The CEO of Google expressed concern over some responses from the company’s Gemini artificial intelligence model, calling them “unlikely” and pointing out issues such as depicting German World War II soldiers as people of color. He described this bias as “totally unacceptable.”
In a memo to employees, Sundar Pichai acknowledged that images and text generated by modern AI tools were causing discomfort.
Social media users highlighted instances where Gemini image generators depicted historical figures of different ethnicities and genders, including the Pope, the Founding Fathers, and Vikings. Google suspended Gemini’s ability to create people images in response.
One example involved Gemini’s chatbot responding to a question about negative social impacts, leading to a discussion about Elon Musk and Hitler. Pichai addressed this issue, calling the responses upsetting and indicative of bigotry.
Pichai stated that Google’s teams were working to improve these issues and have already made significant progress. AI systems often generate biased responses due to training data issues, reflecting larger societal problems.
Gemini’s competitors are also working on addressing bias in AI models. New versions of AI generators like Dall-E prioritize diverse representation and aim to mitigate technical issues.
Google is committed to making structural changes and enhancing product guidelines to address biases. Pichai emphasized the importance of providing accurate and unbiased information to users.
Elon Musk criticized Google’s AI programs, pointing out the bias in generated images. Technology commentator Ben Thompson called for a shift in decision-making at Google to prioritize good product development.
The emergence of generative AI platforms like OpenAI’s ChatGPT presents a competitive landscape in AI development. Google’s Gemini AI chatbot, formerly known as Bard, offers paid subscriptions for enhanced AI capabilities.
Google DeepMind continues to innovate in AI, with breakthroughs like the AlphaFold program for predicting protein structures. The CEO of DeepMind acknowledged the need to improve diversity in AI-generated images.
Source: www.theguardian.com
