Safety of advanced AI under the scientific spotlight

The first iteration of the ‘International Scientific Report on the Safety of Advanced AI’ was recently published, the development of the report being one of the key commitments to emerge from the Bletchley Park discussions in November, coming as part of the landmark Bletchley Declaration, of which Australia is a signatory.

The report was formulated by a diverse global team of AI experts, including an Expert Advisory Panel from 30 leading AI nations from around the world, as well as representatives of the UN and the EU, to bring together the best existing scientific research on AI capabilities and risks. The report aims to give policymakers across the globe a single source of information to inform their approaches to AI safety.

The report recognises that advanced AI can be used to boost wellbeing, prosperity and new scientific breakthroughs — many of which have already been seen in fields including health care and drug discovery, and in how we can tackle climate change. But it notes that like all powerful technologies, current and future developments could result in harm. For example, malicious actors can use AI to spark large-scale disinformation campaigns, fraud and scams. Future advances in advanced AI could also pose wider risks including labour market disruption and economic power imbalances and inequalities.

However, the report highlights a lack of universal agreement among AI experts on a range of topics, including both the state of current AI capabilities and how these could evolve over time. It also explores the differing opinions on the likelihood of extreme risks which could impact society such as large-scale unemployment, AI-enabled terrorism and a loss of control over the technology. With broad expert agreement highlighting that we need to prioritise improving our understanding, the future decisions of societies and governments will ultimately have an enormous impact.

“AI is the defining technology challenge of our time, but I have always been clear that ensuring its safe development is a shared global issue,” said UK Secretary of State for Science, Innovation, and Technology, Michelle Donelan. “When I commissioned Professor Bengio to produce this report last year, I was clear it had to reflect the enormous importance of international cooperation to build a scientific evidence-based understanding of advanced AI risks. This is exactly what the report does.”

The interim publication is focused on advanced general-purpose AI. This includes state-of-the-art AI systems that can produce text and images, and make automated decisions. The final report is expected to be published in time for the AI Action Summit which is due to be hosted by France, but will now take on evidence from industry, civil society and a wide range of representatives from the AI community. This feedback will mean the report will keep pace with the technology’s development, being updated to reflect the latest research and expanding on a range of other areas to ensure a comprehensive view of advanced AI risks.

This report summarises the existing scientific evidence on AI safety to date, and the work led by a broad group of scientists and panel members from 30 nations, the EU and the UN over the past six months will now help inform the next chapter of discussions of policymakers at the AI Seoul Summit and beyond,” said Bengio, International Scientific Report on the Safety of Advanced AI Chair. “When used, developed and regulated responsibly, AI has incredible potential to be a force for positive transformative change in almost every aspect of our lives. However, because of the magnitude of impacts, the dual use and the uncertainty of future trajectories, it is incumbent on all of us to work together to mitigate the associated risks in order to be able to fully reap these benefits.”

“The Interim International AI Safety Report is brave and insightful. It sets a clear direction for future AI safety research, acknowledging where experts disagree,” said Prof Bronwyn Fox, Chief Scientist, CSIRO. “I commend the report to policymakers seeking to inform their responses to AI with rigorous, evidence-based science. I look forward to Australia elevating the voices and experience of First Nations peoples in future reports.”

A final edition of the report is expected to be released ahead of the next round of discussions on AI safety, which will be hosted by France.

The report can be found here.

Image credit: iStock.com/Peach_iStock