Bletchley Park Summit & the Quest for AI Safety

On November 1–2, 2023, heads of government and tech company leaders met at Bletchley Park, north of London, for the first ever Artificial Intelligence Safety Summit. The Summit considered the potential of artificial intelligence (AI) both “to transform and enhance human wellbeing, peace and prosperity” and “for serious, even catastrophic harm, either deliberate or unintentional”. The result of the Summit was The Bletchley Declaration, signed by 28 countries plus the European Union.

The Bletchley Declaration commits to engage in cooperative and preventative measures to inform national policy, research, and technology deployment, to address “frontier AI risk” and “to ensure human-centric, trustworthy and responsible AI that is safe, and supports the good of all through existing international fora and other relevant initiatives.

AI systems have the potential to generate massive amounts of new information, much of it relating not to the state of Earth systems or human need, but rooted in the vagaries of AI itself. This displacement of fact-based real-world data is an important under-explored area of AI safety that could impact human security.

This cooperative effort will focus on:

“identifying AI safety risks of shared concern, building a shared scientific and evidence-based understanding of these risks, and sustaining that understanding as capabilities continue to increase, in the context of a wider global approach to understanding the impact of AI in our societies.”
“building respective risk-based policies across our countries to ensure safety in light of such risks, collaborating as appropriate while recognising our approaches may differ based on national circumstances and applicable legal frameworks. This includes, alongside increased transparency by private actors developing frontier AI capabilities, appropriate evaluation metrics, tools for safety testing, and developing relevant public sector capability and scientific research.”

This international cooperative effort will include future AI Safety Summits. It is worth noting, though, the specific mention of “existing international fora”. This is presumably a reference to the G7 and the G20 processes, but may also include security cooperatives, multilateral trade agreements, and negotiation of sharing of information, technology transfer, and data security provisions of emerging international legal decisions on climate, biodiversity, health, space, the internet, food systems, finance, human rights, and other issues of global concern.

The Declaration notes shared concern about “deceptive content” that could result from commercial or political use of AI, or which could be weaponized by a country to disrupt or destabilize a rival nation, or to undermine democracy, economic health and resilience, or peace and security.

The section on frontier AI also notes both the possibility of future targeted AI systems that could cause harm and which “match or exceed the capabilities present in today’s most advanced models”. It also raised the existential risk of “potential intentional misuse or unintended issues of control relating to alignment with human intent”. The signatory countries are effectively recognizing that advanced AI systems might “decide” to take actions that are not aligned with designers’ intent or with the interests of society or the human species more broadly.

Inclusion is repeatedly cited and there is a general agreement in the document that research is needed to better understand both the projected and as-yet-unimagined risks of advanced AI systems. It is not clearly stated, however, that all countries should develop advanced AI capabilities within the boundaries of what is shown to be “safe” or “responsible”. Instead, the Declaration envisions “development-oriented approaches that could help developing countries strengthen AI capacity building”.

The “inclusive global dialogue” and “research on frontier AI safety” appear to embrace AI as an unavoidable technological leap forward. What is evident, though, from the text and the subtext of the Declaration, is that AI systems must be designed to be managed by human beings tasked with preventing any possibility of a breach of the frontier AI safety barrier. It is also clear that AI systems should be designed to support improved human conditions, not to eliminate the need for human workers and not to downgrade the work available to humans to more physical or harrowing work.

Future AI Safety Summits should aim to make real progress in both strict safety protocols and globally recognized legal response measures, to:

Ensure personal data security and personal data sovereignty;
Prevent the use of AI systems to circumvent independent adjudication of legal questions or to undermine the upholding of human rights;
Guard against masking or proliferation of fabricated or manipulated content through generative AI systems;
Ensure AI systems support enhanced access to evidence and accurate information, rather than replacing clear evidence with AI-generated reformulations that are untraceable or obscure key points of evidence;
Explore uses of AI that are narrow and targeted, where the primary function is enhanced processing of complex pools of information, to improve early warning systems and to make early warning information available to more end users in closer to real time;
Assess the risk — and take measures to prevent — AI systems distorting or obstructing the reliable flow of factual scientific observations into relevant data repositories.

We—at Geoversiv, Liberate Data, and Resilience Intel—also see a need to consider at national and international level what kind of resources would be needed to develop data-driven small and medium-sized enterprises (SMEs) that provide critical services at the community level, in countries and communities of all income ranges. The areas where we see such SMEs generating important benefits include:

Effective right-scaling and translation of data for local application;
Support for early warning systems and related local services;
New financial intermediary services that link financial support to better performance on non-financial metrics like climate, nature, and other sustainability imperatives.

Ultimately, advanced AI processing of data, at all scales, can only be genuinely safe if it operates in line with the rights of all people, equal before the law, with their human rights, dignity, and personal informational sovereignty protected. Because of the power advanced AI systems already demonstrate to fabricate alternate realities filled with distorted content, it is critical that we make binding collective decisions now that will ensure AI systems do not undermine human rights, dignity, and sovereignty.

Failing to put those safeguards in place now may open a future in which it is too late to reverse dangerous decisions made by a few irresponsible actors for short-term gain. The Bletchley Declaration is a good start, in terms of recognizing the general atmosphere of potentially catastrophic risk; future AI Safety Summits need to invite, welcome, and promote the voices of stakeholders, including rights groups and those already marginalized or disadvantaged by the digital divide or other technological and wealth imbalances.

Look for more stories related to the future of AI and data systems at LiberateData.net