HOME / life and art: What really happened.

Gaming the Robot RevolutionA military technology expert weighs in on Terminator: Salvation.

By P. W. SingerPosted Thursday, May 21, 2009, at 3:50 PM ET

(Continued from page 1)

RSS

The third condition for a machine takeover would be the existence of independent robots that could fuel, repair, and reproduce themselves without human help. That's far beyond the scope of anything that now exists. While our real-world robots have become very capable, they all still need humans. For instance, the Global Hawk drone, the replacement for the manned U-2 spy plane, has the ability to take off on its own, fly to a destination 3,000 miles away, and stay in the air for 24 hours as it hunts for a terrorist on the ground. Then it can fly back to where it started and land on its own. But none of this would be possible if there weren't humans on the ground to fill it with gas, repair any broken parts, and update its mission protocols.

Finally, a robot invasion could only succeed if humans had no useful fail-safes or ways to control the machines' decision-making. We would have to have lost any ability to override, intervene, or even shape the actions of the robots. Yet one has to hope that a generation that grew up on a diet of Terminator movies would see the utility of fail-safe mechanisms. Plus, there's the possibility that shoddy programming by humans will become our best line of defense: As many roboticists joke, just when the robots are poised to take over, their Microsoft software programs will probably freeze up and crash.

The counter to all of this, of course, is that a superintelligent machine would figure out a way around each of these barriers. In the Terminator story line, for example, the Skynet computer is able to manipulate and blackmail humans into doing the sorts of things it needs. It's also able to rewrite its own software, a scenario that may be not so far-fetched. There is much work today on "evolutionary" or self-educating artificial intelligence that can even begin to take on its own identity. Just as humanity ended up with both Gandhi and Hitler, there is no guarantee that our machines will evolve to feel only love and compassion.

Most importantly, we rarely take heed of the lessons of science fiction. The military routinely carries out research into systems against which writers and filmmakers have long warned. Indeed, the scientists are often directly inspired by those cautionary tales. For instance, H.G. Wells' dark fantasy of what he called an "atomic bomb" in the 1913 anti-war story The World Set Free actually helped guide the thinkers behind the Manhattan Project. In my book, I mention how one robotics firm was asked a few years ago by the military whether it could design a robot that looked like the "Hunter-Killer robot of Terminator." (It wasn't such a silly request. The design would be quite useful for the sort of fights we face now in Iraq and Afghanistan.)

In my final judgment, however, The Terminator may not be the best guide for how a machine takeover might take place in the real world. Instead, another science fiction series, The Matrix, may be more useful. By this I don't mean that we can look forward to a future of humans living in jelly bubbles and Keanu Reeves' avatar running about in leather pants. Rather, the films give us a valuable metaphor for the technologic matrix in which we increasingly find ourselves enmeshed but barely notice. For all our pop-culture-stoked fears of living in a world where robots rule with an iron (or digital) fist, we already live in a world of technology that few of us even understand. It increasingly dominates how we live, work, communicate, and now even fight.

Why would machines ever need to plot a takeover when we already can't do anything important without them?

Page: 1 | 2SINGLE PAGE

E-MAIL

RECOMMEND... Get Slate RSS Feeds

P. W. Singer is the director of the 21st Century Defense Initiative at the Brookings Institution. In his personal capacity, he served as coordinator of the Obama '08 defense policy task force. He is the author of Wired for War: The Robotics Revolution and Conflict in the 21st Century (Penguin, 2009).

COMMENTS

No, no, worry about gray goo instead.

I don't see any way we could end up having a skynet-style robot enemy on our hands. This scenario is based upon Artificial Intelligence being developed and "choosing" to destroy us. This probably won't and can't happen. What CAN happen, is we could create a fleet of robots that accidentally destroys us because we type in the wrong command.

Maybe a billion trillion self-replicating nano-robots designed to clean up an oil spill get the wrong message and start consumer anything with carbon in it. But I highly doubt we will see a robot ever make a sentient decision to fight humanity.

That's because nothing non-organic has ever been the slightest bit sentient. There's a lot of misleading info about artificial intelligence that gives people the impression that machines are "learning to think." No, they are learning to compute equations faster which makes their reaction time and deductive capabilities appear to be thought-based. But they aren't, they are program based. What is more sentient; a pocket calculator, or a supercomputer? They're both equally un-thinking machines.

But if computers can work ever faster, won't that come to mimic human thought? Or some kind of thought? No, because everything a computer does is predetermined by its programming. The smartest computer in the world will not be able to tell you how to tie your shoes if it isn't programmed to. If the slightest bug hits its programming, the rest of it screeches to a stop. Machines have static programming, and can't appraise situations outside of that. This applies to 100% of all "artificial intelligence" programs or machines that have ever been created.

Real intelligence is the ability to absorb and process new information. No computer in history has ever processed something that it wasn't specifically designed to. Because it can't, and because artificial intelligence is far more artificial than it is intelligent.

In thinking about this, there is only one way I see that such "thought" could emerge. Perhaps is computers were given the ability to constantly reprogram themselves, then perhaps, maybe, some new sort of evolutionary process could spontaneously generate real decision making capability. But that doesn't seem very scientifically sound to me.

-- jwschmidt
(To reply, click here)

Jw,

I think we all agree that computers are not even close to a human level awareness, (even if we knew exactly what software we needed, the calculations per second aren't there yet). However, you seem stuck on this idea that computer programs are inflexible. In fact there are many, many examples of functional "narrow ai" that have "learned" to find patterns from complex inputs sometimes independently. The genetic algorithms I linked above are just one way that programmers do this. Another is with artificial neural networks. Obviously we are still at the point where humans create the parameters, but the results are not always predictable even with the same initial hardware and inputs.

I don't know where along the animal kingdom we could classify the current generation of AI's, and in a way it's meaningless right now. However, as someone else said if the exponential rise of capacity continues, there is no theoretical barrier to software that is smarter than us.

-- blueshift
(To reply, click here)

What did you think of this article?

Join The Fray: Our Reader Discussion Forum

POST A MESSAGE | READ MESSAGES