17: Understanding Robots...
When you see robots in movies or on TV shows,
do you ever wonder: have we done that?
will we do that some
Article 17 provides perspective
for answering these questions.
Sci-fi buffs never tire of discussing
Isaac Asimov’s Three Laws for Robots. As these laws are intended to keep the machines from hurting us, a more apt title
would be Asimov’s Three Laws For Controlling Robots. With a bow to Asimov, I’d like to present my Three Laws for
Understanding Robots. Trust me, this’ll be fun.
By the way, if a machine can be made really smart, experts
speak of “artificial intelligence” or “AI.” If a machine moves around and interacts with the environment,
people speak of robots. Two sides, basically, of the same project. Okay, first the Three Laws and then we’ll see where
they take us.
LAW I: EVERYTHING THAT IS EASY FOR ROBOTS IS HARD FOR US. CONVERSELY, EVERYTHING THAT IS EASY FOR US IS HARD FOR THEM.
This law is not at all obvious and results in endless confusion. We see machines beating the best chess players. We know that
Google can search a billion web pages in a second. These feats are impossible for us, so we assume the machine is hugely superior.
Yes, but in an idiot savant kind of way.
Now make a list of things you could easily do when you were eight. Go to the fridge and get a Sprite. Walk into a room and
recognize all the faces. Cross a busy street or play hopscotch...No machine in the world can do these “easy” things.
Basically, the history of robotics for the past 40 years has been a painful confrontation with the implications of Law I.
The initial attitude--mid-twentieth century--was one of contempt for humans; robots would sweep us off the playing field.
The experts were as fooled as the ordinary person.
LAW II: FOR A ROBOT TO FUNCTION IN THE WORLD AS WELL AS WE DO, IT HAS TO KNOW AS MUCH ABOUT THE WORLD AS WE DO.
Here again, the experts got it wrong. Early theory was that you make a really smart computer, give it a lot of rules and
send it out into the world. Wrong, doesn’t work. The world is too chaotic and unpredictable.
While we’re children, we are learning a million things about the world. Imagine a robot who doesn’t know these
things trying to navigate from one house to another. Streets are full of odd things--gum, spit, feathers, leaves, scraps of
paper, coins, puddles and ballpoint pens--which ones must the robot worry about? Answer: all.
A child knows chewing gum deeply and personally. A robot can’t chew, can’t enjoy gum or spit it out. How do you
explain all these bits of trivia to this huge brain? But without this grounding, how can the robot make sensible deductions
about that lump on the sidewalk?
The experts began to figure this out around 1985, and a cloud descended over robotics. Practical people said, okay, we’ll
just accumulate a universal database of EVERYTHING the typical human knows. Think of the magnitude of this one project! Which
segues to the next Law....
LAW III: A HUMAN-LIKE ROBOT IS NOT ONE THING THAT ONE GENIUS WILL ONE DAY INVENT. A ROBOT IS A THOUSAND SUB-SYSTEMS THAT
WILL BE CREATED ONE BY ONE OVER DECADES AND CENTURIES.
There are two ways to judge a robot’s sophistication. How many tricks can it do? And in how large a setting or domain?
The first robots will do only a few tricks (e.g., vacuum the floor) and in a very limited domain (rooms it has memorized,
rooms with all hazards removed, or rooms with embedded wires that guide the machine). Graduating up to many tricks, and from
a limited domain out into the world, is the dream.
The human eye can distinguish millions of colors; we can pick out individual words in a noisy bar. We are analyzing-- consciously
and subconsciously--a vast array of information and memories as we walk, run and perhaps dance through the world. Adding any
one of our abilities to a robot’s repertoire is difficult. Adding all of them will be more difficult than building a
casino on Mars.
Right now, robotics is advancing steadily on five wide but distinct fronts: smart computers for our businesses; tireless robots
for our factories; artificial limbs and organs for our injured; sensors to detect chemicals and invisible dangers; and, finally,
“expert software” that enables a computer to do a single activity at a high level (e.g. play chess, diagnose illness,
find oil, or translate a language). This galaxy of projects is converging, for the most part coincidentally, toward the robots
imagined by science fiction. The Three Laws explain why the convergence is happening much more slowly than many predicted.
