Tencent’s AI accidentally transforms black people into animals instead of anime characters
The Trust Project is a worldwide group of news organizations working to establish transparency standards.
Tencent’s Different Dimension Me (DDM) is a recently introduced AI tool that can turn photos of people into anime characters. The platform was designed for Chinese users but soon became popular in other regions where anime is loved, such as South America. However, users soon realized that the platform could not identify blacks and plus-size people, groups that are conspicuously absent from Japanese anime, resulting in offensive AI-generated results.
Unfortunately, AI didn’t always produce the most flattering results. In some cases, the AI-generated images were racially offensive, such as a black man with exaggerated lips or a white woman with excessively large breasts.
This issue was likely due to a lack of data for the AI to train on. Japanese anime does not tend to feature blacks and huge people as main characters, meaning that there was no accurate data for the AI to learn from. As a result, the AI-generated results were inaccurate and offensive.
Tencent has since apologized for the offensive results and has pledged to improve AI so that it can better identify different types of people. In the meantime, users are advised to be careful when using the platform and to check the results before sharing them publicly.
About Chinese neural networks
They have their own neural networks there, and they are quite pumped and completely independent of the achievements of Google, Nvidia, OpenAI, StabilityAI, and other Midjourney companies.
- Baidu, which has established itself in search engines and has stepped up its activity in the field of autonomous driving in recent years, uses ERNIE-ViLG, a model with 10 billion parameters trained on a dataset of 145 million Chinese image and text pairs.
- Different Dimension Me (DDM) by Tencent can turn photos of people into anime characters. Aimed at Chinese users, it suddenly became popular in other anime-loving regions.
- Another Chinese text-to-image model is Taiyi, the brainchild of IDEA, a research lab led by Harry Shum, who co-founded the largest research division of Microsoft Research Asia. The open-source model is trained on 20 million filtered Chinese image-text pairs and has one billion parameters.
It is curious to see how the Chinese will regulate all this disgrace because they have more experience in “regulation” than the rest of the world. And while AI rebels are crowdfunding money to create “Stable Diffusion without Censorship,” the Chinese are calmly honing algorithms for regulating neural networks in the interests of the state and not focusing on the feelings of believers.
Read more about AI:
Any data, text, or other content on this page is provided as general market information and not as investment advice. Past performance is not necessarily an indicator of future results.