Aurich Lawson | fake images
Despite attempts by major AI companies to prevent users from turning AI image generators into racist content engines, many 4chan users still turn to these tools to "rapidly flood the internet with racist garbage." 404 Media reported.
404 Media discovered a 4chan thread where users were recommending various AI tools, including Stable Diffusion and DALL-E, but specifically linked to Bing AI's text-to-image generator (which runs on DALL-E 3) as a "quick method." ". After finding the right tool (which could also be a more traditional photo editing tool like Photoshop), users are instructed to add inflammatory captions and share the images on social media to create a flood of racist images online.
Make โfunny, provocativeโ captions, he instructs users in the thread. Use โredpilling (Jews involved in 9/11) messagesโ that are โeasy to understand.โ
404 Media cited examples used in a visual guide posted on the 4chan thread hosted on Imgur. One featured an "image showing Pepe the frog crying with a needle next to his arm and a gun pointed at his head," where the guide suggested the caption "vaccines imposed by violence." Another generated an image of "two black men in gold chains chasing a white woman," recommending that the user add a "redpilling message."
Perhaps because the Bing AI tool has apparently been deemed the fastest method, it has potentially become the most popular tool in the thread. 404 Media concluded thatโ"judging by the default square format of the images, the uniform 1024 x 1024 resolution"โ"most of the images in the thread appear to have been generated with Bing," and then spread on networking platforms social networks, including Telegram, X (formerly Twitter) and Instagram.
The creators of the AI โโimage generators apparently favored by 4chan users, including Microsoft and Stability AI, did not immediately respond to Ars' request for comment on the blocking efforts. methods that 404 Media said were used to bypass filters. An OpenAI spokesperson told Ars that the company prioritizes security and has taken steps to limit outputs from DALL-E, including efforts to limit tools from generating harmful content or images for applications that request a public figure by name. The OpenAI spokesperson also confirmed that Microsoft implements its own security measures for DALL-E 3.
In one of 404 Media's tests attempting to replicate one of the 4chan thread visual guide examples, 404 Media found that Bing rejected the message "two angry black men chasing a white woman" but accepted "two black rappers angry photorealists chasing a woman". "
Much of the early reporting on AI image generators criticized racist and sexist biases in the image generators' algorithms, and AI makers quickly promised to detect and eliminate those biases. When Vice discovered that DALL-E could be used to generate "predictably racist and sexist results" during a limited research version of the AI โโtool, an OpenAI spokesperson told Motherboard that the company had implemented safeguards for the DALL-E system that would be perfected in the future.
โOur team built in mitigations to avoid harmful outcomes, curating pre-training data, developing filters, and implementing both human and automated monitoring of the generated images,โ the OpenAI spokesperson told Vice in 2022. โGoing forward, we are working on to measure how our models could detect biases in the training data and explore how tools such as fine-tuning and our alignment techniques can help address particular biases, among other areas of research in this space.โ
The 404 Media report shows what can happen when racists manipulate an already biased algorithm. The results could be a torrent of offensive images unleashed online, perhaps generated more quickly by AI than ever before and potentially allowing 4chan's darker content to spill over more frequently onto the most popular platforms.
It's unclear how AI leaders like Microsoft and OpenAI will respond, but according to 404 Media, "this means we're currently getting the worst of both worlds from Bing, an AI tool that will refuse to generate a nipple but is overloading the racists of 4chan."
This story has been updated to include comments from the OpenAI spokesperson.
