Efficiency savings. Productivity gains. Headcount reduction. This is the language commonly used to describe the proposed economic benefits artificial intelligence (AI) will bring.
This commercial lens shapes how we view the possible impact of AI on our lives and the investment decisions firms make around the deployment of AI systems and infrastructures, where rate of return on investment frames the conversation.
However, as has become clear, the environmental and social impacts of these technologies are just as pertinent, if not more so, than the economic. Discussion has increasingly turned towards the energy requirements of AI infrastructure and the environmental impacts of data centres, whilst the use of AI systems has been linked to emergent social harms (such as the erosion of civic institutions, cognitive offloading and even episodes of what has been termed ‘AI-induced psychosis’).
The trade-offs of deploying AI are now coming more clearly into view, presaging the need for forms of social impact measurement and assessment that can render the complex social consequences of AI legible to policymakers and communities. This is essential not only for highlighting the potential social harms of AI, but also for encouraging a more responsible deployment of AI that can better deliver benefits to the public.
The limits of existing approaches to social value
This is particularly relevant in the UK, where the use of AI by the public sector is proceeding apace. The UK has long seen the use of social impact frameworks within the context of public sector procurement. Since the UK Public Services (Social Value) Act 2012, the public sector has been required to consider the social value of procurement decisions, referring to the additional social and environmental benefits that might be procured through the tendering of a contract. Oftentimes, this will take the form of measures that the contractor might provide to boost the wellbeing of local communities, such as providing local employment, local educational opportunities or committing to sustainable business practices. These will then usually be given a financial value and expressed as a ratio to demonstrate the social return on investment.
However, as argued by the Ada Lovelace Institute, this kind of approach is limited when it comes to assessing the social value of AI. By treating social value as separate ‘goods’ offered within a contract, existing frameworks can struggle to account for the social outcomes of the use of AI technologies themselves.
The unique challenge AI poses
AI presents unique challenges to social assessment due to its constantly evolving nature, the often-opaque quality of AI systems, the general lack of AI literacy beyond specialist communities, issues around algorithmic fairness and biases, and the yet-uncharted implications of long-term use of and interaction with AI systems and infrastructures. The potential for significant ‘unintended consequences’ emergent from AI thus presents challenges to the rigorous assessment of its social impact over the mid- to long-term.
For example, a council might procure an AI chatbot for frontline services, and the contractor might commit to produce local jobs to provide positive social value. Furthermore, the council might record cost-savings through use of the chatbot, thus freeing up council resources for deployment elsewhere. However, the social impacts on communities that result from using the chatbot are more difficult to account for. This is because such impacts may be less tangible, more subjective or bound by specific cultural differences, and may change over time. For instance, assessing the affective impacts of speaking to a chatbot on diverse user communities, and how this might impact trust in, and the legitimacy of, council services over time is difficult to achieve within existing frameworks.
The need for alternative approaches
So, we are presented with a challenge and an opportunity. Whilst social value is widely understood and accepted in the UK, existing approaches struggle to make legible the specific impacts of AI systems and infrastructures themselves. This calls for supplementing existing approaches with alternative methodologies. Fortunately, there is a diverse field of scholarship dedicated to the measurement of social impact that presents a range of approaches for doing so. The challenge is bringing these into dialogue with the particularities of AI systems and infrastructures.
BRAID will take the first step on this journey by cataloguing the diverse approaches to assessing the social impact of AI and applying this work to the unique challenges of responsible AI use.
The research will draw lessons from different methodologies for measuring social impact and apply these to the use of AI in the UK context, particularly in the areas of local government procurement, the planning and buildout of AI data centres, and the development of an AI skills and literacy pipeline. In particular, by engaging with participatory approaches to social impact, we will seek to foreground the experiences and voices of affected communities. Doing so will provide novel perspectives on how AI might be deployed responsibly for the public good.
Why do this now? We recognise the value of building on and improving best practice around assessing the social value of AI technologies, with a clear emphasis on the less easily captured impacts of AI systems and infrastructures on user communities. With public sector use of AI growing rapidly, it has never been more important to ensure that AI investment is driving responsible practice and delivers positive social outcomes for the public.
Dr Darcy Luke
Research Associate, BRAID
Image credit: Deborah Lupton / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/
