The why, what where and when of AI regulation

There’s no denying that Artificial Intelligence (AI) is the hot ticket in technology right now. The technology isn’t new - it has been hard at work in fields as diverse as credit scoring, recruitment search and self-driving cars (to name but a few) for well over a decade - but over the past few months, and most famously with the release of ChatGPT in November 2022, AI technology has taken a big leap forward both in its capabilities and in public awareness. 

AI presents many potential benefits and opportunities. However, many have warned of the associated risks. Professor Steven Hawking famously warned that: “unless we learn how to prepare for, and avoid, the potential risks, AI could be the worst event in the history of our civilisation.” It is now becoming increasingly important to develop appropriate safeguards.

Governments around the world have been working to address the issues, although the progress of legislation is much slower than the progress of technology. China has implemented some regulations (albeit limited in scope at present) and more general AI focused regulations anticipated in the EU and Brazil this year. Other jurisdictions seem less keen to legislate on AI yet, but there are plenty of discussions taking place about what controls are needed, and pre-legislation frameworks and guidance have been published in countries including the UK, USA, Australia, Canada, Japan and Singapore.

What are the risks?

?       Safety: the main concern is that we have no idea whether AI will always prioritise human safety and wellbeing.  The potential for malfunctions of dangerous equipment present risks to human life, the distortion of human behaviour via illicit manipulation present risks to the stability of political regimes, and there is even the existential threat of an AI system having so much power that it could wipe out humanity entirely. Many of the proposed laws address this with a human-centric risk-based approach.

?       Privacy and security: AI systems can collect and process large amounts of personal data, which could be used to violate an individual's privacy or endanger security. For example, AI-powered devices in our homes and workplaces can capture all manner of information about us, which could be misused or subject to a security breach. Existing regulations address these risks in many jurisdictions and would apply to AI as it would to any technology, but such protection is not yet available worldwide.

?       Bias and discrimination: Much like a human, if AI is trained on information that includes bias, the AI will assimilate that bias. We are already seeing this in, for example, facial recognition software that fails to recognise black faces, and reports of adverts and articles that include pictures of women in sportswear being suppressed by social media algorithms, apparently because they were identified as ‘racy,’ when a male equivalent received no such treatment.

?       Accountability: It can be difficult to determine who is responsible for the actions of an AI system. A complex system with many contributors, a huge range of training materials, and often one or more human operators, will have many possible points of failure. It may be impossible to identify who was at fault if something goes wrong, for example when a self-driving car causes an accident. In the EU at least, specific regulations will likely be created to determine where liability will sit in specified scenarios.