Slashdot Mirror
Can Machine Learning Guess True Emotions From Facial Microexpressions? (cmu.edu)
jbmartin6 writes: Microexpressions are fast, involuntary facial expressions which other people may not consciously recognize, but arise from our real emotions instead of the face we wish to present to the world. Carnegie Mellon University released an interesting blog entry about new approaches to using computers to recognize these microexpressions with a focus on the security and military applications. If you haven't taped over the cameras on your devices, it might be time to start thinking about it. Just imagine how advertisers would (mis)use this sort of technology.
"Our approach uses machine learning features that treat the whole face as a canvas," writes the lead researcher, adding "One challenge we faced for this project was finding a dataset with accurately labeled data to establish ground truth.
"Few existing databases capture subjects' suppressed reactions...."
"Our approach uses machine learning features that treat the whole face as a canvas," writes the lead researcher, adding "One challenge we faced for this project was finding a dataset with accurately labeled data to establish ground truth.
"Few existing databases capture subjects' suppressed reactions...."
Sounds high to me too. Now, how I am going to kill 9 minutes and 59 seconds.
Your ad here. Ask me how!
Here's the thing about emotions. They don't mean what you'd expect. Emotions ARE logic to an extent - short-circuited logic, quicker and made with incomplete information. They're the quick-and-dirty half-logic that makes us able to function in a world of unknowns for millions of years as a species.
They aren't 'serious business' - they're formed and dismissed in fractions of a second, and most of them don't trigger facial emotions - and even then a raised eyebrow muscle can mean confusion, pronounced indifference, mild surprise, or just positioning to move your eyeball, or many other things.
Like most things, this ain't mindreading - it's just more polygraph logic being applied, ironically like emotion itself is applied - for the basis of 'asperational truth', or bullshit they hope might pan out.
Ryan Fenton
... in some cases, it will return an answer of, "pissed off gorilla."
It little behooves the best of us to comment on the rest of us.
Judging from what I saw at NRF in New York City earlier this week the answer is a definite no. Not only did the systems I saw fail to guess my facial emotion but they also decided that I was female. I'm a 53 your old man with short hair and a goatee. It kind of makes you wonder what kind of faces they used to program in what a female face would look like...
I don't think that we are far off but that conference tends to have some pretty decent cutting-edge systems to show off and we are clearly not there yet.
I started getting into ANN and ML recently, just as a hobby. My finding thus far is that the AI we are hearing so much about is nothing more than a sophisticated fuzzy pattern matcher. That's literally all it does: it "learns" by mapping patterns to a limited range of predetermined outputs. The path taken to determine what input should be mapped to what output is reinforced by feeding the machine millions of inputs (and providing the correct output to adjust the weights taken by the path).
It's sophisticated, computationally expensive, and hilariously nowhere near "intelligence". After seeing how these machines (the ANN) functions I'm pretty certain that we're nowhere close enough to intelligent machines. Sure, we can train a particular NN to beat humans at Go, but the first time we put it in front of another game (recognising pictures) it will lose. Give it enough time and error-correction and it will eventually win that too, but the problem is that there are so many "general" tasks that humans and other animals accomplish with much less training that the machine can never hope to keep up.
I'm a minority race. Save your vitriol for white people.