Where we are today:
The media often refer to radio-controlled devices (the Pentagon is developing a lot of these) as “robots.” At
most, these devices are rudimentary, entry-level robots. A machine becomes a robot to the degree that it can act and decide
on its own. At the other extreme, Hollywood movies show robots acting and deciding with abandon, like slightly odd humans.
It would be foolish to say such machines will never be built; so far, however, we have taken only baby steps.
Where we are going...
Kokoro meet-and-greet robot
But these ultimate robots, these human-like robots....will they really be human? That’s the Big Question and, intriguingly,
the AI community is split in half over this issue. Here’s my own guess. Yes, we’ll see astonishing, even unnerving
advances that will increasingly be part of our lives. But no, these things will never be human.
Reflect on what an exquisite meld we are of the physical, the mental, and the emotional. Simultaneously reason-based, feeling-based
and, some would say, spirit-based, we have been in development for millions of years and we are very clever. We can do long
lists of difficult physical things (ski, play pingpong) and subtle emotional things (love a painting, believe in a dream).
The key point here is that robots can learn to mimic only those activities that can be reduced to mathematics. Feelings will
be especially tough for robots to mimic, and tough for them to understand, feelings such as loneliness, fear, pity, boredom,
shame, guilt, and love. We feel panic to the point we sweat and throw up. No machine will ever throw up. We fear death--a
machine never will. Sure, you can program the machine to say it “fears” death, just as you can program it to say,
“I love you.” No machine will ever look up at the stars and feel awe. Robots will endlessly struggle with irony,
ambiguity and deceit--those open-ended, contradictory and sneaky things we humans are so extremely good at!
Consider what “expert software” is and isn’t. Programmers ask a hundred experts how they do what they do,
and the collective wisdom is distilled into a set of rules that let the machine reach smart conclusions. Similarly, nurse
robots, baby-sitter robots, guard robots, romantic robots and so on will be given the combined wisdom of people in those various
fields. But note something crucial. A “medical robot,” to take one example, never knows what disease or health
is, or what life and death is, or who you are. It solves a medical problem in the same way a chess machine solves chess problems--with
formulas and algorithms. Basically, it “looks up” the answer in a vast relational database. This machine could
be quite valuable without thinking or feeling anything.
Alan Turing, robotics pioneer, postulated that “artificial intelligence” would officially exist when a machine
could engage a human in conversation--that is, communicate so that the human never realizes the conversation partner is not
another person. Well, passing the Turing Test will be a big step. But that so-called Turing Machine is hardly human.
Perhaps the Turing Test is not so decisive as people think. Let me propose the Price Test: do they get the joke?? Imagine
a comic telling jokes to an audience of humans and humanoids. The humans--silly physico-emotive creatures that we are--will
reveal themselves by shaking with merriment while the robots sit there. My guess is that robots will never be able to understand
when a joke is funny. (Technically, it will be much easier to teach robots to recognize oncoming laughter in a human face,
so the robots can pretend to laugh at our jokes.) The very best robots, centuries from now, might end up like humans today
who suffer from a form of autism called Asperger Syndrome. These are highly intelligent, even accomplished people. But they
don’t understand humor. If you explain why the joke is funny, they still can’t get it. Robots won’t either.
In short, if a female robot has skin that feels “real,” is it therefore skin? If this robot has a sexy smile,
is she 1) sexy? 2) smiling? Or is “she” best described as an elaborate fake?
On the other hand, a good imitation is often going to be good enough! I have watched with fascination as every other Christmas,
toy companies bring forth another array of overpriced, almost useless “robots.” Two years ago, there were robot
dogs costing $30 that sang! (Not because dogs sing but because that’s a feature that could be added cheaply.) This past
Christmas the price range jumped up close to $100, with Robosapien costing $250. And what do they actually do? Not much. And
yet, and yet. Stop and consider how little we settle for. If your dog wags its tail, you’re hooked. If a parrot says
a few words, wow, a genius. Little girls play all day with dolls that don’t even move; and grown men cohabit with inflatable
dolls. I actually acquired--for research purposes, you understand--a little yellow bird that senses my movement and promptly
moves its wooden head and chirps; and I have actually inquired, “Hey, buddy, how’re you doing?” So, robots
don’t have to be like us; they merely have to be entertaining and useful to us.
