am a machine. machine. a am I’ve held during my 30-year career, this one in particu lar has been central to my research in robotics and and robotics in research my to central been has lar

artificial intelligence. artificial I, family, you, our and friends, dogs—we all are machines. We are really sophisti really are We machines. are all dogs—we cated machines made up of billions and billions of bio R

a od lr a e ne

rful rful ea s udd y y if dy dy odney Brooks, —wegoverning rules the learn ourbrains, B rook a e en ¶ rtifici volvi Accepting this hypothesis opens up a t So are you. are So e s ch n a g g n l i l ologic s n ymbiotic t obot ¶ e Of all the hypotheses hypotheses the all Of llig a l “big b “big l en a c lly with u llywith e wo an ju n g”— ne ’t ’t 2008 • • 2008 s

s pri ­ ­ -

I EEE Sp EEE n g

e ctrum • NA NA • ctrum

71

port e r l a ci e p s | y rit a gul n i s e th SpEciAl rEport |

SOCIABLE MACHINES: Founded by rodney Brooks, mit’s Humanoid robotics group develops robots capable of interacting and cooperating with people. Aaron edsinger built Domo [left] to explore dexterous manipulation and visual perception. mertz [right], created

thE SiNgulArity by Lijin Aryananda, is a robotic head able to learn from its environment. cynthia Breazeal designed Kismet [below] to study human- robot social interactions. photos, from left: aaron edsinger; peter menZel/ photo researchers; liJin aryananda

I’m far from alone in my the rates of progress in various tech- conviction that one day we will nologies and how and why those rates create a human-level artificial of progress are changing. Their argu- intelligence, often called an ments are plausible, but plausibility is artifi cial general intelligence, or by no means certainty. AGI. But how and when we will My own view is that things will get there, and what will happen unfold very differently. I do not after we do, are now the subjects claim that any specific assumption of fierce debate in my circles. or extrapolation of theirs is faulty. Some researchers believe that Rather, I argue that an artifi cial intel- AGIs will undergo a positive- ligence could evolve in a much diff er- feedback self-enhancement until their cal augmentation of the brain, genetic ent way. In particular, I don’t think there comprehension of the universe far sur- engineering, [and] ultrahigh-resolution is going to be one single sudden techno- passes our own. Our world, those indi- scans of the brain followed by computer logical “big bang” that springs an AGI into viduals say, will change in unfathomable emulation” are some of their ideas. They “life.” Starting with the mildly intelligent ways after such superhuman intelligence don’t believe this is centuries away; they systems we have today, machines will comes into existence, an event they refer think it will happen sometime in the next become gradually more intelligent, genera- to as the singularity. two or three decades. tion by generation. The singularity will be Perhaps the best known of the peo- What will the world look like then? a period, not an event. ple proselytizing for this singularity— Some singularitarians believe our world This period will encompass a time let’s call them singularitarians—are aco- will become a kind of techno-utopia, with when we will invent, perfect, and deploy, lytes of Raymond Kurzweil, author of humans downloading their conscious- in fi ts and starts, ever more capable sys- The Singularity Is Near: When Humans nesses into machines to live a disembodied, tems, driven not by the imperative of the Transcend Biology (Viking, 2005) and after-death life. Others, however, antici- singularity itself but by the usual eco- board member of the Singularity Institute pate a kind of technodamnation in which nomic and sociological forces. Eventually, for Artificial Intelligence, in Palo Alto, intelligent machines will be in conflict we will create truly artificial intelli- Calif. Kurzweil and his colleagues believe with humans, maybe waging war against gences, with cognition and conscious- that this super AGI will be created either us. The proponents of the singularity are ness recognizably similar to our own. I through ever-faster advances in artifi cial technologically astute and as a rule do not have no idea how, exactly, this creation intelligence or by more biological means— appeal to technologies that would violate will come about. I also don’t know when “direct brain-computer interfaces, biologi- the laws of physics. They well understand it will happen, although I strongly sus-

