UK ministers propose an “alarm system” for risky AI

Technology Secretary Michelle Donelan strives to establish early warning signals for technology through a safety summit

UK ministers have issued a warning, underscoring the need for a “smoke alarm” system to detect dangerous AI, with the goal of averting severe threats like widespread casualties, cyberattacks, and uncontrolled AI technology. Technology Secretary Michelle Donelan expressed optimism that an upcoming safety summit in the UK would pave the way for an early warning mechanism. This mechanism would necessitate tech companies to proactively identify risks in their AI products and have plans in place to address them. Donelan emphasized, “We need a system comparable to a smoke alarm. Companies should not only detect potential risks but also have a plan to mitigate them. Such a comprehensive system should be universally implemented.

Discussing the upcoming two-day summit scheduled for November at Bletchley Park, Donelan stressed the importance of understanding the risks associated with AI, despite its incredible opportunities. The government announced that the summit would primarily focus on two key areas: the misuse of AI for creating bioweapons or carrying out cyber-attacks, and the challenge of managing highly advanced AI systems.

Reportedly, Rishi Sunak believes that there is limited time to reach a global consensus on defining the most crucial risks associated with AI and formulating strategies to tackle them. This sense of urgency is driven by technology companies harnessing enhanced computing power, technological breakthroughs, and heightened investments to develop more powerful AI models.

The summit is not expected to yield an international treaty comparable to those governing nuclear weapons in the context of AI development. Instead, the aim is to outline the range of significant risks posed by AI systems and propose measures for their mitigation. The primary emphasis will be on “frontier” AI models, denoting state-of-the-art systems whose capabilities rival or exceed the most advanced existing models, potentially posing threats to human life.

The government affirmed that the summit seeks to identify the current state and potential trajectory of the frontier in AI development. Taking place at Bletchley Park in Buckinghamshire, a site historically notable as the workplace of codebreakers such as Alan Turing during World War II, the event will bring together global leaders, AI firms, academics, and civil society groups.

As per a government statement issued on Monday, there is apprehension that models considerably more potent than the current ones, like OpenAI’s ChatGPT powered by the GPT-4 model, could become available soon. These models exhibit capabilities that are exceedingly difficult to foresee, even for their developers, and there is an inherent risk of them being accessible to various entities, including those with malicious intent.

A key concern among AI experts is the prospect of advanced systems surpassing human control. This worry revolves around the potential attainment of artificial general intelligence, denoting AI with intelligence levels equal to or surpassing that of humans, and the theoretical capacity of such systems to evade any safeguards in place.

In a preview of the upcoming summit on Monday, the government’s statement underscored the attention given to systems capable of eluding control. It addressed the risks associated with the “loss of control” arising from advanced systems, stressing the significance of aligning these systems with our values and intentions.