Transcript EECS 690
EECS 690 May 4 LIDA • Most of this chapter concerns specific ways in which a system like LIDA might be an accurate model of human psychology. • This sounds like an interesting project. It would be interesting to see it built and developed. Moral Mimicry • The authors point out that they do not mean to replicate human psychology wholesale, but they do not spend much time on exploring to what degree such mimicry of human moral psychology is necessary for AMAs. (We don’t want to specifically make AMAs who are more generous after finding a dime, but such foibles may be unavoidable in such a complex system) • Questions of how people come to understand morality are called questions of moral epistemology. Moral Facts • In order to talk about coming to understand morality, it is necessary to ask what there is to be understood. • Anti-realists about morality contend that moral language corresponds with no recognizable category of facts. • Realists correspondingly associate moral language with one (or more) of a number of factual categories, a few of which are: – – – – – – – sociological belief human well-being the outcomes of rational procedures interpersonal conventions human well-feeling natural science theology Human moral psychology • Notice that the specifics of human psychology are only related to some of those categories, and are related in different ways. Remember that the question here is not whether these things influence human behavior, but whether what is moral is determined at least in part by these things. – – – – – – – sociological belief human well-being the outcomes of rational procedures interpersonal conventions human well-feeling natural science theology Recognition problems • Because there is not widespread agreement on many aspects of what moral facts are or even what behavior is moral or not, Wallach and Allen leave unaddressed the question of success conditions for LIDA. After all, it does not follow that something with a similar cognitive architecture to ours would have similar moral behavior to ours, or that we would even want it to. Complexity of AMAs • I have several times raised the question of whether to make a McRobot (a machine that worked at McDonalds) that was sensitive to certain ethical concerns it would be most productive to make a machine that could enroll in college, could order at a fancy restaurant, could learn a new game, could criticize a movie, and write a passable limerick and then teach it to work fast food. • Could we deal with a simpler form of ethical sensitivity tailored for specific tasks? Is a better goal AMAs that are as ethically sensitive as well-trained dogs?