In the realm of artificial intelligence, Google’s Gemini, a multimodal generative AI tool, has recently encountered a temporary setback. The tool’s ability to depict people was paused due to its tendency to produce historically incongruous images. For instance, it portrayed the U.S. Founding Fathers as a diverse group rather than solely white men.

Google promptly addressed the issue and provided insights into the underlying cause. Google’s founder, Demis Hassabis, explained that the problem arose from the tool’s inability to recognize when users sought “universal depictions.” He emphasized the nuances associated with advanced AI and the challenges of catering to a global audience with diverse backgrounds and contexts.

The intention behind Gemini’s diversity feature was well-intended, but its blunt application across all prompts led to the incongruous results. Hassabis assured that prompts requesting content about historical figures would be handled with greater accuracy in the future. He expressed optimism that the feature would be restored within the next few weeks.

The conversation then shifted to the broader concern of preventing the misuse of generative AI tools by malicious actors, such as authoritarian regimes seeking to spread propaganda. Hassabis acknowledged the complexity of the issue, calling for a collective effort involving civil society, governments, and tech companies. He stressed the need for open dialogue to define the values and limitations of these systems and to prevent their repurposing for harmful purposes.

Hassabis also touched upon the challenge of open-source, general-purpose AI models, highlighting the desire for users to have full control over the systems they employ. However, he raised concerns about the potential risks as these systems become more powerful. He urged society to address these issues proactively, considering the consequences of widespread proliferation and the potential exploitation by bad actors.

Finally, Hassabis shared his thoughts on the future of AI devices in light of generative AI’s rapid advancements. He envisioned a wave of “next-generation smart assistants” that would be genuinely useful in people’s daily lives, moving beyond the gimmicky nature of previous generations. He speculated that the mobile hardware people carry might evolve to accommodate these AI systems, potentially leading to new form factors beyond the traditional smartphone.