The 2023 AI Safety Summit

Introduction

On 1st - 2nd November 2023, the UK hosted the Safety Summit on Artificial Intelligence (AI) at Bletchley Park, Milton Keynes, with approximately 150 representatives from across the globe in attendance. The two-day event was the first of its kind, and the UK government is confident that the outcomes of the conference will “underpin the international response” towards the regulation of AI.

The overarching purpose of the summit was to facilitate open discussion between countries, allowing each state to express its own concerns regarding developments in frontier AI. Frontier AI refers to highly prolific general-purpose systems, capable of performing a wide range of functions. This approach towards the development of AI demonstrates that the issue is a global problem that requires collaborative action.

During the first day, the countries in attendance agreed to the Bletchley Declaration. This was followed by an agreement made on the second day to support the ‘State of Science’ Report, led by Professor Yoshua Benigo of the University of Montreal.

What is the Bletchley Declaration?

The Bletchley Declaration on AI safety, an agreement reached by all 28 countries in attendance, recognised the mutual consensus regarding the opportunities and risks posed by AI. Moreover, the Declaration represents the countries’ agreement to collaboration and transparency. The shared ambition intends to properly realise and utilise the power of frontier AI to positively transform economies and society. Thus, by regulating the development of frontier AI at a global level, the risks associated with the technology can also be mitigated.

Some of the risks identified in the Declaration include cybersecurity, biotechnology and misinformation. The UK government acknowledged that these risks are global, further affirming the demand for international cooperation.

The Bletchley Declaration is the first formal agreement of its kind. It delivers a clear and unequivocal message: the countries in agreement are committed to regulating frontier AI in tandem. This recognises that it is unfeasible to expect regulations in one country to be adequate if they are developed without consideration for other rules abroad.

What is the ‘State of Science’ Report and who is Yoshua Benigo?

On 2nd November, representatives of the 28 countries in attendance also agreed to support a ‘State of Science’ Report – an independent and inclusive review led by Yoshua Benigo. Benigo is well-versed in the opportunities and risks associated with AI, having won the Association for Computing Machinery (AM) Turning Award in 2018 for his pioneering work on deep learning.

During an interview, Benigo acknowledged the imminency of the risk posed by frontier AI – rapid technological advancements mean that the associated problems will arise much sooner than anticipated. Furthermore, he argues that a passive approach and waiting for a Treaty to be signed is highly unlikely to resolve the issues effectively. By taking the pragmatic course of action through Benigo’s report (and in unison with the Bletchley Declaration) countries can engage in proactive discussions to create remedies with a global impact.

Benigo himself recognises that it is unreasonable to expect all countries to give effect to identical policies in their countries, as there will be discrepancies in particular values between nations. However, he does believe there are “basic principles of safety” that can be agreed upon, and ultimately come into fruition, via open discussion.

By declaring their support for the report, the countries reaffirmed their commitment to evidence-based collaboration, as per the Bletchley Declaration. The report also extends the scope of the discussion to world-leading AI academics and support provided by an international Expert Advisory Panel.

Conclusion

The Bletchley Summit has set the tone for the future of AI and its regulation. With an agreement being reached and support for an inclusive report being confirmed, the UK has taken the first significant step towards the global harmonisation of managing AI. The conference is by no means the end of discussions – the UK has encouraged all parties to continue to consider new, practical regulations ahead of subsequent summits. The Republic of Korea has agreed to co-host a virtual summit, with France hosting an in-person event in 2024.

By Alexander McLean