There are already websites devoted to owning and interacting with robots. A boy in Cleveland rhapsodized, “I love Robosapien!!!
I would recommend this robot to anyone who feels they need a friend.” On another site, an adult (adult enough to have
a website) raves that the $100 Roboraptor “has a well balanced personality that changes intelligently....It’s
almost like having a live pet in your house.”
You can see the future for yourself. Robots will first enter our lives as pets, dolls, playmates, guards, nurses and domestics.
A gigantic vortex of technology, most of it developed for other purposes, is going to spiral down into our living rooms as
Wonder Dog, Baby Betty, Bob the Butler and Pretty Peggy! One can almost predict trouble ahead for the pet food industry, not
to mention Molly Maids, prostitution and disagreeable spouses. Not next year, mind you. But during the next twenty years this
first generation of robots is going to become quite sophisticated. In fifty years, all bets are off. Have you ever seen fake
flowers that seemed more real than real flowers? That’s the world we’re entering.
Many people suppose that evolution is something that happened long ago. Not so. Every species on the planet is evolving as
we speak. But now there is a new species, a form of quasi-life, and it is evolving fastest of all.
Background of this article: more than
20 years ago I read about a tradeshow robot that could move around and talk to people. Always a science buff, I was amazed.
Why hadn't I read about this before? Had we advanced this far?? Long story short: I got an assignment from a magazine to write
about this prodigy. I went to New Jersery and interviewed the owner (the robot greeted me), and then I called experts all
over the country. Little by little, it became clear that the robot had to be a fake. But what
stayed in my mind was how defenseless I (and most others) were before this thing. We just didn't have an intellectual framework
for understanding what it was doing (or not doing). The experts were/are remarkably divided about what's coming
in the future, and remarkably specialized so that it's difficult even for them to judge what some other inventor might have
done, or might do next year. Also, experts are usually trying to raise money--they have to dream big and promise big about
whatever their specialty is...In sum, robotics tends to be an unusually elusive field. So I had long wanted to write a sort
of Beginner's Guide to Robots.
UPDATE: SEPTEMBER, 2007: WIRED just
published a supplement called Geekipedia, which basically endorses everything said in this article. Entry titled "Artificial
Intelligence" notes: "Artificial intelligence is brain-dead."
UPDATE: JANUARY, 2010:
The annoying thing is the way AP (Associated Press) keeps sending out fluff pieces that make it seem that human-like robots
are everywhere, or soon will be. And my local paper keeps running this FLUFF. And the public be damned.
UPDATE: JANUARY, 2011: WIRED, current issue,
has again confirmed everything said in this article by discussing "....the seemingly endless decades-long quest to emulate
human intelligence. That goal proved so elusive that some scientists lost heart and many others lost funding. People talked
of an AI winter..." WIRED, somewhat like the new issue of POPULAR SCIENCE, then tries to find solace in the increasingly
complex weapons, robotics, and management systems we are seeing. But neither magazine is talking about human-like machines any time soon--which is the whole point of
my article. Note the phrase "so elusive." Exactly.
UPDATE: FEBRUARY, 2011: TIME rhapsodizes (in "2045: The Year Man
Becomes Immortal") about Kurweil, the Singularity, the end of aging, and how we'll put our psyches in a chip and live
there forever. A few good points but they are marred by the usual irresponsible blather, such as, "There are more
than 2,000 robots fighting in Afghanistan alongside the human troops." Now, you know any average person reading this
imagines robot-soldiers running wild on the battlefield. Nonsense. What is TIME talking about here? Drone planes? Remote-controlled
bomb-sniffing devices? The word "fighting" is the biggest lie. It suggests active thought and will, as in the sentence.
"The men are fighting the enemy in the mountains." Was TIME always so addled?
The points made in this article
are presented in a short graphic video called "Stop The Hype About Robots" on YouTube.
Click this link:
STOP THE HYPE ABOUT ROBOTS
© Bruce Deitrick Price 2006-11
Price's Three Laws for Understanding Robots