AI is being used now more than ever to moderate human content, helping us preserve our stories to share with future generations. Making sure that content remains respectful, truthful, and safe-while also authentic and human is a delicate balance.

At Hamilton Robson, we’ve tackled this through innovative projects like “We Can Do Better” and our ongoing work with Belfast Stories, taking advantage of our proprietary Marvel platform, we’ve developed scalable, ethical AI moderation frameworks that empower storytelling while safeguarding audiences.

Balancing Authenticity and Safety

“We Can Do Better“, our project delivered in 2024 with our partners Nerve Centre, invited visitors to a historic National Trust site to share personal stories of conflict in Northern Ireland. While the project aimed to collect authentic narratives, we couldn’t guarantee that every submission would align with the experience’s goals or be suitable for all audiences. This required a moderation approach that preserved the raw authenticity of these stories while protecting visitors from potentially harmful material.

Building on this experience, our current project with Belfast Stories, a creative vision to bring people together in a new tourist centre and creative hub in Belfast, Northern Ireland that’s expected to start construction in 2027, has pushed us to refine this further. Over-moderation risks erasing critical historical lessons and removing meaningful voices. On the other hand, under-moderation could expose audiences to offensive or inappropriate content. We know we need to find a thoughtful middle ground.

The Unique Challenge of Historical Moderation

Platforms with broad, diverse audiences face a universal problem: how to present content responsibly without compromising authenticity. Social media offers a familiar example. Users should expect to encounter material they may not agree with, but algorithms gradually personalise feeds to match individual preferences over time. While this approach works for platforms designed for long-term engagement, storytelling projects like “We Can Do Better” lack that luxury. Visitors arrive with varied backgrounds and expectations, requiring immediate content curation that balances safety with historical truth.

This challenge increases when moderating history itself. Over-sanitisation risks reducing complex narratives to bland summaries, stripping away the lessons we aim to preserve. Imagine editing accounts of the Troubles in Northern Ireland to remove all references to violence or sectarianism. That much censorship would obscure the realities people endured, and the progress achieved through peacebuilding.

Our approach recognises that ethical moderation isn’t about erasing difficult truths, but about framing them responsibly. Like adjusting for the Overton Window, a model for understanding how ideas in society change over time.

A Framework for Ethical Moderation

Through these experiences, we’ve developed a framework and pipeline that can scale across projects using our Marvel platform.

Here’s how we approached it:

1. Transparency and Human Oversight

We cannot rely on AI to fully appreciate the nuances of historical stories. That is why we need a human in the loop:

• Content flagged by AI for potential issues undergoes manual review when confidence thresholds are not met.

• Human moderators refine outcomes and provide feedback to improve the AI model over time.
• This hybrid approach ensures context-sensitive decisions while gradually building trust in automated systems.

2.  Accountability Through Customer Collaboration

Ethical standards should reflect the unique goals and values of each project:

• By involving customers in setting thresholds for flagged content categories (e.g., hate speech or graphic imagery), we ensure diverse and project-led perspectives are embedded into the framework.

3. Scalable Technology

By combining robust AI tools with a flexible CMS like Marvel, we’ve created a solution that adapts to diverse projects and scales effortlessly:

• We’ve tested OpenAI’s moderation API against real-world data collected from previous projects. The results have been promising, with accurate identification of potentially harmful content.
• Marvel’s infrastructure incorporates these tools into its live platforms, enabling real-time moderation while preserving the authenticity of user-generated stories.

4. Mixture-of-Experts Approach

Our approach to AI solutions leverages a Mixture-of-Experts model, where specialised AI systems handle distinct tasks:

• One of these “experts” focuses specifically on content moderation, ensuring that flagged material is evaluated against predefined ethical standards.
• This modular design allows us to fine-tune each expert independently.

Ethical Moderation in Practice

1. Respecting Historical Context

Moderating historical content requires sensitivity to its cultural and educational significance. Over-sanitisation risks erasing key lessons from history, undermining the very purpose of storytelling.

We aim to apply models like the Overton Window to ensure that historical narratives remain authentic while adhering to modern ethical standards.

2. Mitigating Bias

Bias is inevitable in any system but can be managed through thorough testing and customer involvement. By allowing customers like Belfast Stories to define acceptable thresholds for flagged categories, we ensure that our framework reflects their specific needs while minimizing unintended bias.

3. Gradual Automation

AI is not yet ready to replace human judgment in storytelling. Our approach starts with human oversight and gradually transitions toward automation as confidence in the system grows. We want to make sure we do right be our customer but also the people telling the stories.

A Platform Designed for Storytellers

Marvel is more than just a CMS—it’s a platform built to address complex content challenges across diverse use cases. From immersive AR experiences like “We Can Do Better” to multilingual visitor guides at Crumlin Road Gaol, Marvel delivers:

• Scalable Infrastructure: Seamlessly handles large volumes of content ingestion and delivery.

• Automated Moderation: Tailored AI tool flag inappropriate content while preserving authenticity.

• Real-Time Updates: Enables instant content management and remote collaboration.

Marvel’s flexibility allows it to adapt to any project’s unique requirements. Whether it’s preserving history or moderating modern narratives, Marvel empowers organisations to tell their stories responsibly.

Conclusion

We have worked with enough visitor attractions and museums to understand that storytelling platforms should amplify voices while ensuring safety and inclusivity for all audiences. By combining cutting-edge AI tools with thoughtful human oversight, we’ve created an ethical moderation framework that balances authenticity with protection.

Marvel stands at the heart of this innovation—offering scalable solutions that adapt to diverse contexts without compromising on ethics or authenticity. As we continue refining our approach through projects like Belfast Stories, we remain committed to empowering visitors to actually commit their stories to these heritage projects – with tools that preserve history in an authentic way, foster connection, and inspire trust.