BIAS: Responsible AI for Labour Market Equality

What do we study? BIAS is an interdisciplinary project to understand and tackle the role of AI algorithms in shaping ethnic and gender inequalities in the labour market, which is now increasingly digitized. The project seeks to understand and minimise gender and ethnic biases in the AI-driven labour market processes of job advertising, hiring and professional networking. We further aim to develop 'responsible' AI that mitigates biases and attendant inequalities, by designing AI algorithms and development protocols that are sensitive to such biases. The empirical context of our investigation includes these labour market processes in organisations and on digital job platforms.

Why is it important? Labour market inequalities need to be tackled because they deny and thwart equitable and sustainable socio-economic development. In both the UK and Canada, access and rewards to work remain patterned around social distinctions, like gender, race, and ethnicity. The deployment of AI in labour market processes is known to exacerbate such inequalities through perpetuation of existing gender and ethnic biases hiring and career progression. From the point of view of policy our proposal speaks directly to the following priorities in both countries-the UK's Industrial Strategy, which has 'putting the UK at the forefront of the AI and data revolution' as one of its grand challenges; the UK's AI sector deal that aims to 'boost the UK's global position as a leader in developing AI technologies'; and the Canadian SSHRC's goal of tackling persistent demographic (ethnic and gender) disparities in workforce selection and development.

Why is it unique? Although we know that AI can exacerbate biases in the labour market, we do not know how AI can mitigate these biases. The project develops responsible and trustworthy AI that reduces labour market inequalities by tackling gender and ethnic/racial biases in job advertising, hiring and professional networking processes. It enhances capacity development for responsible AI applications through training of early career researchers, builds on existing and develops new UK-Canada research partnerships, and develops outputs for multiple stakeholders (researchers, companies and policy units). It speaks to multiple objectives of the funding call.

Why is it intellectually original and challenging? The project is interdisciplinary. It integrates and cuts across three distinct streams of research (the first two are from the social sciences and the third from the computational and mathematical sciences) to tackle the research objective through an interdisciplinary approach. The first includes studies on socio-economic antecedents of labour market inequality, which underscore the persistence and prominence of gender and ethnic/racial inequalities in the UK and Canada. The second includes studies from business management (digitalisation, technology/AI adoption and human resource management). It emphasizes that while we know that the use of AI in the labour market processes of job advertising, hiring and professional networking can strengthen ethnic and gender bias in these processes we do not know what those biases are and how AI algorithms can mitigate them, as opposed to merely (re)producing them. The third draws on computational statistics (Bayesian statistic/machine learning) to design new AI algorithms and development protocols that integrate human and machine inputs/outputs.

What is the work plan? Our project comprises two interlinked work packages that respectively (1) understand the different dimensions of bias from a multi-stakeholder perspective (e.g. employer, employee, digital platform developer) through in-depth data mining and qualitative investigations when AI algorithms are used in the labour market processes of job advertising, hiring and professional networking; and (2) test/design new AI algorithms to mitigate them and create protocols for their development and implementation.

Potential 'biases' produced by AI technologies may significantly undermine labour market equality and stymy equitable and sustainable socio-economic development. BIAS's objectives speak directly to multiple national priority agendas in both the UK and Canada - gender pay gap, ethnic/racial disparity, digital and industrial strategy. As both the UK and Canada look to embrace digital transformations as part of their national (economic and industrial) strategies, our focus on the implications of such transformations for labour market equalities and our objective to reduce such inequalities through the responsible development and deployment of AI promises a broad range of impacts, which are pertinent to the future of labour relations, economic competitiveness, human resource management, and industrial strategies.

Who are the beneficiaries?
We expect our project to achieve impact on and through at least six stakeholder groups: (1) data science developers, computer scientists and technicians involved in the designing, testing and maintenance of AI algorithms and platforms for job advertising, hiring and professional networking, (2) international, national and professional associations responsible for regulating the design and deployment of AI and for skill/labour management, (3) platforms (e.g. LinkedIn) that use AI to broker labour market processes, (4) employers that use AI to facilitate hiring, retention and renumeration, (5) job seekers and employees; (6) the general public who are interested in and undergoing digital transformations in society.

How will they benefit?
The stakeholders will benefit from our co-production of open-source AI development protocol and packages, which promises to set new industrial standards for the workflow of AI development that is sensible to 'biases' and accountable for gender and ethnic inequalities. The project will thus inform policy and regulation developments in the design and use of AI. Our findings will also inform platform developers and owners to operate AI-driven labour market platforms in a transparent and explainable manner to ensure egalitarian labour market opportunities and outcomes. In doing so, employers will be able to combine gender and ethnic equality liability into efficient hiring and human resources management practice in their use of AI platforms. A good understanding of potential AI biases and inequality implications will also empower individual job seekers and employees to devise personalised strategies as they engage with AI. Digital transformations in the labour market and more broadly in society concern all citizens. Our project therefore contributes to citizen digital education in soliciting public opinions to input into AI design and in making the AI workflow explainable to the general public.

What are our pathways to impact?
Our anticipated impact will be achieved through a number of channels. First, we will co-produce with research participants (e.g. public sector organisations, governmental departments, the Cabinet Office, platform owners, employers, individual employees, AI developers) and influential industrial stakeholders (e.g. Output, Profusion) an open-source AI development protocol. We have already secured the named stakeholders to form a learning network to take part in the project. Throughout the life of the project, four (international) knowledge exchange workshops will be hosted for the research team and stakeholders to engage with one another. We will also produce regular newsletters, media engagements, vlogs, policy reports, and a short documentary. Members of the project team will work closely with the Cabinet Office through links such as the Lancaster-Cabinet Office Innovation Partnership to partake in the government consultations and development of national strategies. Project members will also speak at AI developer conferences to job fairs felicitate impact among frontline AI practitioners and users.