Written submission from Dr Selcen Ozturkcan and Dr Inci Toral (AIB0018)
Epistemic Dispossession and Worker Agency: Governing AI Systems in the Future of Work
Dr. Selcen Ozturkcan, Linnaeus University, Sweden & Sabanci University, Turkey (ORCID: 0000-0003-2248-0802)
Dr. Inci Toral, University of Birmingham, UK (ORCID: 0000-0002-4656-7139)
This submission looks at how AI affects the future of work by focusing on digital systems, rules made by algorithms, and worker freedom. Instead of viewing AI as a standalone tool, this study bases its analysis on design choices, platform rules, and governance setups that influence economic and knowledge outcomes (Ozturkcan, forthcoming; Yeung, 2018). The authors bring complementary expertise to this submission. Dr. Selcen Ozturkcan, an Associate Professor at Linnaeus University and a Network Professor at Sabanci University with an extensive research portfolio covering service robots, the impact of AI in marketing, and human-robot collaboration. Dr. Inci Toral, an Associate Professor in the Department of Marketing at the University of Birmingham and a Principle Fellow of the Higher Education Academy, contributes through her expertise in digital marketing and GenAI use in services industries. Her academic focus includes digital marketing, retailing, and the employee perspective on AI integration in service industries.
This submission makes three main, policy-focused arguments:
(1) Current AI systems, including those that influence knowledge production and workplace decisions, must be governed to prevent the dispossession of worker and creator agency.
(2) Harmful outcomes are not inevitable, but require mandatory regulatory intervention and human oversight to mitigate negative effects resulting from specific design and governance choices; and
(3) The UK must establish participatory governance structures that involve affected communities and share decision-making power to ensure worker protection, industry strength, and a fair AI transition (Ozturkcan, forthcoming; Costanza-Chock, 2020).
AI systems are not merely neutral tools in the job market; they embody the objectives and regulations of the companies that employ them (Gillespie, 2018; Couldry & Mejias, 2019). Research indicates that AI systems frequently wrest control from creators, marginalise workers, and obscure accountability. The shift toward 'zero-click' AI marks a significant transformation in how we access information online (Ozturkcan, forthcoming). These systems deliver answers directly on platforms without revealing their sources. In knowledge-based professions, these systems amalgamate various expert inputs into a single response, creating the illusion that the platform itself is the expert. This accomplishes three things: it retains users on the platform, conceals the true providers of information, and discourages users from seeking alternative sources (Mackenzie, 2024; Ozturkcan, forthcoming). The economy depends on individuals such as teachers and researchers who generate and disseminate knowledge online. When AI utilises their work without attribution, these creators lose recognition and opportunities, while platforms gain control over what is deemed knowledge and who receives credit (Ozturkcan, forthcoming; Fricker, 2007). This phenomenon poses a direct threat to the sustainability of the UK's knowledge economy by diminishing the economic value of expert labour.
However, this harm is inevitable. This occurs when three conditions coincide: (i) answers fail to clearly indicate their sources, (ii) new content diverts attention from original authors, and (iii) ranking and financial rules hinder fair exchange and debate (Ozturkcan, forthcoming). Policies can address these issues in several ways: by consistently giving credit rather than occasionally mandating clear citations for created content, enabling users to trace information back to experts, and distributing benefits so that knowledge workers are rewarded for their contributions (Ozturkcan, forthcoming).
Currently, AI systems are making more decisions in the job market, such as hiring and assigning tasks. Studies have shown that these systems make people feel that they have more freedom, but actually limit their real choices (Ozturkcan, forthcoming; Beer, 2017). In hiring and scheduling, these systems focus on efficiency and predictability, not on what workers want (Fourcade & Healy, 2017; Yeung, 2018; Ozturkcan, forthcoming). These systems are often unclear, so workers believe that they have choices, such as applying for jobs or changing schedules. However, their choices are limited by how the system ranks and filters options (Bucher, 2018; Yeung, 2018). This is particularly evident in the UK gig economy, where algorithmic assignment and performance ratings significantly control workers' income and opportunities, often resulting in worker misclassification and limited recourse against managerial decisions.
