Human Automation Digital Leadership, Workforce Transition and New Rights for Human-Machine Work
Who stays accountale when an AI does the work?
Cassandra Gardiner, digital transformation leader, former senior global director at Cable & Wireless, and researcher in technology, organisational governance, workforce policy and vocational education, opened our last WIDForum’s Thematic Working Group 3 (TWG3) webinar with this question. Her answer is the concept she calls Human Automation: emerging relationship between the governance, design and development of human capability alongside increasingly autonomous digital systems.
“As artificial intelligence becomes more capable, how do we ensure humans become more capable too, not less?”
Beyond governing AI
The dominant conversation around AI governance asks how we build better, safer, more trustworthy AI systems. Human Automation asks instead what should be automated, who remains accountable when something goes wrong, and how do we prepare people to work responsibly alongside systems they did not design and are increasingly expected to trust.
Gardiner describes AI as becoming part of the digital fabric of organisations and society, a transformation layer that increasingly makes recommendations that shape decisions and sets priorities in real time. This layer influences who receives care, which job applications are shortlisted, or how a child protection case is prioritised, and prompts immediately the question of human capability in technical governance.
What meaningful oversight requires
The EU AI Act requires meaningful human oversight for high-risk AI systems. Gardiner’s argument is that oversight is a professional capability and that workers receiving only basic AI awareness training cannot exercise it genuinely, regardless of what the system’s interface asks them to confirm.
Meaningful oversight requires people who can understand AI outputs, question recommendations, recognise the limitations and potential biases of the models they are working with, apply professional judgment, and intervene when something is wrong. That is a workforce challenge as much as a regulatory one.
Gardiner drew on her own experience leading a UK government-funded digital leadership programme for local authority leaders to illustrate the gap. Senior professionals from care, planning and public administration did not want to know how to run a scrum or read a Kanban board, but they wanted to know who was accountable and governance frameworks that could help them lead technologies sitting well outside their professional expertise. The question of who is accountable when the system is wrong is one that organisational structures are not yet designed to answer.
When Amazon’s recruitment AI began systematically downgrading applications from women, the system had developed that bias internally, through the logic of its own learning process. The organisation’s human workers were operating downstream of a decision they had not made and could not easily see. This is what Actor Network Theory describes as black boxing: the complexity of the model disappears, the output begins to look like a fact, and the accountability returns to the person at the screen.
“You might find yourself sharing your job with an automated system and the system is telling you what to do, and you’ve lost agency, you’ve lost your decision power, and yet you’re still the one who’s responsible, not the system.”
Who is being left behind
There is a clear gender gap in who, in the process of human automation, is being de-skilled, as women are disproportionately hit by this phenomenon given the gender bias and inequality rooted in most professions.
Care work is the clearest example. It is predominantly carried out by women across Europe, consistently classified as low-skill despite requiring sustained professional judgment, and now one of the primary targets for AI-assisted automation. Gardiner’s own work in adult social care illustrated how an AI system can aggregate data from police, GPs, schools and social services to build a risk picture of a child. But what it cannot do is notice that a child is unwashed, or register the instinct of an experienced worker who has seen these patterns before and knows something is wrong before the data confirms it.
“The data can be wrong. It is not finite. It is dependent on what has been admitted to the system.”
The professional who overrides a system recommendation on the basis of tacit knowledge needs organisational backing to do so. In environments where cost reduction is the primary driver of AI adoption, that backing is rarely guaranteed.
The stewardship gap
The concept Gardiner developed in response is the Human Stewardship pyramid, a progression of capability for people working alongside AI. It begins with the human capacities that AI cannot replace: judgment, experience, communication and ethical reasoning. From there, it builds through understanding AI, collaborating with it, knowing when to challenge it, and having the confidence and organisational support to act when something is wrong.
The shift she is proposing in vocational education is from training people to use digital tools to preparing them to steward the relationship between those tools and the people they affect. That requires AI literacy to be part of professional training as a core competency, built into every profession that will be working alongside high-risk systems.
“Every professional should have some sort of training on AI and digital systems. It’s not a bolt-on. It’s not optional.”
The question of knowledge ownership
Gardiner closed with what she described as the most provocative question in the presentation. AI does not generate knowledge from nothing. It extracts, restructures and recombines information that already exists produced by workers over careers. That knowledge, once ingested into a model, continues generating value for the organisation or the technology company that owns the system, long after the worker who produced it has moved on.
The creative industries are already arguing about this. Gardiner’s position is that every profession is now facing the same question, whether or not it recognises it yet. Should workers have rights when their expertise becomes part of an AI system? Who controls the future use of professional knowledge once it has been modelled? These are not technical questions. They are governance questions, workforce policy questions, and questions about the future of professional identity in an AI-enabled economy.
“As machines become more intelligent, will our institutions evolve quickly enough to protect the humans whose knowledge makes that intelligence possible?”
