By Emma Walker, Andre Barrinha and Julian Padget.
Artificial intelligence is developing rapidly, bringing significant opportunities alongside new risks. As governments, researchers and industry consider how best to govern these technologies, ensuring AI is developed and used safely, ethically and inclusively has become an increasingly urgent challenge.
These issues were at the centre of the United Nations Global Dialogue on AI Governance, held in Geneva in July, where IDSB Deputy Director Dr Julian Padget represented the Institute following a written submission highlighting why effective AI governance matters. Drawing on those discussions, and the expertise of researchers from across the Institute, this blog explores why safe and responsible AI matters more than ever.
Real-world risks are already emerging
In recent years, we have seen examples where potential harms related to AI systems have not been identified or addressed before technologies have been deployed. For example, Grok AI was used to create demeaning images of others, primarily targeting women, with action only being taken following widespread backlash. Generative AI has also been used to influence public opinion on a range of issues through the creation and circulation of fake information or to fraudulently extract money or information from individuals and organisations via targeted family emergency scams and CEO deepfake scams. Large language models were found to expose users to extremist content and politicised communities during conversational interactions, essentially bypassing safeguards designed to reduce potential risks of harm.
And it is not just the way that we communicate and interact online that is being impacted. Generative AI also presents a cybersecurity risk, aiding the development of ransomware and even enabling sophisticated cyberattacks. Although some of these techniques existed before the widespread rollout of AI, the added realism, ease of access and ease of creation that generative AI provides means that the scale and impact of these threats have grown substantially.
The need for collaboration and governance
There are examples of industry collaboration to reduce AI-related risks in specific areas, as seen in the Anthropic-led initiative Project Glasswing. However, these approaches must involve a wider range of voices, including government, civil society and academia. This will enable a fuller consideration of potential impacts at the individual, group and societal levels.
Most importantly, governance approaches need to be interoperable across regions, cultures, providers, deployers and harm types. They should also provide sufficient clarity to minimise so-called ‘avoision’ behaviours where organisations may attempt to operate in the grey area between avoidance and evasion. This is vital if potential gaps between these various areas, and the risks they create, are to be avoided.
Although there can be political reluctance to regulate AI to avoid stifling innovation, we need to consider how these challenges can be overcome most effectively. For instance, is it feasible for individual regulators to oversee AI use within their respective sectors rather than implementing overarching, sector-agnostic regulation?
It’s not enough for technologies with the potential to transform society, law and order to be addressed only through isolated use cases, disparate initiatives, or self-regulatory approaches that are not fully transparent, open or collaborative in approach.
Building a safer AI future
Fundamentally, AI’s potential to support complex decision-making, assist with task completion, and augment human capability can only be realised if it is developed in line with effective ethical and responsible innovation practices. This is currently hindered by uncertainties regarding appropriate governance mechanisms across multiple areas of application and the need for governance frameworks that are sufficiently flexible and rapidly adaptable, to accommodate new capabilities and novel forms of (mis)use.
Going forward, it is crucial that appropriate mechanisms for ethical and responsible governance of both AI development and AI use, including how these mechanisms may change and evolve, are developed and adapted collaboratively over time with a full range of stakeholders. This includes industry partners, academic researchers, policy specialists, end-user communities and those representing the natural world and other non-human systems that may be affected by AI.
One thing is clear, the future impact of AI will depend on the choices we make today. By prioritising responsible innovation, appropriate governance and ongoing collaboration, we can ensure that AI serves as a force for good - enhancing our capabilities while protecting the wellbeing of individuals and society as a whole.
At the Bath Institute for Digital Security and Behaviour, we continue to develop our work in this key area. With expertise spanning policy, computer science, psychology and management, the IDSB takes a multidisciplinary approach to understanding AI that has much to contribute to global discussions on this topic.
