Balancing Ethics and Innovation in AI Decision-Making

As AI becomes more embedded in everyday life, its influence on how we create, decide and connect continues to grow.

Ensuring AI supports human potential while respecting rights and contributions is key to building technologies that people can confidently engage with and help shape.

Manon Dave is an award-winning creative engineer, music producer and entrepreneur, and has worked with artists including Snoop Dogg, will.i.am and Idris Elba. He leads BBC R&D's Future World Design, an initiative aimed at reimagining media experiences for younger audiences in an era dominated by AI, and serves as Chief Product Advisor for TrueRights.

Here, Manon discusses how to balance innovation with human values as AI raises questions of fairness, transparency, accountability and trust.

How would you balance competing objectives ethically in AI decisions?

As a creative technologist, I always start with purpose. Before performance metrics or scale, I ask what human potential we’re trying to unlock and what human value must we protect. In creative work especially, that means putting imagination, identity and human contribution at the centre.

For me, there are clear non-negotiables. Consent must be present when personal data, likeness, or creative work is involved. Accountability must be clear when something goes wrong. Then value must be shared fairly with the people and cultures that power the model.

Balancing objectives is not about choosing innovation over safety. It is about designing systems where innovation and responsibility reinforce each other. That means pressure-testing decisions through multiple lenses – creators, audiences, regulators and the teams operating the technology. Good intentions are not enough. Creativity thrives in environments where people feel respected and protected.

How can AI decisions be made interpretable for all stakeholders?

As someone who works across music, media and immersive technologies, I think of interpretability not just as a technical challenge, but as a cultural one. If a system shapes what you see, hear and create, you should understand what it is doing and why.

For audiences, that means clear labelling when something is AI generated or personalised. For creators, it means traceability so they can see how their work trains, informs or monetises a system. For organisations, it means knowing who is responsible for decisions and outcomes.

In my world, if people cannot understand how system affects them, they cannot trust it. If they cannot question it, they cannot meaningfully participate in shaping it. Interpretable AI is not just about explaining models. It is about giving people visibility and agency.

What methods would you use to detect and reduce bias in AI?

Bias is not only a data issue. It is a design issue and a cultural issue. The questions we ask, the metrics we optimise for and the people in the room all influence outcomes, so bias mitigation has to happen at multiple layers.

Technically, I support testing systems across different groups, monitoring performance over time and stress testing edge cases. But from a creative perspective, inclusion is one of the strongest forms of bias reduction. When diverse voices help shape a system, blind spots become visible earlier.

In creative AI, bias can show up in whose voices are replicated, whose styles are monetised and whose work is invisible in training data. That is why attribution, consent and fair participation need to become structural safeguards, not just ethical add-ons. If value is flowing in only one direction, bias is already present.

How do you decide which decisions should be automated versus human-led?

In my own work, AI is a collaborator. It helps me explore ideas, generate options and move faster.

However, it is me that makes the final call. Taste, judgment and cultural context are human responsibilities.

I believe automation works best for the tasks that free up time for deeper creative thinking. When a decision affects identity, reputation, livelihood, or opportunity, a human should remain accountable.

Choosing what represents you should not be outsourced. Automation should reduce friction but not remove responsibility.

How would you make AI decisions robust under uncertainty or adversarial data?

We have to design systems that admit when they are unsure. Confidence without certainty can cause real harm, especially in cultural and public facing environments.

From a design perspective, that means building in clear signals about confidence levels, escalating high-impact decisions to human review and strong monitoring once systems are live. It also means stress testing for misuse, whether that is deepfakes, impersonation, or manipulation of creative work – this is as important as optimising for best-case performance.

In an AI-shaped world, resilience is not just about technical stability, it is about protecting identity, authorship and public trust. Trust is the foundation that allows creativity and innovation to flourish.