CMSC 491M/691M - Spring 2003

Discussion Questions for Class #15, March 17

Reading: Emotion paper: Sloman, Concar, Archer

What is an emotion? What's the difference between an emotion, an attitude, and a mood (according to Sloman, at least)?
What are primary, secondary, and tertiary emotions? Is this distinction crisp or soft?
What does Sloman mean by shallow vs. deep models of emotion?
What function do you think emotions serve in human thought and cognition? What are the possible purposes of modeling human emotion? For what purposes/applications are shallow vs. deep models appropriate?
How could you evaluate a model of emotion? (That is, how would you know whether you had a "good" model of emotion?)
Similar to the "intentional stance," we can take the "emotional stance" to ascribe emotions to agents. When and why might it be appropriate to take such a stance?
Can an agent have emotion without having consciousness? (We can leave this discussion for Wednesday, when we talk about consciousness...)
What is the purpose of the horizontal and vertical layers in the CogAff architectures? Where are emotions in this architecture? Do they emerge from the organization of the system or are they imposed onto the organization?

Sloman says that Damasio's claim that intelligence requires emotion is a fallacy; essentially Sloman's argument is that this conclusion is only one interpretation of the evidence. Concar's article starts from the premise that Damasio is correct: that feelings are needed (at least in humans) in order to make judgments among alternatives. Without having read Damasio's actual research, what are your intuitions on this question?
The examples that Concar gives all seem like very shallow models by Sloman's definition: recognizing emotional events from surface features in a video, adapting to a human performer's musical style, designing "emotional agents" to model simple social interactions. Is it that these examples are shallow, or are these just victims of the "as soon as you model it computationally, the magic is gone" problem that pervades AI?

The focus in this article is not on building intelligent agents with emotion, but on modeling human emotions to improve interactions with an agent. Do you think that the same emotional models are appropriate for these two tasks?
Picard et al.'s system can recognize eight different emotions with a reasonable degree of accuracy. Is this level of granularity sufficient for sophisticated social interactions?
Do you really want your computer monitoring your pulse rate and skin conductance? Don't you think that would get awfully annoying? (Isn't the paper clip already annoying enough?)