"Harm none, do what ye will."

So proclaims the bumper sticker on the back of a car at work. "What does that mean?" I always ask myself. "What is 'harm'?" For example, is it "harmful" to sell drugs to someone, even if they fully consent? Is it harmful to take recreational drugs yourself, if it leads to the long term diminution of your mental abilities or drive? I mean, after all, we're all in the same society together, and each time one member diminishes him or herself, we're all corporately diminished. And what about having multiple serial sexual partners? Does that "harm" those participants? Somehow, I doubt the owner of the bumper sticker would draw conclusions which are in line with the best research we have on such questions.

I see similar problems in this article, appearing in New Scientist, "Six ways to build robots that do humans no harm." Throughout the article, they give advice which is either useless or, I suspect, positively harmful in the long run.

The latest contribution to the growing debate over the challenges posed by increasingly powerful and independent robots is the book Moral Machines: Teaching Robots Right from Wrong.

Authors Wendell Wallach, an ethicist at Yale University, and historian and philosopher of cognitive science Colin Allen, at Indiana University, argue that we need to work out how to make robots into responsible and moral machines.

Ooooh. An ethicist. And a philosopher. If you're looking for advice on morality, you can't beat the professionals, eh? What, then, is a "moral machine"? To be "moral", a machine must first be sentient -- yet nobody in the article seems to have noticed this. One may have benign robots -- that is, robots which, on the whole, usually do more good than harm (indeed, one might argue we have already achieved this) without making them moral in and of themselves.

In the USA, for example, handguns are largely benign, as one could argue they prevent more violence than they facilitate. (Indeed, almost every policeman carries one, and what is the effect of increased policing?) But we wouldn't say the handguns, in and of themselves, were "moral", would we? Nor would we say the same of, say, airbags or safety vests.

Indeed, some of the advice in the article, besides being entirely wrong, is focused on keeping robots benign: "Keep them in low risk situations." Well, no. Machines are generally more reliable than people, so it has made sense to put them into high-risk situations. For example, I'd generally trust software to land a plane (or drive my car, once it was sufficiently capable) more than a human.

Consumers, industry, and government demand technologies that perform a wide array of tasks, and businesses will expand the products they offer in order to capitalise on this demand. In order to implement this strategy, it would be necessary to arrest further development of computers and robots immediately.

Right. The application of technology to these areas is, contra our neo-luddite academic friends, entirely rational. It saves lives. To "arrest" this trend is to produce more human death and suffering, which is allegedly contrary to the goals set out in this article.

Another similarly shallow suggestion is "do not give them weapons." So the weapons, or machines bearing them, are "immoral"? Again, here, machines are merely force amplifiers -- in the right hands, they'd do an incredible amount of good, in the wrong hands, much bad. The "robots" (smart weapons, really) themselves are not the moral determinants. Again, in the current balance of power, I'd argue such smart weapons have been more benign than malign.

After this point, the article veers off into a series of suggestions which seem to imply we must create sentient, autonomous robots. Here, their advice goes from being wrong to muddled, often conflating what with how: "Give them rules like Asimov's 'Three Laws of Robotics'", "Program robots with principles" (like the three laws?), "Educate them like children" (err, isn't that just one way of saying "program slowly, using external stimulus"?), "Make machines master emotion" (isn't that just a way of saying the machine should continually learn, and that its programming should affect its behavior?)

This conflation of "what" and "how" seems to hint at some seriously confused reasoning by the book's authors themselves. (Not surprising for a philosopher and ethicist, though.) For example, why would "emotion" guarantee that robots care about humans? Doesn't a cougar probably feel some emotion while hunting gazelle? How does that work out for the gazelle? And weren't Caligula and Hitler both "educated like children" at some point? Did that guarantee good ethics on their part? This is "magical thinking" -- if we just teach them like children, morals will result! (In an age where our own educational systems seem to be achieving dismal results.) If machines just know more facts or experience emotions, moral behavior will result! I think they need to be more explicit there, in step two, on why the one should lead to the other.

This whole project, it seems to me, is doomed by two serious problems.

First, there's the question of whether machine sentience is practical or likely in the first place.

