In the film, HAL makes several plans: to test whether the AE35 unit was in fact faulty, to navigate the ship, to search for extraterrestrial life around Jupiter, and to kill the crew. Such planning, Dave Wilkins points out in chapter 14, requires the identification of subgoals, anticipating obstacles, retracing steps, and so on. His review of the progress in planning by computer demonstrates the difficulty of solving even apparently simple problems in a rarefied and idealized world of stacking blocks. The bottom line is that planning is hard! HAL was not particularly good at it, and neither are current computer systems.

As the characters in the film admit, it is hard to tell whether HAL's emotions are programmed in to make him easy to talk to, or whether they are genuine. Rosalind Picard (chapter 13) discusses HAL's emotions -- his pride, anger, fear, paranoia, concern -- as well as his ability to recognize emotion in crew members -- and how current research approaches similar problems. If emotions are essential for computer cognition, as she and Minsky and Norman argue, how will we deal with such "affective" computers? Would you trust an affective computer with your spreadsheets?

Don Norman (in chapter 12) underscores the importance of emotion, and points to a notion of machine (and human) intelligence much broader than the kind of logical intellect that preoccupies the field. He looks at what it might be like to live in space and work with computers for extended periods. His discussion stresses the need to take into consideration the "softer" aspects of cognition, those related to emotions, making mistakes, and so forth.

It's interesting that Kubrick and Clarke seem to want us to have stronger feelings toward HAL than we do toward the crew. HAL is the only one in the film to show emotions: "I'm afraid. I'm afraid. I'm losing my mind. I can feel it. I can feel it." In contrast, the dull, robotlike astronauts sleepwalk through boring meetings and chat about ham sandwiches. (Three other characters are known solely through the trace of their biological functions during hibernation.) When the BBC announcer, Mr. Amer, greets the crew, they mumble indistinctly in response. HAL, however, answers with clarity, animation, and interest: "Everything is going extremely well." By the middle of the movie, we have grown accustomed to HAL and accept the fact that he has more personality than the crew. While we may be shocked and surprised at the death of Frank, HAL's plaintive "I'm afraid .... I'm afraid" evokes our sympathy.

In fact, the audience's sympathy (or anger or both) toward HAL makes more poignant a number of ethical questions. Does HAL murder Frank and the three hibernating crewmen? If so, who (or what) should be punished? Is it immoral to disconnect HAL (without a trial!)? Daniel Dennett argues that higher-order intentionality and responsibility are necessary conditions for moral responsibility -- and hence blame -- and that HAL exhibits such intentionality. After all, he tells Dave "I want to help you" (though, of course he says this to try to stop Dave from dismantling him).

A broad question touched on throughout this book is whether we should try to make computers intelligent by mimicking a human brain or, instead, exploit their particular strengths -- such as rapid search and large memories. So far, different domains have tried different approaches. Researchers in computer chess, for instance, began by trying to reproduce the methods of human grandmasters -- in particular recognizing key configurations of pieces on the board -- but found that this approach quickly led to such difficulties as determining and representing the key properties of the arrangements. Massive and rapid searches of possible sequences of moves have proven more successful. This approach is used for Deep Blue, which has rapid search capacities that are distinctly unlike those of human grandmasters (although grandmaster Garry Kasparov has commented that Deep Blue plays like an intelligent human). Likewise, the best computer speechreading systems operate not at all like humans. Both of these applications are limited and -- as valuable and interesting as they may be -- do not lead us toward true AI.

Kurzweil's brain-scan approach is the most extreme of the duplicate-a-human-brain approaches. Certain cognitive abilities -- in particular, language and our species-specific common sense -- are distinctive to human beings. There are strong arguments for trying to copy these capacities in computers. While this would not require duplicating a brain to the level of nerve cells, it would, presumably, involve identifying the specific methods we use to carry out myriad tasks.

Another theme suggested by 2001 is society's perception of computers, contemporaneous and future. Were public fears about computers in the 1960s borne out? The fact that Frank takes his loss to HAL at chess without the slightest surprise reflects an attitude that is radically different from the public's perception in that decade when the thought of a computer becoming the world chess champion evoked anger and hostility.

As we approach 2001, we might ask why we have not matched the dream of making a HAL. The reasons are instructive. In broad overview, we have met, and surpassed, the vision of HAL in those domains -- speech, hardware, planning, chess -- that can be narrowly defined and easily specified. But in domains such as language understanding and common sense, which are basically limitless in their possibilities and hard to specify, we fall far short. Perhaps too we need to ask whether, as a culture, we are willing to support the undertaking of producing artificial intelligence.