In a bid to grapple with the formidable challenges posed by cutting-edge artificial intelligence (AI) technologies, the UK government’s Frontier AI Taskforce is forging a path toward the establishment of an AI safety research unit. The mission at hand is clear: to assess the associated risks in the evolving landscape of AI innovation, this was announced this Wednesday October 18 in a press release from the Department for Science, Innovation and Technology.
The Frontier AI Taskforce is positioning itself as a pivotal player in the realm of AI safety. The pioneering group acknowledges that achieving excellence in AI safety research does not necessitate reinventing the wheel or toiling in isolation. This is underscored by its inaugural progress report released on September 7, 2023, revealing the taskforce’s strategic collaborations with prominent technical organizations, including ARC Evals, RAND, and Trail of Bits.
Since the release of the progress report, the taskforce has embarked on new partnerships with an additional three leading technical organizations: Advai, Gryphon Scientific, and Faculty AI. These collaborations are poised to delve into pivotal questions surrounding the capacity of AI systems to enhance human capabilities within specialized domains, while scrutinizing the existing safety safeguards.
The fruits of this research will be shared through informative presentations and engaging roundtable discussions. Participants will include government representatives, civil society groups, leading AI corporations, and esteemed researchers, with the culmination being the AI Safety Summit scheduled for November.
Understanding Frontier AI
Frontier AI is all about the most advanced AI tech. Think super-smart AI that understands languages and solves tough problems. These AI systems have a big impact on our lives.
But, the smarter AI gets, the more it can bring new challenges. For instance, if AI can write computer code really well, it might open doors to more cyberattacks. And if it’s excellent at understanding biology, there could be new risks in medicine and health.
To ensure we use this powerful tech safely, we need experts to check it. These experts must be impartial, meaning they don’t favor one side or company. We don’t want AI companies grading their AI homework!
What the Taskforce is Up To
The Frontier AI Taskforce, now officially called the Frontier AI Taskforce, is quite busy. Here’s a glimpse of their actions:
- Gathering the Brainpower: They’ve brought together a team of experts from different fields like AI research, safety, and national security. These experts help make smart decisions about AI safety. For example, they have folks like Yoshua Bengio, a deep learning guru, and Paul Christiano, an AI alignment expert. They also have experts from national security and the medical community.
- Hiring AI Whizzes: They’re not only focused on AI safety; they’re also beefing up their technical skills. They’ve hired experts like Yarin Gal and David Krueger, who know a lot about AI and machine learning. This way, they have experts on their team who can check if AI systems are safe.
- Teaming Up: They’re not going solo. They’re partnering with organizations like ARC Evals, Trail of Bits, and The Collective Intelligence Project. These partners help them figure out the risks of advanced AI systems.
- Getting Government AI-Ready: They’re making sure government researchers have the same tools and access to AI as top companies. This helps them work on AI safety effectively.
- Moving Quickly: Time is of the essence. They’re hosting the first-ever AI Safety Summit in the UK soon, and they want as many experts and organizations as possible to join in. It’s all about making AI safer and better for everyone.
Taskforce’s latest partners and their respective areas of expertise
1. Advai: This UK-based company has a singular focus on enabling simple, safe, and secure AI adoption. Its technology and research endeavors revolve around the identification of vulnerabilities and constraints within AI systems, with the ultimate goal of fortifying and enhancing these systems. The Frontier AI Taskforce is collaborating with Advai to unearth vulnerabilities within frontier AI systems.
2. Faculty AI: As an applied AI enterprise, Faculty AI offers a suite of services encompassing software, consulting, and more. With nearly a decade of experience in collaborating with the UK government, their projects extend to groundbreaking efforts such as partnering with the NHS to construct an early warning system for COVID-19 and working with the Home Office to combat the dissemination of ISIS online propaganda. The taskforce is joining forces with Faculty AI to evaluate the extent to which Large Language Models (LLMs) can elevate the capabilities of novice actors with malicious intent, as well as the potential risks associated with future systems.
3. Gryphon Scientific: This organization specializes in research and consulting within the realms of physical and life sciences, with a unique focus on public health, biodefense, and homeland security. Drawing from a wealth of experience in scientific advancement, they have worked in collaboration with governments, including the United States and nations in the Middle East and North Africa. Gryphon Scientific’s collaboration with the Frontier AI Taskforce centers on unlocking the potential of Large Language Models (LLMs) as tools for rapid progress in the life sciences.
Today’s announcement builds on the foundation laid out in the September 7 progress report. At that time, the Frontier AI Taskforce unveiled the establishment of an expert advisory panel, the appointment of two research directors, and the initiation of several strategic partnerships with other organizations.
Frontier AI Taskforce Structure
The taskforce is led by Ian Hogarth and Director Ollie Ilott, who play pivotal roles in steering the taskforce’s mission. Ollie’s extensive experience in government positions him as a valuable asset for team-building. The board comprises experts in AI research, safety, and national security, providing a diverse range of perspectives and knowledge. This multidisciplinary approach enhances the taskforce’s capabilities in addressing AI safety concerns. The recruitment of top AI researchers from renowned institutions like Oxford and the University of Cambridge demonstrates the taskforce’s commitment to technical excellence and capacity building. Being part of DSIT ensures the taskforce’s adherence to government standards, accountability, and compliance. It highlights the importance of aligning AI safety research with broader government objectives.
- Leadership:
- Taskforce Chair: Ian Hogarth
- Director: Ollie Ilott
- Expert Advisory Board:
- Yoshua Bengio
- Paul Christiano
- Matt Collins
- Anne Keast-Butler
- Alex van Someren
- Helen Stokes-Lampard
- Matt Clifford (Vice Chair)
- Additional members (to be announced)
- Recruitment of Expert AI Researchers:
- Yarin Gal (Research Director)
- David Krueger (Research Group Leader)
- Other technical AI experts
