Data is omnipresent. Our digital interactions throughout the day represent a wealth of information about our habits, preferences, interests, and activities. Our physical movements are captured by our phones and, increasingly, the buildings and cities where we work and live. 

Considering the abundance of data, how do we curate and analyze it purposefully? 

In December, the AIANY Social Science and Architecture Committee hosted AI and Sensemaking: Human-Centered Design in the Age of Abundant Data, where experts from the worlds of design, technology, data science, and organizational sociology examined this question. Organizations are turning to AI for the “sensemaking” that helps us understand the significance of the data.  The speakers—Daniel Pittman, partner for strategy and innovation at TAD; Will Shapiro, cofounder and CEO of Topos Inc.; Andreas Hoffbauer, founder and director of Atelier Kultur; Melissa Marsh, founder and executive director of PLASTARC and senior managing director of occupant experience at Savills; and moderator Nitzan Hermon, founder and consultant at Future-of.Agency—reflected on AI and architecture through case studies, current projects, and potential scenarios.  

 

Daniel Pittman showed how TAD, which specializes in the integration of architecture and technology, is exploring AI’s relevance for user experience, data analysis, and messaging. To illustrate how AI could provide new perspectives on complex information, he described a generative content project that relies on AI to create a real-time depiction of financial markets using boids; the room-height, multi-wall installation is both an aesthetic statement and a means of providing insight into market data. 

The firm is exploring other arenas where AI could foster insight or greater simplicity in settings ranging from corporate offices to medical facilities. Referencing the firm’s work to integrate AI into different projects, Pittman considered the implications of natural language interfaces and authentication; of process automation, in which intelligent systems help clients operate more efficiently and gain insights from their technology networks; and of the firm’s efforts to create “frictionless environments” in which technology enables seamless experiences in workspaces. 

Takeaways:
There is a disjunction between rhetoric and the reality of AI; his firm’s purpose is to distinguish between the two. Beyond that, the firm has three roles in this arena: it helps clients best prepare for what lies ahead; determines what are legitimate, pragmatic steps forward for AI; and provides insight about how clients should be thinking about AI and why it’s relevant to their people and customers. 

 

Throughout his consulting work, Will Shapiro has taken powerful data analysis tools to, as he described it, “understand place holistically.” Topos relies on AI to discover the dynamics of people and urban environments that can be meaningful for governments, planners, and businesses. As a demonstration of the potential for discovery, he discussed his firm’s “Five Boroughs for the 21st Century,” an effort that used AI to cluster New York City regions not by geography, but by data. Using an AI technique called “k-means clustering,” he analyzed publicly available urban data to identify 17 key dimensions that define the city. The correlations that emerge—between nightlife and dollar-pizza eateries, or between a neighborhood’s pizza topping options and residents’ median household incomes—reveal surprising parallels and a basis for creating the new mappings. 

Takeaways:
AI creates the possibility for us to understand cities in a more granular, temporal way in the face of rapid shifts in urban life and dynamics. This analysis could be powerful in determining how cities are designed and zoned—in conceiving the way we organize elements and neighborhoods of the urban environment. AI, Shapiro said, would “allow people to use data to make those decisions, instead of intuition or arbitrary standards.” 

 

Melissa Marsh highlighted the use of data in defining and improving design, placing a strong emphasis on curating information that is truly relevant to the design endeavor. Her firm PLASTARC focuses on strategic design for workplace performance and innovation, and she noted that buildings now gather a plethora of data, yes—but this data often says nothing about behavior, which is the key factor in human-centered design. “What makes data in buildings relevant is people’s actions and interactions,” she observed. With that in mind, practical applications of data in architecture must focus on human dynamics.

Case Study:
To highlight PLASTARC’s data-sifting approach, Marsh spoke about a consulting project to examine why a newly completed office was often 50% empty. By amassing data points about physical space characteristics, resources, and social drivers, the firm revealed the intersection between and the characteristics of places and the choice to occupy them. The firm relied on the assessment to improve the design of the spaces, using a combination of social, technological, and physical interventions—a mode of assessment that is ripe for AI. 