72 NA • iEEE SpEctrum • juNE 2008 www.spectrum.ieee.org www.spectrum.ieee.org juNE 2008 • iEEE SpEctrum • NA 73 pect it won’t happen before 2030, the engage in a dialogue using complete years ago someone asked me at a talk I year that some singularitarians predict. clauses and can handle irregularities, was giving, “Isn’t the brain just like the But I expect the AGIs of the future— idiomatic expressions, a vast array of World Wide Web?” embodied, for example, as robots accents, noisy environments, incom- We use these metaphors as the basis that will roam our homes and work- plete utterances, and interjections, and for our philosophical thinking and even places—to emerge gradually and sym- they can even correct nonnative speak- let them pervade our understanding of biotically with our society. At the same ers, inferring what was really meant in what the brain truly does. None of our time, we humans will transform our- an ungrammatical utterance and refor- past metaphors for the brain has stood selves. We will incorporate a wide range matting it. Most of these capabilities are the test of time, and there is no reason of advanced sensory devices and pros- still hard or impossible for computers. to expect that the equivalence of current thetics to enhance our bodies. As our n The manual dexterity of a 6‑year‑old digital computing and the brain will machines become more like us, we will child. At 6 years old, children can grasp ­survive. What we might need is a new become more like them. objects they have not seen before; manip- conceptual framework: new ways of And I’m an optimist. I believe we will ulate flexible objects in tasks like tying sorting out and piecing together the bits all get along. shoelaces; pick up flat, thin objects like of knowledge we have about the brain. playing cards or pieces of paper from Creating a machine capable of effec- ike many AI researchers, I’ve a tabletop; and manipulate unknown tively performing the four capabilities always dreamed of building objects in their pockets or in a bag into above may take 10 years, or it may take the ultimate intelligence. As a which they can’t see. Today’s robots can 100. I really don’t know. In 1966, some longtime fan of Star Trek, I have at most do any one of these things for AI pioneers at MIT thought it would wanted to build Commander some very particular object. take three months—basically an under­ LData, a fully autonomous robot that we n The social understanding of an graduate student working during the could work with as equals. Over the 8‑year-old child. By the age of 8, a child summer—to completely solve the prob- past 50 years, the field of artificial intel- can understand the difference between lem of object recognition. The student ligence has made tremendous prog- what he or she knows about a situation failed. So did I in my Ph.D. project ress. Today you can find AI-based capa- and what another person could have 15 years later. Maybe the field of AI will bilities in things as varied as Internet observed and therefore could know. The need several Einsteins to bring us closer search engines, voice-recognition soft- child has what is called a “theory of the to ultraintelligent machines. If you are ware, adaptive fuel-injection modules, mind” of the other person. For exam- one, get to work on your doctorate now. and stock-trading applications. But you ple, suppose a child sees her mother can’t engage in an interesting heart-to- placing a chocolate bar inside a drawer. grew up in a town in South Australia power-source talk with any of them. The mother walks away, and the child’s without much technology. In the late We have many very hard problems to brother comes and takes the chocolate. 1960s, as a teenager, I saw 2001: A Space solve before we can build anything that The child knows that in her mother’s Odyssey, and it was a revelation. Like might qualify as an AGI. Many problems mind the chocolate is still in the drawer. millions of others, I was enthralled by have become easier as computer power This ability requires a level of perception I the soft-spoken computer villain HAL has reliably increased on its exponential across many domains that no AI system 9000 and wondered if we could one and seemingly inexorable merry way. has at the moment. day get to that level of artificial intelli- But we also need fundamental break- But even if we solve these four prob- gence. Today I believe the answer is yes. throughs, which don’t follow a schedule. lems using computers, I can’t help won- Nevertheless, in hindsight, I believe that To appreciate the challenges ahead dering: What if there are some essential HAL was missing a fundamental compo- of us, first consider four basic capabili- aspects of intelligence that we still do not nent: a body. ties that any true AGI would have to pos- understand and that do not lend them- My early work on robotic insects sess. I believe such capabilities are funda- selves to computation? To a large extent showed me the importance of coupling mental to our future work toward an AGI we have all become computational bigots, AI systems to bodies. I spent a lot of time because they might have been the founda- believers that any problem can be solved observing how those creatures crawled tion for the emergence, through an evolu- with enough computing power. Although their way through complex obstacle tionary process, of higher levels of intelli- I do firmly believe that the brain is a courses, their gaits emerging from the gence in human beings. I’ll describe them machine, whether this machine is a com- interaction of their simple leg-control in terms of what children can do. puter is another question. programs and the environment itself. n The object-recognition capabili- I recall that in centuries past the brain After a decade building such insectoids, ties of a 2-year-old child. A 2-year-old was considered a hydrodynamic machine. I decided to skip robotic lizards and can observe a variety of objects of some René Descartes could not believe that cats and monkeys and jump straight to type—different kinds of shoes, say— flowing liquids could produce thought, humanoids, to see what I could do there. and successfully categorize them as so he came up with a mind-body dual- My students and I have learned a shoes, even if he or she has never seen ism, insisting that mental phenomena lot simply by putting people in front of soccer cleats or suede oxfords. Today’s were nonphysical. When I was a child, a robot and asking them to talk to the best computer vision systems still make the prevailing view was that the brain machine. One of the most surprising mistakes—both false positives and false was a kind of telephone-switching net- things we’ve observed is that if a robot negatives—that no child makes. work. When I was a teenager, it became has a humanlike body, people will inter- n The language capabilities of a an electronic computer, and later, a mas- act with it in a humanlike way. That’s one 4‑year-old child. By age 4, children can sively parallel digital computer. A few of the reasons I came to believe that to

