AI in Design
Transmission Model of Teaching Leads to Conservation of Injustice
Current worries about AI focus mostly on LLMs and on policing subjective qualitative content and narrative expression in the humanities (e.g., essays or presentations).
These worries are anthropocentric and Eurocentric. They privilege dominant forms of cultural capital and economic power. They also defend subjective and optional lifestyle skills that have a poor track record and lack an evidentiary basis.
Examples of these concerns include:
- Preservation of the prestige and financial models of higher education.
- Policing of intellectual property.
- Withdrawal of access to tools from less powerful groups, including non-experts, speakers of non-English languages, students, non-elite institutions, non-urban settings, and citizens engaged in scrutiny.
- Teaching the dated and provincial canon.
The Need for Innovation as a Way to Respond to Uncertainty
These concerns are disconnected from the key challenges of design. Those challenges centre on creating novel and lasting benefits in ways that are both real and fair.
Designers need evidence-based practice to understand what is happening and what is possible. Without evidence, architecture and design remain marginal and poorly funded. They also fall behind scientific, engineering, medical, economic, political, and ethical research, including work that integrates traditional knowledges and nonhuman knowledges and practices.
Designers must also work under conditions of uncertainty and redundancy if they want to engage multiple ontologies, epistemologies, and forms of practice.
The field also lacks integration across basic knowledge, training, community culture, shared practices, and tool development and adaptation.
This situation is especially challenging in design and architecture because the work demands an unusually broad range of overlaps and interconnections.
Examples of these gaps include:
- Lack of site, project, and stakeholder information collection and sharing.
- Lack of standards, quality assurance, and risk management beyond professional self-protection.
- Lack of longitudinal studies and monitoring of outcomes.
- Poor engagement with evidence-based approaches in areas such as health and conservation.
Automation as Co-learning Infrastructure
In this situation, automation tools can act as co-learning infrastructure. This includes not only LLMs, but also machine learning, expert systems, data analysis, simulation, visualisation, statistical projections, big data parsing, quality control, trend analysis, and scenario planning. These tools can shorten the learning curve. They can orient designers, as perpetual novices, to established, tested, and reusable disciplinary practices, sites, and everyday interactions. They can also build interfaces across disparate knowledge bases and data types, support experimentation and innovation, enable iterative development in silico and through living labs, demonstrate potential impacts and benefits, and track drawbacks and failures so that practitioners can reverse or adapt their actions.
Automation lowers the threshold for these activities and can justify its use by multiplying beneficial impact.
At the same time, LLM use is highly profligate and carries serious environmental and ethical costs. Any future use therefore needs a way to assess risks and benefits with evidence, which returns us to the same underlying logic.
Examples and cases include:
- Use APIs to obtain multidimensional data from multiple sources, including open data, government data, and data behind paywalls already licensed by the university, across spatial, media, economic, geographical, and environmental domains.
- Visualise complex and dynamic relationships as evidence-driven data stories or scenarios of alternative futures that connect with high-profile conversations at IUCN, COPs, and similar forums.
- Create evidence-design-fabrication-installation-monitoring loops that connect many, or all, forms of expertise: professional and lay, trained and traditional, human and nonhuman.