Stork: Did it seem plausible?
Dennett: It did seem plausible to me. It seemed as if a lot of attention had gone into trying to imagine how to extrapolate the matters of time into the future, of course it didn't all come out, but they did a pretty good job.

Stork: Did Hal commit murder?
Dennett: The question of whether Hal committed murder really comes down, as it does in human cases, to whether Hal had the right kind of mind. Did Hal have a guilty mind? Now, just being the causal nexus of a homicide isn't enough to be called murder. Industrial cranes kill people every week and they don't commit murder. An industrial robot in Japan once was causally implicated in the death of a worker, but that wasn't considered murder because an industrial robot doesn't have the right kind of mind. Now, the question is whether Hal's mind has the right structure and the right contents to be up for the charge of murder. Or whether it would have to be just another industrial accident.

Stork: So how do we tell whether Hal's mind had such contents?
Dennett: Well, Hal is definitely what I call an intentional system. Hal had beliefs and desires, who manifestly, if you listen to the things that Hal says and see what Hal does, Hal has a huge amount of information about the world, those were its beliefs, and it has goals, it has projects. Hal has a mission. So there is an elaborate and structured and largely rational, well working, this is not a completely deranged believer and desirer, this is largely rational believer and desirer. That's the first requirement, you have to be rational agent. But more than that you have to have the right kinds of complexity in your rationality. You have in particular to be able to reflect on your own beliefs and desires, and there are some passages in the film, in which what Hal says makes it very clear, that Hal can remember what he used to think, can be worried about making a mistake, can be worried about things that don't seem to add up to him, so Hal manifestly exhibits higher order states, that is, just not beliefs, but beliefs about his own beliefs, and desires about his own desires. So, Hal gives the evidence of meeting the conditions as far as that goes. So then the other condition is, is there some excusing condition? If there is, its for the same sort of reasons that we would be excused. Was he under terrible duress? Was he massively misinformed in ways that he had no control over? Was his own life threatened for instance? And, if any of those conditions are met, if I were to defend him in court, I would push those issues pretty hard.

Stork: So given that Hal committed murder, who's to blame?
Dennett: If I committed murder, I'm to blame. There may be many factors that contribute to this, maybe people want to blame my parents to a certain to degree, or maybe my associates more likely, but if I am functioning, basically sane human being, then I'm to blame, and if Hal is a basically functional, sane, rational robot agent then Hal is to blame. We're just going to have to get used to the fact that this is a new feature of our world.

Who beat Kasparov in Chess? Not Murray Campbell. Deep Blue was the one to blame in effect there. Nobody else can take responsibility for that. And, so, moral responsibility is just one notch up from the sort of responsibility that you get in games.

Stork: So what would be a reasonable punishment?
Dennett: Ah. One of the curious features about most treatments of robots is that we ignore the fact that maybe if we're a robot, we're clever enough and sensitive enough to be morally responsible, then we would also be clever enough and sensitive enough to be punishable. And, there is evidence, of course, right at the end, when Hal seems to be in considerable distress, and so Hal is capable of distress, or so it would seem, and so the punishment would be pretty much the same as for human beings, of course, we might do something really cruel and unusual, and require Hal behind the machinery behind some stupid video game or perhaps handle the airline reservations all over the world, something extremely tedious and boring and mind deadening. I hadn't thought before what a good punishment would be, but I am sure we could dream up some pretty horrific ones.

Stork: Public service?
Dennett: Public service is not typically a punishment as a means of atonement, and of course there would be lots of ways in which Hal could benefit the other sentient beings around, so that would be an appropriate use of Hal's talents.

Stork: So would it be murder, if one of the crew men without provocation, disconnected Hal permanently?
Dennett: One of things about robots in general, and Hal in particular, is that they are so easily resembled if they are taken apart. I think that would be murder, would be destroying the memory banks beyond repair. Simply taking the memory apart in a way that could permit it be put back together later, that's not murder, that's just putting it into a coma, that's just a time out. It might have very evil consequences, but not in particular for Hal.

