The UK’s AI Security Institute is collaborating with several global institutions on a global initiative to ensure artificial intelligence (AI) systems behave in a predictable manner.
The Alignment Project, backed by £15m of government funding, brings together an international coalition including the Canadian AI Safety Institute, Schmidt Sciences, Amazon Web Services (AWS), Anthropic, Halcyon Futures, the Safe AI Fund, UK Research and Innovation, and the Advanced Research and Invention Agency (ARIA).
This reflects a growing global consensus across government, industry, academia and philanthropy that alignment is one of the most urgent technical challenges society faces, and that expanding the field is a shared international responsibility.
In January, the government published its International AI safety report ahead of the AI Action Summit, which took place in Paris on 10-11 February 202. The report notes that AI experts are undecided as to when major societal risks of AI will appear. Some, according to the report’s authors, predict they are decades away, while others think that general-purpose AI could lead to societal-scale harm in the next few years.
“Recent advances in general-purpose AI capabilities – particularly in tests of scientific reasoning and programming – have generated new evidence for potential risks such as AI-enabled hacking and biological attacks, leading one major AI company to increase its assessment of biological risk from its best model from ‘low’ to ‘medium’,” the report’s authors noted.
AI alignment is focused on making sure AI systems behave in the best interest of humanity. Peter Kyle, secretary of state for science, innovation and technology, said this is at the heart of the work the AI Security Institute has been leading since day one, which involves safeguarding the UK’s national security and ensuring the British public are protected from the most serious risks AI could pose as the technology becomes more and more advanced.
Given the pace of AI advancement, the Department for Science, Innovation and Technology (DSIT) notes that today’s methods for controlling AI are likely to be insufficient for tomorrow’s more capable systems as the technology continues to develop, which, it said, is why there is the need for co-ordinated global action to ensure the long-term safety of citizens.
Kyle said: “Advanced AI systems are already exceeding human performance in some areas, so it’s crucial we’re driving forward research to ensure this transformative technology is behaving in our interests.”
The funding includes up to a grant of up to £1m for researchers across disciplines from computer sciences to cognitive science. It also provides dedicated compute resources from AWS and Anthropic and access to investment from private funders to accelerate commercial alignment.
Geoffrey Irving, chief scientist at the AI Security Institute, said: “AI alignment is one of the most urgent and under-resourced challenges of our time. Progress is essential, but it’s not happening fast enough relative to the rapid pace of AI development.
“Misaligned, highly capable systems could act in ways beyond our ability to control, with profound global implications. By providing funding, compute resources and interdisciplinary collaboration to bring more ideas to bear on the problem, we hope to increase the chance that transformative AI systems serve humanity reliably, safely and in ways we can trust.”