This is important because the loss of the ability to make choices affects workers’ feelings about their jobs. Studies have shown that when workers allow AI to make decisions, they become less aware of other options and rely more on the AI's decisions. The distinction between helping and replacing workers has become unclear (Yeung, 2018).
In addition, these systems reflect the values of their creators (Gillespie, 2014; Verbeek, 2005). They are designed to meet company goals, such as efficiency and productivity, not worker needs, such as fair pay and skill growth (Ozturkcan, forthcoming). This means that AI often favours employers over workers (O'Neil, 2016; Pasquale, 2015).
Another dimension of worker agency and the challenges of AI integration is explored from the employee perspective in service industries (Akter et al., 2023). Research reveals that workers often prioritise immediate operational consistency over the long-term innovative potential of AI solutions. Initial findings from this context suggest that AI co-workers are not always fully trusted because of the potential for failures and machine injuries (Toral et al., 2025). Furthermore, the high costs of AI investments, including maintenance and updates, suggest that firms may need to revert to human employees during economic downturns, highlighting a conditional dependence on AI that is subject to economic realities. This perspective demonstrates that worker decisions are often shaped by a prioritisation of reliability and cost-effectiveness, offering a ground-level counterpoint to the organizational metrics of optimisation and efficiency that typically drive algorithmic design.
Policy intervention requires four essential actions: (1) Organisations employing algorithmic workforce systems must disclose their methods for ranking, optimising, and providing feedback (Burrell, 2016; Yeung, 2018). (2) Evaluations should be conducted to determine how these systems affect worker autonomy, job quality, and job security (Mittelstadt et al. 2016). (3) Workers must be actively involved in the development, testing, and management of systems that influence their employment (Ozturkcan, forthcoming). (4) Regulations should be established to ensure human involvement in decisions regarding worker opportunities, compensation, and job stability (Sunstein, 2015; Yeung, 2018).
AI has different effects on various workers. Research indicates that automation and algorithms often disproportionately disadvantage women, people of colour, migrants, disabled individuals, and those in precarious employment. These groups already contend with job and wage disparities (Ozturkcan, forthcoming; Noble, 2018). The UK's existing Equality Act (2010) framework must be extended to cover algorithmic decision-making to address this problem.
Several factors contribute to this situation. First, AI systems learn from historical data, which inherently contain existing biases and job inequalities, perpetuating unfair patterns (Barocas, Hardt, & Narayanan, 2019; Ozturkcan, forthcoming). Second, AI applications in hiring and job evaluations can exacerbate these biases by disseminating discrimination more broadly (Caliskan, Bryson, & Narayanan, 2017; Noble, 2018). Third, demand for new skills predominantly affects those with limited resources. Individuals with unstable jobs cannot afford extended periods without employment or training.
Such workers are frequently excluded from the development or management of AI systems, which influence their employment. Involving these communities in AI-related decisions can uncover the risks and solutions that experts might overlook (Costanza-Chock, 2020; Ozturkcan, forthcoming).
To address these challenges, four actions are essential: (1) mandate assessments of AI's impact on employment, with a focus on women, people of colour, migrants, disabled individuals, and those in precarious jobs (Ozturkcan, forthcoming); (2) establish governance structures that empower affected workers to have a voice in AI applications within their industries (Costanza-Chock, 2020); (3) invest in training and support for workers in sectors experiencing significant job losses, ensuring accessibility and inclusivity; and (4) enforce regulations that hold organisations accountable when AI leads to discrimination, providing remedies for affected workers (Noble, 2018; Pasquale, 2015).
