What Architects can learn from Musicians about Machine Learning.
Introducing Creative Training Data Sets into Architecture
Machine learning models are rapidly transforming creative disciplines, from music to architecture. After listening to a recent RA Exchange with musicians Holly Herndon and Mat Dryhurst (RA Exchange 789) about AI’s implications for artistic autonomy, ethics, and authorship, it made me wonder if there was as much potential in architectural practice. My conclusion was that their approach to AI-generated music offers valuable insights for architects, particularly in how they think about training data and control over digital assets.
Architects and the Machine Learning Wave
Machine learning is increasingly embedded into architectural workflows, from generative design tools to predictive urban analysis. Yet, much of the architectural profession has been passive in shaping how these tools function. Architects often rely on pre-trained AI models developed by large tech firms, without critically engaging with the data these models are built on or questioning who benefits from them.
Herndon and Dryhurst, on the other hand, have taken a more proactive stance. Instead of allowing their work to be scraped and repurposed without consent, they have experimented with ways to control and curate training datasets. This offers a model for architects who want to ensure that machine learning enhances, rather than erodes, their creative agency.
The Importance of Training Data Curation
Herndon and Dryhurst provocatively propose that ‘all media is training data’ and therefore artists should take control of the datasets that shape AI models. Their work with Choral Data (see: Serpentine Arts Technologies White Paper) explores ways artists can build models trained on ethically sourced data that align with their artistic vision rather than relying on black-box systems.
For architects, this raises an important question: Who is creating the datasets that inform AI-driven design? The current state of AI-assisted architecture, through models like Mid Journey, relies heavily on datasets pulled from historical projects, internet scraping, and proprietary archives. If architects don’t intervene, they risk working with models that reinforce outdated norms or serve commercial interests rather than expanding creative possibilities.
Data as an Artistic and Ethical Act
Herndon and Dryhurst training data provocation ****suggests a paradigm shift where designing the dataset itself becomes an integral part of the creative process.
For architects, this could mean:
Curating architectural datasets from previous projects rather than relying on generic, publicly available ones.
Creating AI tools trained on diverse, experimental design processes rather than defaulting to market-driven aesthetics.
Ensuring transparency in how AI models interpret and generate spatial data, rather than accepting AI as a ‘neutral’ tool.
Toward Greater Control Over Architectural AI
The architecture profession has a history of embracing technological innovation, yet it has been slow to engage with AI on a critical level. If architects adopt the mindset of musicians like Herndon and Dryhurst, they could begin to shape AI tools that align with architectural values rather than simply adapting to existing ones. I see some key takeaways for architects:
Architects should see data curation as a design act. Just as musicians control the datasets that train AI models to generate sound, architects should actively shape datasets that inform customised machine learning models for architecture.
Machine learning models must reflect diverse and critical design perspectives. If architects don’t intervene, AI will reinforce the biases inherent in mainstream datasets, such as historical symbols or
The future of architectural AI should be open and participatory. Platforms like Choral Data offer a blueprint for how architects can create collective AI infrastructures that benefit practitioners rather than large tech corporations.
This is why I feel learning from musicians who have taken charge of AI in their domain can help architects reclaim agency over machine learning tools that will shape the future of design.