72 NA • iEEE Spectrum • june 2008 www.spectrum.ieee.org www.spectrum.ieee.org june 2008 • IEEE Spectrum • NA 73 port e l r a ci e p s

| rit y a gul n i s e th

ANDROIDS ARISING: Clockwise from bottom left: Honda’s Asimo can dance and climb stairs and has even conducted an orchestra; the Actroid female humanoid was developed by the Japanese firm Kokoro andO saka University; Chinese roboticist Zou Ren Ti, of the Xi’an Chaoren Sculpture Research Institute, sits next to his trumpet and the violin; Osaka android twin [right]; Anybots, in University researchers built the CB2 Mountain View, Calif., is developing robot to mimic the appearance tele-operated mechanical and behavior of a toddler. servants; Toyota’s Partner robots photos, clockwise from bottom left: Honda; Yves Gellie/Corbis; Indiana University; Anybots; include little droids that play the Toyota; Androniki Christodoulou/WPN

build an AGI—and its predecessors—we’ll perform chores like taking out the gar- emotional behavior can be seen as the need to give them a physical constitution. bage, cleaning the bathtub, and carry- real thing. We are made of biomolecules; At this point, I can guess what you’re ing groceries. the robots will be made of something wondering. What will AGIs look like and Will they have complex emotions, per- else. Ultimately, the emotions created in when will they be here? What will it be like sonalities, desires, and dreams? Some each medium will be indistinguishable. to interact with them? Will they be socia- will, some won’t. Emotions wouldn’t be In fact, one of my dreams is to develop a ble, fun to be around? much of an asset for a bathtub-cleaning robot that people feel bad about switch- I believe robots will have myriad robot. But if the robot is reminding me ing off, as if they were extinguishing a life. sizes and shapes. Many will continue to to take my meds or helping me put the As I wrote in my book Flesh and Machines be simply boxes on wheels. But I don’t ­groceries away, I will want a little more (Pantheon, 2002), “We had better be care- see why, by the middle of this century, personal interaction, with the sort of feed- ful just what we build, because we might we shouldn’t have humanoid robots with back that lets me know not just whether end up liking them, and then we will be agile legs and dexterous arms and hands. it’s understanding me but how it’s under- morally responsible for their well-being. You won’t have to read a manual or enter standing me. So I believe the AGIs of the Sort of like children.” commands in C++ to operate them. You future will not only be able to act intelli- will just speak to them, tell them what to gently but also convey emotions, inten- any of the advocates of do. They will wander around our homes, tions, and free will. the singularity appear to offices, and factories, performing certain So now the big question is: Will those the more sober observers tasks as if they were people. Our envi- emotions be real or just a very sophisti- of technology to have a ronments were designed and built for cated simulation? Will they be the same ­messianic fervor about our bodies, so it will be natural to have kind of stuff as our own emotions? All I M their predictions, an unshakable faith these human-shaped robots around to can give you is my hypothesis: the robot’s in the certainty of their predicted future.

