Regaining Agency over AI Requires Putting Humans Back at the Centre of Its Development

Artificial Intelligence (AI) has become a fact of life for most people; it is now embedded across industries and increasingly present in our daily lives. As its capabilities continue to expand, it is often framed as a force acting on society rather than something shaped by it. While headlines rightly spotlight fears of job displacement, these concerns only capture part of a more profound shift in the relationship between humans and AI.

What is changing more fundamentally is the question of agency. As AI systems become more capable and more widely deployed, they are beginning to influence how decisions are made, how information is interpreted, and how outcomes are determined across a range of contexts. This raises a more structural question about the role that humans play in shaping these systems, not just using them.

The Imbalance at the Centre of AI Development

AI systems did not emerge in isolation, fully formed and ready for deployment; they were trained on vast reservoirs of human-generated data – our language, behaviours, and decision-making habits – gathered from the internet and through direct user interactions. This process has enabled remarkable progress, but it has also created an imbalance in how value and control are distributed.

The people whose knowledge and lived experiences fuel these systems often lack visibility into how their data is used, let alone meaningful influence over the systems’ evolution. Today’s paradigms reduce human input to a passive commodity: collected, aggregated, and funnelled into training pipelines, with little connection back to the individuals who generated it. This dynamic has sparked high-profile legal cases, such as the case of The New York Times v. OpenAI and Microsoft over the use of the former’s articles, or the collective of artists who sued Midjourney, DeviantArt, and Stability AI for training image generators on their creative work without consent.

As AI becomes more embedded into environments where judgment and context matter, that disconnect becomes more significant. The question is not only what AI can do, but who has a role in determining how it develops.

From Passive Input to Active Participation

Addressing this imbalance does not require a retreat from AI, nor does it depend entirely on regulatory intervention. The most compelling way forward positions humans as active participants in training AI, enabling us to reclaim the value of our contributions.

Today’s AI models draw on static datasets that capture what people do on the surface, but they struggle to truly grasp the thinking behind it – the personal judgments and real-world context that shape our decision-making. These human elements are key to good choices, yet most training methods overlook them entirely. Anyone who has used generative tools like Copilot or ChatGPT has likely seen this gap: responses, while often impressive in fluency, can fail to match basic standards of human intelligence, sometimes resulting in frustrating or even hilarious outputs.

A better approach would treat humans as ongoing collaborators, integrating not only what we know but how we think – the step-by-step reasoning, trade-offs, and situational insights that inform our choices. In practice, this means designing AI systems that can incorporate structured human input, so they can learn from contextual thinking as they evolve, rather than relying solely on historical data.

Re-establishing Agency Through Contribution

If human knowledge and behaviour are central to the development of more advanced AI systems, then recognising and compensating that contribution becomes a practical consideration as much as an ethical one. Human input is a resource, one with great value, and one that should give people power over the future of AI.

This perspective also reframes how we think about the impact of AI on work itself. While certain roles will be displaced or reshaped, there is also the emergence of new forms of work centred on training, evaluating, and improving AI systems. These roles form part of the infrastructure required to advance AI, rather than sitting alongside it.

There is already early work in this direction. Platforms such as Humanix are exploring how human reasoning and behavioural input can be integrated into AI training in a structured, scalable way. The specific implementations will vary, but the underlying principle remains: AI development becomes a process in which human contribution is explicit and compensated, rather than assumed and stolen.

Regaining agency over AI, in this sense, does not need to limit its ultimate capabilities. The systems can still grow while being connected to the people whose knowledge and behaviour they reflect. As AI continues to develop, the degree to which humans can steer this process will profoundly shape AI’s performance and its coexistence with us in the years ahead – inviting us to choose collaboration over extraction.