The effects mentioned are happening in real life. In retail and logistics, systems using algorithms manage workers' schedules and tasks with little transparency (Ozturkcan, forthcoming). In jobs such as ride-hailing, delivery, and gig services, algorithms control how workers are managed (Kellogg, Valentine, & Wolff, 2020; Rahman, 2021). The UK's transport and logistics sector, for instance, faces particular challenges as algorithmic management blurs the lines of employee supervision. In customer service, back-office support, and knowledge work, AI systems are replacing human workers without adequate support for adjustment (Ozturkcan, forthcoming; Acemoglu & Johnson, 2023).
What is missing is public and regulatory infrastructure adequate to manage this transition. Policy discussions lack several critical components (Yeung, 2018; Pasquale, 2015; Ozturkcan, forthcoming):
● Mandatory disclosure requirements for algorithmic employment practices (Burrell, 2016; Yeung, 2018)
● Independent capacity to audit and assess algorithmic systems in employment contexts (Pasquale, 2015)
● Sectoral governance frameworks that allow workers, unions, and communities to participate in decisions about AI deployment (Ozturkcan, forthcoming)
● Public investment in transition support tied to actual labor displacement (Acemoglu & Johnson, 2023)
● Effective enforcement mechanisms with sufficient deterrent effects. Current regulatory frameworks impose penalties too small to deter harmful practices (Pasquale, 2015)
We should focus on building this system. The steps to take include: setting up a public agency to check algorithm-based hiring systems, similar to past ideas for algorithm checks (Pasquale, 2015); forming groups in different sectors that include workers (Ozturkcan, forthcoming); supporting independent studies on how AI affects jobs, broken down by demographic group (Ozturkcan, forthcoming); and changing job laws to require checks on algorithm impacts and involve workers in decisions (Yeung, 2018).
The harms identified in this submission are real and measurable. They are also conditional rather than inevitable (Ozturkcan, forthcoming). The following criteria offer a framework for distinguishing acceptable AI deployments from practices that should be prohibited or reformed:
Attribution and Transparency: AI systems in employment and knowledge work should operate with attribution by default, visible provenance mechanisms, and transparent ranking criteria (Ozturkcan, forthcoming). Opacity should trigger regulatory intervention (Pasquale, 2015; Yeung, 2018).
Worker Participation: Algorithmic systems that directly affect employment should be designed, tested, and governed by substantive worker participation (Ozturkcan, forthcoming; Costanza-Chock, 2020). Consultation without veto power was insufficient.
Intersectional Equity: Before deployment, organisations should conduct intersectional impact assessments that project the effects on workers in different demographic categories (Ozturkcan, forthcoming). Systems predicted to exacerbate existing inequalities should not be deployed without mitigation measures (Barocas et al., 2019; Noble, 2018).
Deliberation Preservation: Algorithmic systems should be designed to preserve human deliberation in decisions affecting worker prospects, compensation, and the continuity of employment (Ozturkcan, forthcoming; Sunstein, 2015). Optimisation should never override worker choice (Yeung, 2018).
Accountability: Organisations deploying algorithmic employment systems should be accountable for documented harm, with remedies available to affected workers (Pasquale, 2015). Penalties should be sufficient to deter noncompliance (Yeung, 2018).
To ensure AI deployment is worker-centric and promotes inclusive growth within the UK, we urge the Committee to adopt the following specific policy recommendations:
The future work on AI is not set in stone. It depends on choices, rules, and policies made the next year or two (Ozturkcan, forthcoming; Verbeek, 2005). Evidence shows that current trends towards more secretive, efficient, and independent AI systems can harm workers, especially those already at risk in the job market (Ozturkcan, forthcoming; Yeung, 2018; Acemoglu & Johnson, 2023).
These harms can be prevented. Policy tools are available, such as required transparency, shared decision-making, impact checks considering different groups, worker power to reject AI use, public funding for job changes, and strong enforcement (Ozturkcan, forthcoming; Yeung, 2018; Costanza-Chock, 2020). What is needed is the political will to steer AI development toward outcomes that focus on worker control, industry strength, and inclusive growth.
