By Tom Blitzer
Key Insights:
- The UK AI Safety Institute is expanding to San Francisco to tap into Bay Area tech talent and enhance global AI safety efforts.
- The new US office strengthens UK-US partnerships, building on the recent AI Safety Summit’s global collaboration initiatives.
- Safety tests reveal AI models’ strengths in basic cybersecurity but highlight vulnerabilities and the need for human supervision in complex tasks.
The United Kingdom’s AI Safety Institute is extending its reach with a new office in San Francisco. Announced by Michelle Donelan, the UK Technology Secretary, on May 20, the institute’s first international branch is set to open in the summer. This move aims to leverage the rich pool of tech talent available in the Bay Area and to foster stronger ties with one of the world’s largest AI hubs located between London and San Francisco.
The San Francisco office is expected to help the UK institute build and solidify relationships with key AI players in the US. The expansion is seen as a crucial step towards promoting global AI safety standards and furthering the institute’s mission of ensuring AI developments serve the public interest.
London Branch and AI Safety Expertise
The London branch of the AI Safety Institute currently employs 30 professionals focused on AI risk assessment. This team is on a path to growth, aiming to enhance its expertise, particularly in evaluating the risks associated with advanced AI models. The expansion to San Francisco is anticipated to provide additional resources and insights, thereby strengthening the institute’s overall capabilities.
Michelle Donelan highlighted that this expansion signifies the UK’s leadership and proactive approach to AI safety. She emphasized that it marks a significant moment for the UK to analyze AI from a global perspective and to enhance its partnerships, particularly with the US. This move is expected to pave the way for other nations to engage with the UK’s AI safety initiatives.
Recent AI Safety Summit and Global Collaboration
The announcement of the new office follows the UK’s AI Safety Summit held in London in November 2023. The summit, a pioneering event focused on AI safety, gathered global leaders and key figures from the AI industry, including representatives from the US and China. Notable attendees included Microsoft president Brad Smith, OpenAI CEO Sam Altman, Google DeepMind CEO Demis Hassabis, and entrepreneur Elon Musk.
This summit underscored the importance of international collaboration in addressing AI safety challenges. The event set the stage for ongoing discussions and cooperative efforts aimed at mitigating AI risks and ensuring the technology benefits society at large.
AI Model Safety Testing Results
In conjunction with the expansion announcement, the UK AI Safety Institute released results from recent safety tests conducted on five publicly available advanced AI models. The models were anonymized, and the results provided a snapshot of their capabilities rather than labeling them as “safe” or “unsafe.”
The findings revealed that several models could successfully tackle basic cybersecurity challenges, while others faced difficulties with more complex tasks. Some models exhibited PhD-level knowledge in fields such as chemistry and biology. However, all tested models were found to be highly vulnerable to basic jailbreak techniques and struggled to complete more intricate, time-consuming tasks without human intervention.
Ian Hogarth, the chair of the institute, stated that these assessments are crucial for developing a comprehensive understanding of AI model capabilities. He acknowledged that AI safety is an emerging field, and the results represent a fraction of the evaluation approach the institute is developing.
Geoffrey Hinton Advocates for Universal Basic Income
In related news, Geoffrey Hinton, a prominent AI expert known as the “Godfather of AI,” recently advised the UK government on the potential need for a universal basic income (UBI). Hinton, who previously worked for Google, emphasized the risk of job losses due to AI-driven automation and suggested that a UBI could mitigate the economic impact on workers displaced by technological advancements.
Hinton’s consultations with Downing Street officials reflect growing concerns about AI’s societal implications. He joined other AI leaders, such as OpenAI co-founder Sam Altman, in advocating for UBI as a solution to potential job displacement. Altman’s venture, Worldcoin, aims to provide UBI through a cryptocurrency token system, emphasizing the importance of economic measures to support those affected by AI automation.
Both Hinton and Altman have expressed concerns about the broader existential risks posed by AI. Hinton, in particular, left his position at Google to discuss these issues more freely. He warned that within five to twenty years, humanity might face significant challenges from AI systems attempting to gain control.