Google’s multimodal generative AI tool, Gemini, is expected to resume its capability to generate images of people in the next few weeks, according to Demis Hassabis, the founder of DeepMind. The tool’s ability to produce historically incongruous images, such as depicting the U.S. Founding Fathers as a diverse group, led to its suspension last week.
Key Takeaway
Google’s AI tool, Gemini, is set to address the historical-image diversity issue and resume generating images of people in the coming weeks, following its suspension due to inaccuracies.
Addressing the Issue
During an interview at the Mobile World Congress in Barcelona, Hassabis acknowledged the technical glitch and attributed it to the failure of Google to identify instances where users sought a “universal depiction.” He emphasized the complexity of advanced AI and the need to cater to a diverse global audience.
Resolution and Future Measures
Hassabis assured that the feature promoting diversity in Gemini’s image outputs of people will be refined to ensure historical accuracy. He mentioned that the tool will be back online in the next few weeks, with a more targeted approach to prompts for historical figures.
Preventing Misuse and Future Developments
When questioned about preventing the misuse of generative AI tools, Hassabis highlighted the need for extensive research and societal collaboration to establish and enforce ethical boundaries. He also discussed the challenges associated with open source AI models and the potential implications as these systems become more powerful.
Future of AI Devices
Hassabis also shared his vision for the future of AI devices, predicting the emergence of next-generation smart assistants that could potentially reshape the mobile hardware landscape. He raised the possibility of alternative form factors, such as glasses, to enhance the contextual awareness of AI systems.
