[dandanqu82]

As someone from a less represented group in tech, I would want to know exactly who is classifying emotions present on the dataset. I don’t want to be mistaken as feeling angry anytime I don’t look happy or sad. There’s definitely unconscious bias to how certain people’s feelings are interpreted that lead to inaccurate perceptions.

[randyrand]

Note: as long as you are well represented in the dataset your representation in tech won’t matter.

[jakelazaroff]

Historically, these two things have been correlated.

[dandanqu82]

I can be well represented in the data set, but here people can unwittingly classify my emotions incorrectly due to unconscious bias. What good will the data do if every time, I appear confused, I am labeled and angry?

[FooBarBizBazz]

The biases of labelers are tricky, because that work gets outsourced to low-wage countries.

[dandanqu82]

That makes things worse, but we would still have a problem with the work being done primarily by us tech workers. I wouldn’t place the blame solely on unskilled low-wage workers. I’ve worked with Ivy League graduates that think I am angry when I ask a question and am just confused. The US education system is moderately segregated and many of our most educated students are not taught to be more critical of stereotypes perpetuated in the media.

[coffeefirst]

Agreed. This all smells like snake oil to me. If they've gotten it to work at all, there's no chance it works for all people in all circumstances.