Rosie, C-3PO, HAL, WALL-E, Bender, Optimus, Asimo. We have envisioned
building humanoid machines in our likeness with the ability to walk,
talk and think like us — or better than us — since long before the
creation of the microchip.
In 322 BC, Aristotle postulated that automatons created from garden
statues might bring about the end of slavery. Two-year-olds are
captivated by these mostly mythical creatures we call robots. From the
time we are little kids, when we dream about the future, we dream about
robots. It’s as simple as that.
But here’s the thing — humanoid robots are never going to happen. To
understand why, you have to understand the shape of the technical
problems and compare them to market forces.
First, the mechanics of the human body are spectacularly complex.
Depending on your view of creationism and evolution, the human design
either required the hand of God or it was the output of a one in a
gazillion experiment. Either way, we are dealing with a mechanical form
that isn’t easy to replicate. Sure, we have built machines with arms and
legs, but we are a long way from having a unified platform that can
walk the Earth with the dexterity and reliability of a human.
Second, we need major advancements in input/output software. Anyone
who’s gotten into an argument with Siri can well understand how far we
are from having a dependable conversation with a machine. It’s amusing,
if frustrating, to ask Siri to “Send Jenn a text” and hear her reply
“What song would you like me to play?” You can imagine what might happen
when your robot misunderstands you when you say “Take the trash to the
curb.” Speaker-independent, open-ended speech recognition with better
than 99 percent reliability isn’t part of our near future.
Third, artificial intelligence is the biggest software challenge in
the universe. The human brain’s ability to process information and learn
and make decisions is something we can barely understand, let alone
reproduce. Replicating human thought in a bounded problem — like the
game of chess — is barely within our grasp. Software that enables
machines to act like us, which requires them to think like us, isn’t
something we’ll be able to develop for generations, if ever.
So all we have to do to build a humanoid robot is build a machine to
replicate the human form. Then solve the input/output problem so we
always understand each other. And finally, replicate the thought process
of the human brain. Any one of those three tasks is nearly impossible.
The intersection of all three represents one of the greatest technical
challenges known to man.
However, it’s not just the scale of the technical challenge that
makes a humanoid robotic future impossible; it’s the intersection of
those three challenges with simple economics.
A humanoid robot would be a multipurpose platform that could do
almost anything. It could do things like get a beer from the fridge or a
newspaper off the front lawn. But those are sub $20 problems — meaning
realistically you wouldn’t pay $20 for someone to solve those problems
for you. That same humanoid robot could do more useful things like
vacuum your house or soothe your baby or drive your car. Those things
are worth more than $20 and we already use robots to solve those
problems – they just aren’t humanoid.
You can pay $400 to have a robot vacuum your house. For around $250
you can buy an infant seat to soothe your baby. When buying a new car
you can select robotic options on the fringe of self-driving like lane
departure warning and collision avoidance and dynamic cruise control.
It’s easy to see that fully autonomous self-driving cars are on the
horizon; it’s just a question of how far off. The drones used by the
military are another example of non-humanoid robotics.
The point is, we don’t need to build a humanoid robot to do these things. We can build the robots into the things themselves.
The computer industry offers an interesting analogy. For decades,
computers were multipurpose platforms that got faster and cheaper but
otherwise did not change much. However, in the early 2000s, for the
first time in history, a few cents’ worth of computing power was really
powerful. You could call it the second coming of Moore’s Law. That gave
us the ability to start integrating intelligence into devices at a rapid
rate, and it reduced the need for super powerful, multipurpose
platforms like PCs.
Today, tablets and smartphones do much of the work that our PCs used
to do and we have loads of new devices, such as DVRs, fitness trackers
and intelligent thermostats, that are essentially special-purpose
computers. Because computer technology got so powerful and inexpensive,
the answer wasn’t to keep extending the reach of the multipurpose PC. We
added the intelligence of the PC into the devices around us – a march
that is going to continue indefinitely.
The idea that we’ll have robots in the future to assist in daily
tasks isn’t wrong. The idea that we need a multipurpose humanoid
platform is. As costs continue to decline and technology continues to
advance, we’ll be able to make lots of special-purpose robots to solve
real consumer problems. Unfortunately for Aristotle, and the little kid
in all of us, there’s just no need to make them humanoid.
