Neil Blomkamp's new film CHAPPiE, which hits US theaters this weekend, follows the unlikely transformation of a defective robot into a one-of-a-kind conscious machine. The movie inserts the audience into a Johannesburg, South Africa, that's protected by a fully robotic police force. The brilliant designer of these bots, Deon Wilson (played by Dev Patel), isn't quite satisfied with soulless automatons--so he secretly works after-hours to instill his creations with consciousness.
Thus CHAPPiE, the first thinking robot, is born shortly after being slated for destruction. The machine begins its new life awed by what we'd normally find mundane. He (it?) soon begins to understand more about the humans who shelter him and alters his way of thinking based on his artificially rapid upbringing.
Blomkamp's flick ultimately rejects the warnings of bright minds such as Stephen Hawking and Elon Musk and makes an optimistic statement about the future of artificial intelligence. If a robot starts with a blank slate, yet is given consciousness to help it learn and grow, Blomkamp says it would be a benevolent bot--a being interested in academic and creative pursuits, and not some SkyNet-like apocalypse seed. Such a robot would also trust the humans that created it, emulate them, and even love those who show it friendship.
This is a rosy and distant take from the usual ways moviemakers present A.I.--all-powerful, robotic beings bent on destroying or enslaving humans.
If we ever manage to create conscious robots, what type of system can we expect? A robot determined to rid the Earth of its pesky human population? Or one that'd graciously integrate into our society?
We spoke with Dr. Wolfgang Fink, an A.I. researcher at the California Institute of Technology, for some answers. He says the bigger questions will go unanswered for quite some time--we're nowhere close to creating the robot envisioned in CHAPPiE. According to Fink, who consulted on the film, researchers are still missing the biggest piece of the A.I. puzzle: how to make a system that is both self-aware and environmentally aware. When those traits can be programmed, then we'll have a truly autonomous system--and Fink says it will look like nothing we've ever seen before.
What is A.I.?
Fink says three different levels of robotic intelligence exist. The first is human-controlled, which is the dynamic behind most robots today. In CHAPPiE, for example, Vincent Moore (played by Hugh Jackman) dons a sensor-embedded helmet to control a giant military robot called MOOSE. “It's nothing but a mechanical extension or a tele-presence of the human,” Fink explains. “The intelligence is instilled by the human.”
The next tier is what Fink actually considers to be A.I.: a machine largely guided by a pre-installed, rule-based system. Equipped with a list of situations the system might encounter, the robot can react (appropriately or disastrously) to a variety of pre-programmed scenarios. CHAPPiE plays up this level of intelligence with Johannesburg's fictional, robot-only police force. The bot cops follow a set of rules defining how a law enforcement officer should behave to keep citizens safe; there's no real improvisation.
The hurdle with this type of intelligence--if you're striving for higher machinery, anyway--is that it lacks the fundamental aspects that make beings sentient: It's not truly aware of itself and its surrounding environment.
“If there's a situation where someone draws a gun, then [the robot] will draw a gun,” says Fink, referring to the police bots. “But what happens when the situation changes? If civilians run onto the scene, they might not know what to do and harm an innocent person in the process. They're not self-aware or situation-aware.”
Creating a robot with both flavors of awareness would raise the bar to the third level of robotic intelligence: true autonomy. According to Fink, that would signify the first artificial consciousness. Like CHAPPiE, such a robot would be capable of modifying its thinking and personality, taking cues from its environment and its interactions with humans. And most of all, it would understand that it is a thinking being.
“It makes that statement: 'I am CHAPPiE,'” says Fink. “That's a profound statement because it symbolizes self awareness, and that's something nobody has been able to figure out.”
From intelligent to autonomous
In all likelihood, the “brain” of an autonomous system will look like no other structure yet conceived by robotics or biology. To make it a reality, Fink says we should look to--wait for it--geology. No, autonomous systems don't need to be experts on rocks; Fink argues the value of geology rests in its reliance on inductive reasoning, a skill that sometimes takes years to strengthen. "You will find a stellar genius physicist at 15 or 16, but not a geologist," Fink says. "Geology works on a different level."
So far, robots are stuck with only deductive reasoning skills, or top-down logic. With deductive reasoning, a conscious being works from general information to a more specific conclusion. It's a process that essentially determines “if A then B.” For example, a scientist will start with a theory or broad statement, such as “all astronauts have flown in space.” Then when told that Buzz Aldrin is an astronaut, the scientist will conclude that Buzz has flown in space. With their reliance on rule-based systems, A.I. bots have this process nailed down.
It's the opposite form of reasoning--induction--that has roboticists stumped. It's bottom-up logic that enables drawing a broad conclusion from few observations. For example, if a person met one astronaut with brown hair, he or she might inductively conclude all astronauts have brown hair. That's not true, of course; the person just hasn't met enough astronauts. And that's what makes inductive reasoning so tricky: It's intelligence that accommodates new information and strengthens through experience.
Inductive reasoning is a process geologists must use every day, making the profession difficult to master. "You see a rock that shouldn't be there, a geologist will think, 'Well there may have been a flood,'” says Fink. “So they might look for evidence of a riverbed. If that's not being corroborated, they must reject the idea and start over again.”
For that reason Fink works closely with geologists, since he hopes to untangle the elements of the inductive reasoning process. It's what makes humans both self-aware and situation-aware, because it insists we pay attention to our life experiences and surroundings and use that information to shape our future actions. For a robot to really be conscious, it will need to do the same.
“It all starts with observing something in your environment, thinking what could have caused this, coming up with a hypothesis, and then that tells you what to do next,” says Fink. “My claim is this will be the path to truly autonomous systems.”
What would a truly automated system look like?
Let's say a robot with inductive reasoning comes to fruition. How would it behave, exactly? Fink has a simple answer: We won't know it until we see it.
“If you give a robotic system the capabilities and tools to absorb the environment, be taught and learn, and be judged by its actions and then modify its thinking,” Fink says, “then the sky's the limit as what it will develop into.”
In Fink's world, CHAPPiE gets that part right. The robot has inductive reasoning skills, and his personality is largely based on his experiences with the people who “raise” him. (Including, unfortunately, some gangsters who kidnap him.) Yet even if the most benevolent of people trained an autonomous system, Fink isn't convinced it'd lead to either a compassionate entity or an evil one--it'd just be different.
“It's not governed by these core values. It may have a way of thinking and acting that may be incomprehensible to us, but it's a totally valid way of thinking," he says. "It just may be devoid of ethical behavior and moral behavior.”
It's that unknown outcome that scares public figures like Hawking and Musk. Most, if not all, human behaviors stem from our desire to obtain a certain goal, be it survival, finding love, protecting our families, etc. A robot can be programmed with the same kinds of optimization goals, which would help us predict the ways in which it would behave. But if the robot is truly autonomous, it will have the ability to update or change these goals whenever it chooses--a luxury we lack, given our biology (and biological motivations)--leaving us in the dark about what it really wants. We'll never be able to anticipate what it might do.
Fortunately, Fink says we have some time before we see an autonomous system in action. Researchers haven't even come close to mastering artificial inductive reasoning skills. But as Blomkamp depicted in CHAPPiE, radical technologies often seem to arrive out of nowhere. “It will not be an incremental process, but a disruptive process,” says Fink. “It'll be a discontinuous jump to true autonomy.”
