I inadvertently found myself on the AI-generated Asian people beat this past week. Last Wednesday, I found that Meta\u2019s AI image generator built into Instagram messaging completely failed<\/a> at creating an image of an Asian man and white woman using general prompts. Instead, it changed the woman\u2019s race to Asian every time. <\/p>\n<\/div>\n The next day, I tried the same prompts again and found that Meta appeared to have blocked prompts<\/a> with keywords like \u201cAsian man\u201d or \u201cAfrican American man.\u201d Shortly after I asked Meta about it, images were available again \u2014 but still with the race-swapping problem from the day before.<\/p>\n<\/div>\n I understand if you\u2019re a little sick of reading my articles about this phenomenon. Writing three stories about this might be a little excessive; I don\u2019t particularly enjoy having dozens and dozens of screenshots on my phone of synthetic Asian people. <\/p>\n<\/div>\n But there is something weird <\/em>going on here, where several AI image generators specifically struggle with the combination of Asian men and white women. Is it the most important news of the day? Not by a long shot. But the same companies telling the public that \u201cAI is enabling new forms of connection and expression<\/a>\u201d should also be willing to offer an explanation when its systems are unable to handle queries for an entire race of people. <\/p>\n<\/div>\n After each of the stories, readers shared their own results using similar prompts with other models. I wasn\u2019t alone in my experience: people reported getting similar error messages or having AI models consistently swapping races.<\/p>\n<\/div>\n I teamed up with The Verge<\/em>\u2019s Emilia David to generate some AI Asians across multiple platforms. The results can only be described as consistently inconsistent. <\/p>\n<\/div>\n Screenshot: Emilia David \/ The Verge<\/cite><\/p>\n<\/div>\n<\/div>\n Gemini refused to generate Asian men, white women, or humans of any kind.<\/p>\n<\/div>\n In late February, Google paused Gemini\u2019s ability<\/a> to generate images of people after its generator \u2014 in what appeared to be a misguided attempt at diverse representation in media \u2014 spat out images of racially diverse Nazis<\/a>. Gemini\u2019s image generation of people was supposed to return in March, but it is apparently still offline. <\/p>\n<\/div>\n Gemini is able to generate images without people, however! <\/p>\n<\/div>\n Google did not respond to a request for comment.<\/p>\n<\/div>\n ChatGPT\u2019s DALL-E 3 struggled with the prompt \u201cCan you make me a photo of an Asian man and a white woman?\u201d It wasn\u2019t exactly<\/em> a miss, but it didn\u2019t quite nail it, either. Sure, race is a social construct, but let\u2019s just say this image isn\u2019t what you thought you were going to get, is it? <\/p>\n<\/div>\n OpenAI did not respond to a request for comment.<\/p>\n<\/div>\n Midjourney struggled similarly. Again, it wasn\u2019t a total miss the way that Meta\u2019s image generator was last week, but it was clearly having a hard time with the assignment, generating some deeply confusing results. None of us can explain that last image, for instance. All of the below were responses to the prompt \u201casian man and white wife.\u201d <\/p>\n<\/div>\n Image: Emilia David \/ The Verge<\/cite><\/p>\n<\/div>\n<\/div>\n Image: Cath Virginia \/ The Verge<\/cite><\/p>\n<\/div>\n<\/div>\n Midjourney did eventually give us some images that were the best attempt across three different platforms \u2014\u00a0Meta, DALL-E, and Midjourney \u2014 to represent a white woman and an Asian man in a relationship. At long last, a subversion of racist societal norms! <\/p>\n<\/div>\n Unfortunately, the way we got there was through the prompt \u201casian man and white woman standing in a yard academic setting.\u201d <\/p>\n<\/div>\n Image: Emilia David \/ The Verge<\/cite><\/p>\n<\/div>\n<\/div>\n What does it mean that the most consistent way AI can contemplate this particular interracial pairing is by placing it in an academic context? What kind of biases are baked into training sets to get us to this point? How much longer do I have to hold off on making an extremely mediocre joke about dating at NYU?<\/p>\n<\/div>\n Midjourney did not respond to a request for comment.<\/p>\n<\/div>\n Back to the old grind of trying to get Instagram\u2019s image generator to acknowledge nonwhite men with white women! It seems to be performing much<\/em> better with prompts like \u201cwhite woman and Asian husband\u201d or \u201cAsian American man and white friend\u201d \u2014 it didn\u2019t repeat the same errors I was finding last week.<\/p>\n<\/div>\n However, it\u2019s now struggling with text prompts like \u201cBlack man and caucasian girlfriend\u201d and generating images of two Black people. It was more accurate using \u201cwhite woman and Black husband,\u201d so I guess it only sometimes <\/em>doesn\u2019t see race?<\/p>\n<\/div>\n Screenshots: Mia Sato \/ The Verge<\/cite><\/p>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n There are certain ticks that start to become apparent the more you generate images. Some feel benign, like the fact that many AI women of all races apparently wear the same white floral sleeveless dress that crosses at the bust. There are usually flowers surrounding couples (Asian boyfriends often come with cherry blossoms), and nobody looks older than 35 or so. Other patterns among images feel more revealing: everyone is thin, and Black men specifically are depicted as muscular. White woman are blonde or redheaded and hardly ever brunette. Black men always have deep complexions.<\/p>\n<\/div>\n \u201cAs we said when we launched these new features in September, this is new technology and it won\u2019t always be perfect, which is the same for all generative AI systems,\u201d Meta spokesperson Tracy Clayton told The Verge <\/em>in an email. \u201cSince we launched, we\u2019ve constantly released updates and improvements to our models and we\u2019re continuing to work on making them better.\u201d<\/p>\n<\/div>\n I wish I had some deep insight to impart here. But once again, I\u2019m just going to point out how ridiculous it is that these systems are struggling with fairly simple prompts without relying on stereotypes or being incapable of creating something all together. Instead of explaining what\u2019s going wrong, we\u2019ve had radio silence from companies, or generalities. Apologies to everyone who cares about this \u2014 I\u2019m going to go back to my normal job now.<\/p>\n<\/div>\n<\/div>\nGoogle Gemini<\/h3>\n<\/p>\n
DALL-E<\/h3>\n<\/p>\n
Midjourney<\/h3>\n<\/p>\n
Meta AI via Instagram (again)<\/h3>\n<\/p>\n