New research on AI fairness: How co-creation can mitigate bias

Artificial intelligence is now a part of everyday decision-making, from loan approvals to hiring processes. Building fairness into these AI tools matters more as they spread. However, companies still struggle to get it right.

As public awareness grows, fairness is no longer just an ethical concern—it’s a business imperative. Companies failing to address AI bias risk losing trust, damaging their reputations, and facing increasing regulatory scrutiny. Overall, addressing AI fairness isn’t just about correcting outputs; it requires rethinking how we design and develop AI systems.

New research from our team explores a powerful solution: co-creation—bringing diverse stakeholders into AI development—paired with foundational AI training to ensure meaningful contributions. This research provides a practical framework for businesses looking to proactively embed fairness in their AI systems.

New Research: Co-Creation as a Path to AI Fairness

The need for a greater focus on AI fairness is growing as AI adoption increases. Our researchers, Ian Slesinger and Evren Yalaz, recently published a study exploring how co-creation, when combined with early-stage training, can significantly reduce bias in AI models​.

Their findings highlight two key insights:

1. Involving diverse participants early helps to identify biases before they become deeply embedded in AI models.
2. Providing AI literacy training ensures that participants can offer meaningful feedback rather than surface-level observations.

This approach balances technical rigour (data selection, algorithm tuning) with social perspectives (insights from underrepresented groups in tech development). The research demonstrates that structured and inclusive co-creation leads to more equitable AI outcomes.

How Co-Creation Strengthens AI Fairness

Many AI development processes involve collecting feedback only after a system is built. However, this retrospective approach limits the ability to correct bias effectively. Instead, co-creation encourages continuous collaboration from the start. Here’s how it makes a difference:

1. Uncovering Hidden Biases Early

AI models often inherit biases from training data. By involving individuals from diverse backgrounds, organisations can identify problematic assumptions before they become embedded in the system. This reduces the need for costly bias mitigation later.

2. Strengthening Public Trust

AI fairness is not just about eliminating bias—it’s also about transparency and accountability. When affected communities participate in AI design, they better understand how decisions are made and trust the system’s outputs.

3. Aligning with AI Regulations

With global AI regulations evolving, businesses must demonstrate fairness and inclusivity. Co-creation provides a structured method to document fairness efforts, making it easier to comply with emerging guidelines such as the EU AI Act.

Training: The Missing Piece in AI Fairness

Our research reveals that co-creation alone isn’t enough—it must be paired with AI literacy training. Without a foundational understanding of how AI models learn and where biases emerge, participants may struggle to provide valuable input. In order to make co-creation successful, teams should look to implement the following training elements:

• Simplified AI Explanations – Using everyday examples (e.g., movie recommendation algorithms) to help participants grasp AI concepts.
• Real-World Case Studies – Demonstrating bias in action through tangible examples like fraud detection or customer service chatbots.
• Ongoing Support – Ensuring that stakeholders stay engaged through workshops, follow-up sessions, and continuous learning resources.

Practical Steps for Businesses to Implement AI Fairness

Companies looking to embed AI fairness into their models can take these actionable steps:

1. Identify Key Stakeholders Early – Consider the peoples your AI system will impact. Whether it’s job applicants, customers, or internal teams, involve them at the design stage.
2. Commit to Iteration – AI fairness is not a one-time fix. Schedule multiple feedback loops at different development stages to refine models continuously.
3. Document the Process – Keep records of bias findings, stakeholder feedback, and adjustments made. This documentation will support compliance efforts and demonstrate a proactive approach to fairness.
4. Measure Fairness Outcomes – Use demographic fairness metrics to track improvements over time, ensuring that AI models perform equitably.
5. Balance Technical & Social Solutions – AI fairness requires both algorithmic improvements and human collaboration. Invest equally in technical fixes and inclusive design processes.

Building a Fairer AI Future

As AI adoption accelerates, fairness cannot be an afterthought. Businesses must proactively embed ethical considerations into AI design—leveraging co-creation and structured training to prevent bias before it arises.

At Trilateral Research, our goal is to create AI systems that reflect societal values of equity and inclusivity. By applying research-driven methods, we help organisations move beyond theoretical discussions and implement practical fairness strategies.

To explore the full findings from our research, download the paper here​. If you're interested in developing a fairness-first AI strategy, contact our team today.