A New Approach to AI-Generated Images Puts Communities in Charge

Microsoft researchers have developed a new platform that allows communities to participate in the creation of AI-generated images, ensuring they are portrayed accurately and fairly. The Community Library Creator is an effort to address the issue of biased data used in AI development, which often perpetuates negative stereotypes and misrepresentations of marginalized groups.

The foundation for AI knowledge comes from data most people never see or interact with, including technical processes that determine what’s included or left out. As AI becomes increasingly influential in daily life, it’s essential to involve communities in the creation of this data to ensure accurate representation. Anja Thieme, a principal researcher at Microsoft with a background in social psychology and human-computer interaction, emphasizes the importance of collective definition and negotiation when it comes to how people should be represented.

Communities are building their own libraries instead of relying on patchy online data using the Community Library Creator tool. This platform provides a structured way for groups to define what good representation looks like and build the data around it in a way AI models can learn from. Each library is created by a group of people with shared experiences, such as disability or identity, working together through advocacy organizations.

A community library is essentially a collection of images or videos created by community members that reflect their shared experiences. These libraries are paired with detailed descriptions explaining what matters about each image, providing context for AI systems to understand not just visual representations but also the underlying meaning and significance from the community’s perspective.

The Community Library Creator software was developed by Thieme and her team alongside Microsoft’s Accessibility Team. This tool provides a step-by-step method for communities to curate images or videos organized around key themes that represent their shared experiences. For example, a community of Black people with albinism might highlight elements like wearing hats for sun protection or sitting close to work due to vision disabilities common in their condition.

Each library aims to gather about 400 ‘real-world’ images from community members, capturing a range of experiences with a focus on quality over quantity. Thieme emphasizes the importance of this approach: ‘It’s putting people and their data and their rights first.’ The result is a resource that can be used to help AI systems learn from more diverse lived experience, not just patterns in online data.

The Community Library Creator platform enables communities to actively shape both the data and evaluation standards that direct AI development. This matters because representation isn’t just technical; it needs to be defined by the people it affects, drawing on their lived experiences that others can’t fully replicate. By giving control back to those represented, this approach addresses a significant issue in today’s AI landscape.

The process of building community libraries starts with choosing images that feel meaningful and explaining why they’re important using the Community Library Creator’s step-by-step method. This reflection helps communities zero in on key themes like family life, work, or everyday routines that represent their shared experiences. From there, they curate a larger collection of images or videos organized around those themes.

A community library is not just a collection of images; it’s also paired with detailed descriptions explaining what matters about each scene. These annotations help AI systems understand the context and significance behind visual representations, enabling them to better reflect the experiences of marginalized communities. For instance, a community might highlight elements like wearing hats for sun protection or sitting close to work due to vision disabilities common in their condition.

The images and annotations become training material used by AI models to learn from more diverse lived experience. Prompts generated from the library are used to create AI images that community members review and rate based on how well they match their desired representation. Over time, these ratings create a feedback loop giving AI systems a clearer sense of what ‘good’ looks like as defined by the community.

A key part of this model is that the community itself owns the data through the advocacy organization that built the library. This means communities retain control over how their data is shared and used, including whether to make it available on platforms like Hugging Face or place limits on its use. The images are collected with consent, and if someone wants their data removed later, the community can make that change.

This approach puts people and their rights first in a space where much of today’s AI data is scraped from immense amounts of information online without visibility into its origins or control over how it’s used. Thieme emphasizes: ‘It needs to be collectively defined and negotiated.’ By giving communities more agency, this platform helps address the issue of biased data used in AI development.

The Community Library Creator tool is currently being used in a controlled way with specific advocacy organizations. Thieme and Cecily Morrison co-lead a team working on engineering, safety, and legal controls needed to expand to more communities. The libraries are a way to widen who gets to shape AI in the first place – not just systems themselves but also data and criteria used to define success behind them.

The work of these community libraries could involve more communities and help train and evaluate future AI models. Thieme emphasizes: ‘We’re really trying to widen participation in AI and who gets to have a voice in shaping its future.’ By providing tooling for others to create the future of AI, this platform opens up opportunities for marginalized groups to contribute their perspectives and experiences.

The Community Library Creator is an essential step towards creating more inclusive AI systems that reflect diverse lived experience. As Thieme notes: ‘We need to see this a lot more in general in the AI space.’ By putting communities at the forefront of data creation, we can ensure accurate representation and address issues of bias in AI-generated images.