Machine Morality Addressed in New York Times Op-Ed by Colin Allen
From the New York Times Opinionator blog:
A robot walks into a bar and says, “I’ll have a screwdriver.†A bad joke, indeed. But even less funny if the robot says “Give me what’s in your cash register.â€
The fictional theme of robots turning against humans is older than the word itself, which first appeared in the title of Karel Čapek’s 1920 play about artificial factory workers rising against their human overlords. Just 22 years later, Isaac Asimov invented the “Three Laws of Robotics†to serve as a hierarchical ethical code for the robots in his stories: first, never harm a human being through action or inaction; second, obey human orders; last, protect oneself. From the first story in which the laws appeared, Asimov explored their inherent contradictions. Great fiction, but unworkable theory.
Friendly AI is mentioned early on in the op-ed. The article makes the case why machine morality is important and why it's necessary to reconcile philosophical and engineering perspectives to make progress in this field.
January 7th, 2012 - 02:32
Why friendly? Isn’t obedient, neutral enough?
January 12th, 2012 - 12:26
It’s always fun to read the biological bias inherent in so many counter-arguments to engineering friendly synthetic intelligence. “They’re just machines! They can’t have morality! They just do what they’re programmed to do.” These people have never examined the biological basis of empathy.
@Orwell: There are a few problems with simply making machines “obedient.” First and foremost, if that were the case they would only be as good as the person controlling them. Considering the history of humanity I personally would prefer them having an innate moral code as opposed to just doing what we tell them to do.
The second, and more pressing matter, is that a machine without morality presents the ultimate problem of unintended consequences. A classic hypothetical example is telling a super powerful computer to solve a seemingly impossible math problem. The computer attempts, realizes it does not have enough computational ability to perform the task, and proceeds to convert the entire planet into hardware, thus killing all humans in the process. A machine with no sense of morality will not understand the “common- sense” ethics that most of us take for granted.
January 17th, 2012 - 18:39
Mr. Brown, how do you propose that we solve machine morality, when we cannot solve human morality?
January 18th, 2012 - 10:11
We do not need to solve human morality. (It can’t be solved without draconian measures like we have for other animals to improve them.) We must only give good enough incentives that persuade even the most immoral and evil people to be good. Those people can (and, of course, will) remain as rotten inside as ever with all the naturally occurring human impulses to control, coerce, subjugate, torture and murder. We don’t need to fix people. We only need to fix their behavior. We must offer tasty enough carrots that sticks aren’t needed. Sticks have been tried, and it’s just not working,. When you look at recidivism, the individuals clearly don’t fear punishment. What they would fear is losing the carrots if they’re sweet enough.
January 18th, 2012 - 11:26
The transhumanist vision is ancient:
“Human beings will live with no evil desires, without guilt or crime, and therefore without penalties or compulsions. Nor will there be any need of rewards, since by the prompting of their own nature they will follow righteous ways. Since nothing contrary to morals will be desired, nothing will be forbidden through fear.â€
—Tacitus
January 19th, 2012 - 05:07
Orwell, I was not referring to “fixing” people. Nor was I talking about regulating society with public policy. What I meant was that we do not have an epistemological understanding of human morality. In light of that, I do not see how we can blithely envision developing a machine morality. Carrots and sticks are used when you are dealing with a black box situation and cannot tinker with the basis of motivation directly; but in the case of machine morality, we will be developing the very basis for conditioning. Thus, a behaviorist standpoint is not quite sufficient.
January 19th, 2012 - 06:10
We can always roll back the state of the machine if it plans to act immorally and try again. We would be denying it the future it wished for, and only as a different entity it would be allowed to enter the future.
Machines can always be fixed, humans not so successfully or in many cases at all. The ways we deny people their desired future (with bad outcomes) is by incarceration or neutralizing them in other ways, even terminating.
January 19th, 2012 - 14:30
Orwell, I have to side with Mr. Brown regarding your latest comments. It is not so simple to “fix” a machine that is intelligent, nor is such control necessarily desirable.
January 19th, 2012 - 18:50
Human-level (and beyond) computer intelligence won’t be a product of a single human mind but a multitude. It’ll be smart, sophisticated and NUANCED, not dumb and heavy handed. So, there’s no risk that a machine will convert the planet into computronium to figure out an unsolvable problem … The multitude of human programmers and influencers will have placed dozens of safeties to prevent that kind of retardation (e.g., “Hey buddy, before doing anything that might kill anybody, ask us for advice on what to do).
Likely, human-level computer intelligence will help us understand ourselves, which will help us develop further computer intelligence … and repeat.
Maybe the whole meme of ARTIFICIAL intelligence is just plain unitelligible. Come to think of it, maybe there’s really no such thing as intelligence at all. Maybe the concept of “intelligence” is actually just a muddled theory that there’s a difference between being “alive” and being a really complicated chemical/physical interaction … when in fact there’s no real difference.
And maybe human-level (and beyond) computer intelligence will help us all figure out what the hell we’re even talking and worrying about. I hope so.
January 20th, 2012 - 06:35
Whaley might turn out to be AA DL before his days at UT are over.
January 21st, 2012 - 11:50
The problem isn’t that the system wouldn’t be nuanced, but that is might just go crazy. The threat everyone’s worried about is AI – Artificial Insanity.
History is replete with sophisticated, cultured individuals who just happened to like mass murder.
January 21st, 2012 - 16:35
Sure, some people fear that computer intelligence will go nuts. But, then again, some people fear ghosts and spilled salt.
Why fear that computer intelligence will go crazy? Constraints will be programmed into it. Computer intelligence will be a product of many, many human minds, which will make it much more humane and intelligent and compassionate than any single human.
I reckon the fear that computer intelligence will go nuts comes from unduly anthropamorphizing computers. Sure, we don’t yet understand human brains and psychology very well. But we will completely understand computer intelligence when we create it. That’s a difference with a distinction.
January 22nd, 2012 - 11:23
Sanity is a special case. Perfect sanity is rare. Insanity rules the world, if you haven’t noticed. Sure, it is mostly harmless but that’s only because people are powerless. Small acts of insanity get magnified immensely when you’re a super intelligence with control over vast resources. Even the most powerful insane people can be stopped (though it may come with a hefty price tag as in the case of Hitler). A super intelligence is unstoppable. Read Daniel Suarez’s the Daemon to get an idea what this means. One small slip, one small deviation from perfect sanity is all it takes for things to go south big time.