"As is typical for AI systems, DALL-E 2 has inherited biases from the corpus of data used to train it. That means for all the delightful images that DALL-E 2 has produced, it’s also capable of generating a lot of images that are not delightful ... here’s what the AI gives you if you ask it for an image of lawyers <pictures of men>"
"just like those previous experiments, DALL-E suffers from the same racist and sexist bias AI ethicists have been warning about for years. Machine learning systems almost universally exhibit bias against women and people of color, and DALL-E is no different"
Here's OpenAI themselves on DALL-E bias:
"Use of DALL·E 2 has the potential to harm individuals and groups by reinforcing stereotypes, erasing or denigrating them, providing them with disparately low quality performance, or by subjecting them to indignity"
Their primary complaint is that the pictures DALL-E draws (i.e. symbols of reality) don't reflect their social agendas. They worry about this a lot because they fear that if DALL-E draws men when asked to draw lawyers, and women when asked to draw flight attendants, that this by itself will actually make the drawings come true.
"Also, how is the most prominent US propaganda that was all around movies - that Russians are the enemy, red scare, etc — leftist?"
That's hardly the most common US propaganda in movies, is it, and almost certainly is a legacy of the cold war. It's much more profitable to annoy the Russian market than the other very large communist country that might be depicted as a generic enemy of the state.
At any rate very little US output features Russians as the enemies compared to movies/TV shows that depict e.g. women being able to kung-fu fight off a group of 10 men 3x heavier than themselves, or which depict gay relationships as being far more prevalent than they actually are, or in which the Queen of England is black. That's the kind of "inclusivity" bias that's being discussed here.
https://www.vox.com/future-perfect/23023538/ai-dalle-2-opena...
"As is typical for AI systems, DALL-E 2 has inherited biases from the corpus of data used to train it. That means for all the delightful images that DALL-E 2 has produced, it’s also capable of generating a lot of images that are not delightful ... here’s what the AI gives you if you ask it for an image of lawyers <pictures of men>"
Or: https://www.vice.com/en/article/wxdawn/the-ai-that-draws-wha...
"just like those previous experiments, DALL-E suffers from the same racist and sexist bias AI ethicists have been warning about for years. Machine learning systems almost universally exhibit bias against women and people of color, and DALL-E is no different"
Here's OpenAI themselves on DALL-E bias:
"Use of DALL·E 2 has the potential to harm individuals and groups by reinforcing stereotypes, erasing or denigrating them, providing them with disparately low quality performance, or by subjecting them to indignity"
Their primary complaint is that the pictures DALL-E draws (i.e. symbols of reality) don't reflect their social agendas. They worry about this a lot because they fear that if DALL-E draws men when asked to draw lawyers, and women when asked to draw flight attendants, that this by itself will actually make the drawings come true.
"Also, how is the most prominent US propaganda that was all around movies - that Russians are the enemy, red scare, etc — leftist?"
That's hardly the most common US propaganda in movies, is it, and almost certainly is a legacy of the cold war. It's much more profitable to annoy the Russian market than the other very large communist country that might be depicted as a generic enemy of the state.
At any rate very little US output features Russians as the enemies compared to movies/TV shows that depict e.g. women being able to kung-fu fight off a group of 10 men 3x heavier than themselves, or which depict gay relationships as being far more prevalent than they actually are, or in which the Queen of England is black. That's the kind of "inclusivity" bias that's being discussed here.