The UK Launches a New State-Backed AI Safety Platform ‘Inspect’
The UK has taken a pioneering step in AI safety with the release of ‘Inspect,’ a state-backed platform designed to evaluate AI models. This groundbreaking toolset is now available globally as an open-source resource, fostering collaboration and ensuring safer AI development.
We open-sourced Inspect, our framework for large language model evaluations. We're excited to see the research community use and build upon this work! https://t.co/tRVCEnKL20
— AI Safety Institute (@AISafetyInst) May 10, 2024
Overview of Inspect Platform
The AI Safety Institute, a UK-based organisation, announced the global release of ‘Inspect’ on 10 May. The platform aims to bolster AI safety evaluations worldwide, providing a consistent framework for assessing AI models.
Inspect is a comprehensive software library allowing users to test AI models’ capabilities, including core knowledge, reasoning abilities, and autonomous functions. By standardising evaluations, Inspect ensures that AI models are robust and secure.
Open Source Accessibility
The platform’s release under an open-source MIT licence marks a significant milestone. This accessibility allows a broad range of users—from startups to government bodies—to integrate and enhance AI safety tools, ensuring widespread adoption and improvement.
Michelle Donelan, Secretary of State for Science, Innovation, and Technology, emphasised the UK’s leadership role: “This puts UK ingenuity at the heart of the global effort to make AI safe.” She highlighted the extraordinary rewards of managing AI risks, benefiting the healthcare and transportation sectors.
The UK AI Safety Institute collaborates with the Incubator for AI (i.AI) and Number 10 to gather top AI talent for developing new safety tools. Ian Hogarth, Chair of the AI Safety Institute, expressed pride in the open sourcing-of Inspect, aiming to build a shared, accessible approach to AI safety evaluations.
This initiative aligns with international efforts, including the UK-US partnership announced last month, which focuses on developing advanced AI testing methods. The collaboration aims to establish a global standard for AI safety, reflecting commitments made at the AI Safety Summit.
Amanda Brock, CEO of OpenUK, stressed the importance of the UK’s position in AI safety. She noted that the success of Inspect would be measured by its adoption and efficacy in the AI community, with global attention now on future safety summits.
Conclusion
The launch of Inspect signifies a crucial advancement in AI safety, positioning the UK as a leader in this field. The UK is paving the way for safer and more reliable AI innovations worldwide by providing a robust, open-source platform for AI evaluations.
