There was an old suggestion of making an AI learn human values by training it on happiness and unhappiness in human facial expressions, making it happy when humans are happy and vice versa. Besides its other problems, now there’s this...
We present an exploratory study of analyzing and visualizing player facial expressions from video with deep neural networks. [...] As an additional contribution, we show that although there has been tremendous recent progress in deep neural networks and computer vision, interpreting the results as direct read-outs of experiential states is not advised. According to our data, getting killed appears to make players happy, and much more so than killing enemies, although one might expect the exact opposite. A visual inspection of the data reveals that our classifier works as intended, and our results illustrate the limitations of making inferences based on facial images and discrete emotion labels. For example, players may laugh off the death, in which case the closest label for the facial expression is “happy”, but the true emotional state is complex and ambiguous. On the other hand, players may frown in concentration while killing enemies or escaping a tight spot, which can easily be interpreted as an “angry” expression.
People make the weirdest faces when they play video games, it’s hilarious to watch. :-)
Lisa Feldman Barrett has a bunch of papers / talks / books / etc. about how facial expressions are difficult to interpret. (I read her book How Emotions Are Made (discussed a bit in my post here), and her article Emotional Expressions Reconsidered) She makes a lot of good points in the “Emotional Expressions Reconsidered” article, but I think she takes them too far...
The article brings up a lot of relevant facts, but the way I would explain them is:
1. Labeled emotional concepts like “happiness” that we use in day-to-day life don’t perfectly correspond to exactly one innate reaction, and vice-versa;
2. Our innate subcortical systems create innate facial expressions, but at the same time, our neocortex can also control our face, and it does so in a way that is learned, culturally-dependent, unreliable, and often deceptive. (Hence Paul Ekman’s focus on trying to read “facial microexpressions” rather than reading facial expressions per se.)
3. Most people (including me) seem to be kinda bad at consciously inferring anything about a person’s inner experience based on even the most straightforward and stereotypical facial expressions. (You might think we have a lifetime of experience, but not really. We have almost no experience with seeing faces out of the context of other body motions and knowledge of the situation, and points #1 and #2 also make for a difficult learning task.)
By contrast, the article’s position seems to be something like “facial expressions are completely arbitrary, not innate at all”. Actually, I don’t think she thinks that exactly, but if not, she doesn’t seem to try very hard to avoid giving that impression.
Meta: I’m obviously shoehorning in an old discarded blog post draft outline into the conversation here :-D
There was an old suggestion of making an AI learn human values by training it on happiness and unhappiness in human facial expressions, making it happy when humans are happy and vice versa. Besides its other problems, now there’s this...
People make the weirdest faces when they play video games, it’s hilarious to watch. :-)
Lisa Feldman Barrett has a bunch of papers / talks / books / etc. about how facial expressions are difficult to interpret. (I read her book How Emotions Are Made (discussed a bit in my post here), and her article Emotional Expressions Reconsidered) She makes a lot of good points in the “Emotional Expressions Reconsidered” article, but I think she takes them too far...
The article brings up a lot of relevant facts, but the way I would explain them is:
1. Labeled emotional concepts like “happiness” that we use in day-to-day life don’t perfectly correspond to exactly one innate reaction, and vice-versa;
2. Our innate subcortical systems create innate facial expressions, but at the same time, our neocortex can also control our face, and it does so in a way that is learned, culturally-dependent, unreliable, and often deceptive. (Hence Paul Ekman’s focus on trying to read “facial microexpressions” rather than reading facial expressions per se.)
3. Most people (including me) seem to be kinda bad at consciously inferring anything about a person’s inner experience based on even the most straightforward and stereotypical facial expressions. (You might think we have a lifetime of experience, but not really. We have almost no experience with seeing faces out of the context of other body motions and knowledge of the situation, and points #1 and #2 also make for a difficult learning task.)
By contrast, the article’s position seems to be something like “facial expressions are completely arbitrary, not innate at all”. Actually, I don’t think she thinks that exactly, but if not, she doesn’t seem to try very hard to avoid giving that impression.
Meta: I’m obviously shoehorning in an old discarded blog post draft outline into the conversation here :-D