● Acemoglu, D., & Johnson, S. (2023). Power and progress: Our thousand-year struggle over technology and prosperity. PublicAffairs.
● Akter, S., Hossain, M. A., Sajib, S., Sultana, S., Rahman, M., Vrontis, D., & McCarthy, G. (2023). A framework for AI-powered service innovation capability: Review and agenda for future research. Technovation, 125, 102768.
● Barocas, S., Hardt, M., & Narayanan, A. (2019). Fairness and machine learning. fairmlbook.org.
● Beer, D. (2017). The social power of algorithms. Information, Communication & Society, 20(1), 1–13.
● Bucher, T. (2018). If... then: Algorithmic culture. Polity Press.
● Burrell, J. (2016). How the machine 'thinks': Understanding opacity in machine learning algorithms. Big Data & Society, 3(1), 1–12.
● Caliskan, A., Bryson, J. J., & Narayanan, A. (2017). Semantics derived automatically from language corpora contain human-like biases. Science, 356(6334), 183–186.
● Costanza-Chock, S. (2020). Design justice: Community-led practices to build the worlds we need. MIT Press.
● Couldry, N., & Mejias, U. A. (2019). The costs of connection: How data is colonizing human life and appropriating it for capitalism. Stanford University Press.
● Fourcade, M., & Healy, K. (2017). Seeing like a market. Socio-Economic Review, 15(1), 9–29.
● Fricker, M. (2007). Epistemic injustice: Power and the ethics of knowing. Oxford University Press.
● Gillespie, T. (2014). The relevance of algorithms. In T. Gillespie, P. J. Boczkowski, & K. A. Foot (Eds.), Media technologies: Essays on communication, distribution and difference (pp. 167–193). MIT Press.
● Gillespie, T. (2018). Custodians of the internet: Platforms, content moderation, and the hidden decisions that shape social media. Yale University Press.
● Kellogg, K. C., Valentine, M. A., & Wolff, A. (2020). Algorithms at work: The new contested terrain of control. Academy of Management Annals, 14(1), 366–410.
● Mackenzie, A. (2024). Machine learners: Archaeology of a data practice. MIT Press.
● Noble, S. U. (2018). Algorithms of oppression: How search engines reinforce racism. NYU Press.
● O'Neil, C. (2016). Weapons of math destruction: How big data increases inequality and threatens democracy. Crown Publishing.
● Ozturkcan, S. (forthcoming). Zero-click AI, epistemic injustice and the governance of digital knowledge infrastructures. Discover Artificial Intelligence.
● Ozturkcan, S. (forthcoming). Robots do not arrive empty-handed: A call for reflexive and inclusive HRI. In M. F. Özbilgin, C. Erbil, & D. Groutsis (Eds.), Methods for researching global challenges: An interdisciplinary guide. Routledge.
● Ozturkcan, S. (forthcoming). Choice without choosing: Algorithmic marketing and the illusion of consumer autonomy. In P. M. Pardalos, D. Skandali, & M. Tsioufis (Eds.), Springer Handbook on AI and Optimization in Marketing. Springer.
● Pasquale, F. (2015). The black box society: The secret algorithms that control money and information. Harvard University Press.
● Rahman, K. S. (2021). The algorithmic state. Georgetown Law Review, 109(2), 1–87.
● Sunstein, C. R. (2015). Choosing not to choose: Understanding the value of choice. Oxford University Press.
● Toral, I., De Kervenoael, R., & Peronard, J.-P. D. C. (2025). Navigating AI Integration in Service Industries: Employee Perspectives on their AI Co-workers. Abstract from Recent Advances in Retailing and Consumer Sciences Conference , Zagreb, Croatia.
● Verbeek, P.-P. (2005). What things do: Philosophical reflections on technology, agency, and design. Pennsylvania State University Press.
● Yeung, K. (2018). Hypernudges: Private digital choice architecture and public regulation. Regulation & Governance, 12(1), 3–23.