Takeaways:
Just because buildings are becoming “smarter” doesn’t necessarily mean that we are bringing greater intelligence to architectural design. In order to create more meaningful analyses, we need more time to gather data from pre- and post-occupancy periods. 

AI is pushing us to conceive how analytic insights might apply to future projects, but “our first responsibility is to look at what we’ve already built before we focus on what the next buildings will be.” 

 

When Andreas Hoffbauer visits an organization’s offices, people will frequently remark to him, “Look how collaborative we are.” Yet for Hoffbauer, an organizational sociologist who helps businesses foster innovation, there is often a disconnect between the way people aspire for a space to perform and how it actually performs. His role is to consider why these workplaces aren’t fulfilling their purpose and to help the organization address the problem. 

Hoffbauer’s analysis has revealed the limitation of relying on data—because the under-performing workplaces were often designed using analytic insights. Yet the data and the AI analyses, Hoffbauer has found, have some invalid assumptions. Among them:

  • Workplace interactions are dematerialized to a limited set of variables that do not capture the full range of employee behavior and work culture.
  • The analyses assume, often incorrectly, that past practices are a valid basis for predicting future practices and thus designing new workspaces.
  • Most importantly, the analytical models and data ignore the behaviors, practices, norms, and social contexts that allow certain kinds of work to happen. 

Takeaways:
Buildings designed around current practices and current user data will not necessarily perform in the long-term as our behaviors and workplace culture change. Instead, organizations and designers have to consider how behaviors, practices, and norms fit with where an organization expects or aspires to be in the future. In this effort, sociologists can help in determining what behaviors would predominate. As an example of such a collaboration, Hoffbauer pointed to the library Snøhetta designed for Ryerson University in Toronto, “a library without books” that cultivates collaboration, conversation, and interaction, the activities anticipated to characterize the campus athenaeum of the future.

 

Guardrails: The Ethics of AI

A recurring theme throughout the discussion focused on the ethical dimension of AI as it related to architecture.  The panelists noted several potential issues:

  • AI can provide insight, but it can also perpetuate biases and self-indulgent creative loops; the result—spaces that are stagnant and employees who are disconnected or alienated.
  • Over time, analyses derived from building data can become stale because, when the information set does not change, it perpetuates a certain view of the world.
  • Organizations and governments have begun to use AI to uniquely identify individuals, creating the possibility for nefarious ends, as seen in China’s use of AI to identify Uighurs, among other initiatives.
  • Research into AI’s application for detecting emotions through physiological measures, facial expressions, and body movement could lead to enforced conformity and retribution for straying from the norm, in both organizations and societies at large.

Takeaways:
Designers and technologists need to consider how they can introduce a contrarian view into the data-gathering mechanism and analysis—a view that questions the assumptions built into the data and analytics.

In using AI technology—whether natural language systems or facial recognition algorithms—it is the responsibility of design and technology firms to consider the potential impact on social norms, privacy, and other ethical considerations, and present these issues to the client. 

 

Data Curation 

In considering how to curate information that would be most meaningful for a particular project, Hermon noted that he prefers to “work with deep context as opposed to wide data.” He investigates the content that would have the greatest relevance, based on the organization’s questions or goals, rather than accumulating a wide range of data that doesn’t apply to those aims. 

On a similar note, Marsh noted, “We need to be faster at figuring out what is impactful and not impactful and have better hypotheses going in so we’re being more thoughtful before we go into the data, or we could be just boiling the ocean.”

 

The Future of AI

Among the beneficial impact of AI, Hoffbauer noted that it could play a role in “trying to make spaces that people like to work in, spaces that make a positive impact in people’s lives.” Marsh suggested that AI could contribute to the paradigm of thinking of buildings as “living organisms that we’re always learning from and adjusting to over time.” 

The panelists agreed that the integration of AI and architecture is still in its infancy, and, as such, we haven’t yet identified appropriate measures of success and progress; we haven’t identified the rules and objectives in this field. “We have work to do as a profession or industry about what is progress and what does good look like,” Marsh said. “Once we are able to identify what identifies success or performance within that context or environment, then we could get to better architecture and have a possibility of using computation to make a better architecture.”