Stork: You've written a lot about evolution, and how that has influenced and been the source of human consciousness, perhaps machines don't have to go through an evolutionary sequence the way humans do, perhaps it's a sense of self, perhaps it's type of intelligence would be different. Do you want to comment on that?
Dennett: No matter how Hal came into existence, just to take care of himself in his ongoing day he's going to have to have a very robust sense of self, narrative self, because he's going to have to keep track of what he already knows, what he doesn't need to know, what he doesn't want to know, he's got a flood of information coming on, he's has to keep track of that, and keep track of his keeping track of it. So Hal will have a very robust sense of self or it won't work at all.

Stork: Why is that a sense of self?
Dennett: Because that is what a self of sense is. A sense of self is the organizing principle that gets you through the day. In fact I think this is an ancient idea that goes back to Aristotle, who talked about plants having what he called a vegetative soul. And what he didn't mean is that they had a little pearl of plant soul that was immortal. What he did meant was that a plant has an organization, it has a control structure that has to be in order, otherwise the plant will die. It has to keep its act together. A self or a soul is simply the organizing principle, its whatever they are, however they are embodied, that keeps the act together. So the overall control organization of Hal is itself.

Stork: Would you tell us about the Turing test.
Dennett: In the middle of the 17th century, Rene Descarte said there was a fool proof test for distinguishing a machine from a human being. He said have a conversation with it. Only a human being, with an in-material mind, with a soul, could understand words, not just a few words, but the vast number of different things that we can say, just using a few words. Hundreds of years later, Alan Turing re-invented this idea, as a test for intelligence, have a conversation. Now what Turing proposed was an addition, in effect, where you conceal the contestants behind a screen so that our biases are not in play. So we can't use the fact that we like our intelligent beings to have soft skin, or to have hair on the top of their heads rather than being made out of metal, we shouldn't allow such biases, such potentials for discrimination to interfere. We want just what matters to pass through the screen. What matters is intelligent conversation, says Turning. So he proposed a test where put either a human being or machine behind the screen and have the judge conduct a conversation.

Approximately 50 years ago, Alan Turing proposed a test which he thought should settle to anybody's satisfaction the question of whether or not a robot was intelligent, whether or not a robot was conscious. He said put it behind a screen and let it have a conversation with a human judge. If it can convince the human judge that it is a human being, then its intelligent. To make it a proper test, he supposed that you might have another contestant behind the screen who was a human being. So now the judge's task is which one of these is more likely to be the real human being. If the robot could convince the judge that it was the human being, that it was a better human being in conversation than the real human being, then it had to be intelligent. Now Turing didn't put this forward as a proper methodology for research into artificial intelligence, he put it forward as a conversation stopper, a philosophical thought experiment, but it has subsequently been often treated as if this was the goal of artificial intelligence, which is to make a robot that can pass the Turing test. Well in one sense, I think that is just fine, because in fact the test is extraordinarily difficult, because Turing said, on any topic. If you open up so that any topic of conversation is available, the demands, the strain on any robot that is trying to present itself as human are just extraordinary. Now sometimes people sometimes thought, but a computer. . . couldn't see, couldn't hear, doesn't have any sense organs at all, doesn't have a body, aren't these important? If you stop and reflect, no, a computer that probably that didn't have sense organs and so forth probably couldn't pass, just physically couldn't the Turing test, because so many questions that we could can raise with people draw on their sensory experience, on their bodily experience, and the only realistic way to get that information, is to have a body, and have used it, and gained experience from the real world. So as a matter of brute physical fact, it's impossible for anything to pass the Turing test that isn't pretty much like us.

Stork: Hal passed the Turing test. I am presuming you would agree?
Dennett: Certainly, Hal gives evidence of being fully capable of passing the Turing test.