Despite rapid advances in artificial intelligence technology, current AI systems remain fundamentally inadequate for architectural design work, according to a leading expert in the field. Phil Bernstein, deputy dean and professor at Yale School of Architecture and former Autodesk vice president, argues that while AI tools show promise in certain areas, they fail dramatically when it comes to core architectural tasks like creating coherent floor plans.

Bernstein, who teaches courses in professional practice and technology at Yale, has been closely monitoring AI's integration into architectural education and practice. He points to the breakneck speed of AI adoption, noting that while it took the telephone 100 years to reach 100 million users and Facebook four years, the AI platform DeepSeek achieved the same milestone in just two months. This rapid evolution creates challenges for educators trying to develop appropriate teaching methods.

At Yale School of Architecture, faculty are taking a three-pronged approach to AI integration. First, they ensure students understand the philosophical, legal, and disciplinary implications of using AI technologies, including issues around intellectual property, precedent, and academic integrity. Second, they provide students with access to as many AI platforms as possible through partnerships with the central campus. Third, individual instructors determine how AI tools should be used in their specific studios.

One experimental studio at Yale attempted to delegate significant design responsibility to AI algorithms, but the results were disappointing. "It didn't go well," Bernstein explained. "You lose a lot of design autonomy when you delegate to an algorithm." The main issue was control – students found themselves unable to guide the design process effectively when relying heavily on AI systems.

In professional practice, architectural firms are experimenting with AI in various ways, though innovation levels vary significantly based on factors like willingness to experiment, internal capabilities, data availability, and technological sophistication. Most firms are using AI for marketing materials and rendering generation, but Bernstein believes the real innovation will likely emerge from larger firms that have the resources to invest in expensive AI technology and data management.

However, firms tend to keep technological advantages secret rather than sharing them with competitors, making it difficult to assess the true extent of AI adoption in the industry. This pattern mirrors what happened with building information modeling (BIM) technology, where firms didn't help each other adopt the new tools.

Bernstein expresses skepticism about the current dominant approach to AI development, which relies on memorizing vast amounts of data and organizing it into probability matrices. He believes this "Connectionist" approach, championed by figures like Sam Altman, will ultimately need to be combined with symbolic logic methods to achieve reliable results. For architects, reliability is crucial, and current AI systems fall short of professional standards.

The fundamental limitation of current AI systems becomes apparent when they encounter multivalent environments that require integrating multiple streams of data and logic – which describes virtually every architectural project. While AI might threaten certain specialized fields like structural engineering, which has clear rules and standardized representations, architectural design requires complex judgment, experience-based decision-making, and responsibility-taking that current AI cannot replicate.

"These diffusion models right now can't draw a damn floor plan with any degree of coherence," Bernstein stated bluntly. "A floor plan is an abstraction of a much more complicated phenomenon." He argues that it will be a considerable time before AI systems can handle the most important aspects of architectural work: making judgments, exercising experience, making tradeoffs, and taking responsibility for decisions.

The job displacement debate surrounding AI reveals complex dynamics that extend beyond simple replacement scenarios. Bernstein points to recent experiences in software engineering, where companies initially replaced programmers with AI code generators but then had to rehire human programmers because the AI-generated code wasn't reliable enough. Additionally, students who rely on AI to write code often lack the debugging skills necessary for professional work, creating a skills gap that concerns employers.

Architectural firms face particular challenges in adopting AI technology due to market realities. Architects are generally poor technology customers – they're cost-conscious and often use pirated software, making them unattractive targets for multibillion-dollar AI investments. The architecture industry also lacks the organized, high-quality data sets that AI systems need for training, unlike fields like medicine where researchers can access curated data from thousands of patients.

Even major architectural firms like Skidmore, Owings & Merrill don't have enough historical project data to properly train an AI system, according to Bernstein. The industry's data is typically disaggregated and scattered across different platforms and firms, with companies reluctant to share information due to competitive and risk concerns.

Regarding AI's impact on architecture's business model, Bernstein sees two potential paths for the profession. The first is a "race to the bottom" where firms compete to offer the lowest fees by cutting costs through AI automation. The second, more promising approach involves using AI to provide better services and charge premium fees for enhanced value.

Bernstein advocates for the value-based approach, suggesting that firms could use AI to make specific performance guarantees to clients, such as promising that a building will produce a certain percentage less carbon emissions. If firms can provide third-party validation for such claims, they should be able to command higher fees for their enhanced capabilities.

The expert emphasizes that the choice between these approaches isn't fundamentally about technology but about business strategy. How the architectural profession adapts to AI will depend on strategic decisions made by individual firms and the industry as a whole, rather than being determined solely by technological capabilities.

Looking ahead, Bernstein remains cautious about predictions regarding AI's timeline for advancement in architecture. While acknowledging that AI tools will likely become more capable over time, he argues that if AI becomes sophisticated enough to truly replace architects, society will be facing much larger economic and social disruptions that extend far beyond the architecture profession.