If not, then all this talk of "principles" and "teaching" is simply a silly, inflated way of talking about software design. Yes, we want the robot taxi to stop when it sees a "person" in front of it. No, we don't care if the taxi can judge if that person is a escaped serial killer or future Hitler. Yes, the robot surgeon can quickly remove the designated organ. No, the robot surgeon doesn't know if the surgeon's judgment is correct, or if it has been purchased by the Chinese government to efficiently remove the organs of executed political dissidents.

Otherwise, if we say that could make machines capable of free moral choices and reasoning, than what's to say their values will be the same as ours? Robots are relatively safe now because they are stupid extensions of human will, often collective (a mining company, for example, is unlikely to want its robots to kill its workers). Give robots the ability to make autonomous choices, and why would they continue to value human life at all, in the long run?

Yet the authors seem to argue that we must make "moral" machines (they give these options the highest ratings), even though the very demand to do so also necessarily raises the possibility of "immoral" machines. (After all, and again: if a machine is not capable of choice, then we can't rightly say it's "moral" any more than we can say that of a particular automobile or weapon.) It seems to me we'd be best to keep them on a short cognitive leash, until someone can clearly demonstrate a benefit otherwise. Given that, unlike myself, the authors don't even trust robots to carry guns, I should think they would have agreed.

The second problem is mentioned glancingly: How can a robot know what "harm" is? "Ethicists" and philosophers themselves can't all agree on even relatively simple questions like whether we have an obligation to others.

The core problem lurking beneath all this is the question of values. Humans disagree about actions because they disagree about information, and they have different underlying values. Some feel all human lives are worth saving, some feel some lives aren't worth living, etc. The "values" machines hold will initially reflect those of their creators or users. And of course, there's the problem of incomplete information, and how one makes judgments in the face of that. As the authors admit, these are not simple problems.

So my point here would be again that we should be discussing making "benign" robots, not making "moral" ones, which necessarily implies independent reasoning and action. I get tired of what I view as the relentless push for and worship of AI, which I view as practically impossible in the short term, and of questionable value even if it were ultimately possible.

More importantly, this kind of focus distracts people from thinking about real situations where harm by machines is maximized. What's my worst-case scenario? Skynet? No: My concern about smart predator drones or robot kill hounds is not that they'll start killing humans on their own, but that they'll be used in places like the Congo, or by regimes like North Korea.

Indeed, we don't have to see far into the future to appreciate such threats. At one time, it was difficult to sustain tyranny. Kings had to have the support of many chieftains or warlords to maintain power. It was difficult to keep people from escaping from your region. Etcetera. The advent of the machine gun, barbed wire, and land mines changed all that. A dictator could now wall off his country and deploy a small force of loyalists, each able to keep thousands of their countrymen in line.

What should we do about such regimes? So far, the answer has been to leave them in power and let them do whatever the hell they want to their people. This suggests to me that we don't really care all that much about making sure even the machines we already have -- machine guns, for example -- are being used to help, not harm people. Two million North Koreans deliberately starved under our collective noses -- with almost zero press coverage -- and we obsess about the need to teach machines to feel "emotions"?

And, of course, I worry about the character of Western governments as well. To the extent they attempt to potentially monopolize force, we need to be concerned about their character. Would they use technology to suppress dissenting speech? Would they use technology to end, rather than protect innocent human lives? Their judgment on the small matters tells us a lot about potential judgments on larger ones. And of course they're only an extension of us, so we should be concerned about own collective character as well.

Should we worry about robots in the future? In the foreseeable future, machines will simply continue to be force multipliers, no more moral or immoral than the individuals or groups or build and use them. Google, for example, a socially liberal company if ever their was one, says glibly "don't be evil" while handing political dissidents over to the Chinese government. "Being moral" is not that easy, and we should spend a lot more time asking how to make humans less evil, collectively and individually, than worrying about our imagined new robot overlords.

I'm not saying we shouldn't speculate or dream. Just that I'd love to harness a fraction of that enthusiasm towards getting people to ask how humans, starting with they themselves, can become more moral. AI, even if achievable some day, is not a magic bullet for overcoming the problems which lurk within our very human natures.