The controversy highlights the ongoing struggle of AI researchers to balance the need for diversity and accuracy in AI outputs. Companies like OpenAI have previously faced similar issues and have developed techniques to insert diversity into image-generation prompts in a way that is hidden from the user. However, these techniques have also led to awkward and sometimes controversial results. As a solution, some experts suggest the need for a diverse set of AI assistants that reflect the diversity of languages, culture, value systems, and political opinions across the world.
Key takeaways:
- Google has paused its Gemini AI image-synthesis feature due to criticism that it was inserting diversity into its images in a historically inaccurate way, such as depicting multi-racial Nazis and medieval British kings with unlikely nationalities.
- OpenAI had previously invented a technique to insert terms reflecting diversity into image-generation prompts in a way that was hidden from the user, a technique that Google's Gemini system seems to use as well.
- The controversy reflects the ongoing struggle in which AI researchers find themselves stuck in the middle of ideological and cultural battles online, with different factions demanding different results from AI products.
- Google could potentially fix the issue by modifying its system instructions to avoid inserting diversity when the prompt involves a historical subject, and is currently working on improving these kinds of depictions.