74 NA • iEEE Spectrum • june 2008 www.spectrum.ieee.org www.spectrum.ieee.org june 2008 • IEEE Spectrum • NA 75 To an outsider, a lot of their convictions to change ourselves from purely biologi­ seem to have many commonalities with cal entities into mixtures of biology and religious beliefs. Many singularitarians technology. My prediction is that we are believe people will conquer death by more likely to see a merger of ourselves downloading their consciousnesses and our robots before we see a stand- into machines before their bodies give alone superhuman intelligence. out. They expect this option will become Our merger with machines is already available, conveniently enough, within happening. We replace hips and other their own lifetimes. parts of our bodies with titanium and But for the sake of argument, let’s steel parts. More than 50 000 people have Expert View: accept all the wildest hopes of the singu- tiny computers surgically implanted in Esther Dyson laritarians and accept that we will some- their heads with direct neural connec- WHO ShE IS: how construct an AGI in the next three or tions to their cochleas to enable them to Commentator and four decades. My view is that we will not hear. In the testing stage, there are retina evangelist for emerging live in the techno-utopia future that is so microchips to restore vision and motor technologies, investor and board member for fervently hoped for. There are many pos- implants to give quadriplegics the abil- start‑ups; currently sible alternative futures that fit within the ity to control computers with thought. focused on health care, genetics, private aviation, themes of the singularity but are very dif- Robotic prosthetic legs, arms, and hands and commercial space. ferent in their outcomes. are becoming more sophisticated. I don’t Ran PC Forum conference One scenario often considered by think I’ll live long enough to get a wire- until 2007; currently hosts the annual Flight School singularitarians, and Hollywood, too, less Internet brain implant, but my kids conference. is that an AGI might emerge spontane- or their kids might. ously on a large computer network. But And then there are other things still THOUGHTS “The singularity I’m perhaps such an AGI won’t have quite further out, such as drugs and genetic interested in will come the relationship with humans that the and neural therapies to enhance our from biology rather than machines. We won’t be singularitarians expect. The AGI may senses and strength. While we become building things; we’ll be not know about us, and we may not more robotic, our robots will become growing and cultivating know about it. more biological, with parts made of arti- them, and then they will grow on their own.” In fact, maybe some kind of AGI ficial and yet organic materials. In the already exists on the Google servers, future, we might share some parts with probably the single biggest network of our robots. computers on our planet, and we aren’t We need not fear our machines aware of it. So at the 2007 Singularity because we, as human-machines, will Summit, I asked Peter Norvig, Google’s always be a step ahead of them, the chief scientist, if the company had noticed machine-machines, because we will any unexpected emergent properties in adopt the new technologies used to build its network—not full-blown intelligence, those machines right into our own heads but any unexpected emergent property. and bodies. We’re going to build our He replied that they had not seen any- robots incrementally, one after the other, thing like that. I suspect we are a long, and we’re going to decide the things long way from consciousness unexpect- we like having in our robots—humility, edly showing up in the Google network. empathy, and patience—and things we (Unless it is already there and cleverly don’t, like megalomania, unrestrained concealing its tracks!) ambition, and arrogance. By being care- Here’s another scenario: the AGI ful about what we instill in our machines, might know about us and we know we simply won’t create the specific con- about it, but it might not care about us at ditions necessary for a runaway, self- all. Think of chipmunks. You see them ­perpetuating artificial-intelligence explo- wandering around your garden as you sion that runs beyond our control and look out the window at breakfast, but leaves us in the dust. you certainly do not know them as indi- When we look back at what we are viduals and probably do not give much calling the singularity, we will see it not thought to which ones survive the win- as a singular event but as an extended ter. To an AGI, we may be nothing more transformation. The singularity will be than chipmunks. a period in which a collection of technolo­ From there it’s only a short step to the gies will emerge, mature, and enter our question I’m asked over and over again: environments and bodies. There will be Will machines become smarter than us a brave new world of augmented people, and decide to take over? which will help us prepare for a brave new R ick ick I don’t think so. To begin with, there world of AGIs. We will still have our emo- S molan will be no “us” for them to take over from. tions, intelligence, and consciousness. We, human beings, are already starting And the machines will have them too. o

74 NA • iEEE Spectrum • june 2008 www.spectrum.ieee.org www.spectrum.ieee.org june 2008 • IEEE Spectrum • NA 75