Question 1: What technologies are you working with, or have you worked with?
For quite a few years, the focus of my research has surrounded the large-scale empirical analysis of experiential and emotional qualities of urban environments. I’ve always been interested in being able to analyze how citizens “feel” across different parts of cities, and how this knowledge can be used to inform design and planning. Specifically, I’ve been developing design assistance tools and software to allow urban-designers, planners, and policy-makers to tap into these complex and ‘intangible’ urban qualities, for empirically-grounded decision-making. My past research has involved the training of machine-learning models for real-time evaluation of the experiential qualities (such as liveliness, beauty, safety, stress, etc.) of urban scenarios being designed. My current work however marks a departure from such quantitative methods, and focuses rather on innovative ways to represent, communicate, and evaluate complex urban qualities beyond numbers and scales. I am currently working within a very different technological paradigm, where such qualities are handled relationally through examples and analogies, rather than through explicit quantities.
I rely on a number of diverse methods, techniques and data-sources for different parts of my research and development workflows. Since I work primarily with geospatial data, I depend a lot on Geographic Information Systems (GISs) for data collection, management and spatial analysis, and many of these systems are integrated into the tools that I develop. I use supervised Computer Vision models (i.e. Convolutional Neural Networks (CNNs) or Vision Transformers) to extract and analyze urban features from geolocated Google street-view imagery or crowdsourced photographs. Publicly available census records and land-use data are useful for analyzing demographic and use-related features. I use unsupervised-learning methods for clustering urban locations based on their features, and for assessing experiential similarity between environments. Textual descriptions gathered from Google Places or X (Twitter) can also provide rich information on perceived urban qualities. I’ve experimented with Natural Language Processing (NLP) methods on these descriptions to analyze topics of discussion and associated sentiment across what I consider to be the emotional landscape of the city.
Question 2: How do you take account of MIT’s obligation to pursue the public interest in the work that you do?
With the rise of machine learning methods over the past decade, the automation of various decision-making processes has increased. Here, the question of ‘interest’ is a crucial one. In the context of my research, the biggest question is ‘who’s interest?’. Urban experiential qualities are subjective in nature. No two individuals experience the city in the same way. Cities, in this sense, form the backdrop for a rich tapestry of feelings and emotions. Data-driven analyses of these qualities often succumb to reductionist approaches, and run the risk of homogenizing experience into cut-and-dry numbers. Design evaluation and decision support tools relying on data-driven methods are often not in the true interest of the ‘public’, because they do not account for how the public realm is plural by definition.
My recent approach towards the question of ‘public interest’ and plurality has been to treat urban experience not as a set of quantities that can be ‘measured’ or predicted in an absolute manner, but rather as a relational quality that can be handled through references, examples and analogies to other urban environments which share similar features. For example, many of the qualities of the narrow lanes in Beacon Hill, Boston, can be evaluated through references to some of the cobblestone lanes in the Old City of Philadelphia. In this sense, keeping the idea of experience open-ended serves the interest of the public to a far greater extent, because analogies can capture and communicate subjective qualities in a way that numbers can never do. The obvious challenge here, of course, is the development of systems and tools that can retain this relational openhandedness within a data-driven paradigm, such that they are still able to provide valuable decision-support. How can an urban experience-analysis tool still be useful to designers and planners if the units for measuring experience are not fixed or objective, but rather fluid and highly referential. My current work attempts to address some of these questions through a data-driven framework for what I call ‘evaluation by analogy’. In this regard, I think it is really important for the MIT community to engage with questions of knowledge-representation in the context of design evaluation, and think about ‘experiential-big-data’ beyond explicit quantities.
Question 3: What more could you and others do to help MIT meet its social obligation to pursue Public Interest Technology?
In the age of big data, I think the question of ‘whose data’ is as important as ‘whose interest.’. While this has very much been at the center of contemporary discourse surrounding the ethics of Artificial Intelligence, I believe we still have a long way to go. As researchers, we must take the first step of always remaining critical of the degree to which our data represents the ‘public’ and being aware of the conscious or unconscious decisions that have gone into generating the datasets our work relies upon. No dataset is a ‘true’ representation of the external conditions it stands in for. Datasets are abstractions, capturing only certain aspects of the world. The analyses that we as designers and planners run on our datasets, and the results and recommendations that we derive thus also inevitably betray a specific way of ‘seeing’ society. A constant acknowledgement of our own positionality is very important.
At the same time, however, it is also important for us to pass on this awareness to the users of the tools and software products we develop as part of our research. The subjectivity involved in the creation of these systems are often not apparent to the users, who use them for critical decision-making. It is important to rethink the ways in which tools are designed, such that users don’t simply use them as black-box models. Rather, developers and researchers can create products that encourage users to explore the datasets that go into the training of such models, and also the processes of data collection, and dataset preparation.
Rohit Priyadarshi Sanatani is a designer, researcher, and currently a PhD scholar in Design Computation and Advanced Urbanism at the Massachusetts Institute of Technology (MIT). His work is supported by a Leventhal Center for Advanced Urbanism (LCAU) Fellowship. Rohit’s interests lie in data-driven lines of inquiry into the experiential and affective qualities of urban environments. His recent work has focused on the creation of design assistance and evaluation systems that help designers and planners to systematically engage with such complex and subjective qualities of the public realm. He has also been associated with the MIT Senseable City Lab, where his work has revolved primarily around large-scale urban visual analytics.
Rohit graduated with a dual Master of Science in Design Computation and in Computer Science from MIT. Prior to that, he was an Assistant Professor at the Department of Urban Design, School of Planning and Architecture (SPA), New Delhi, and also worked on multiple masterplan development schemes across India as Project Lead with Studio Lotus, New Delhi. He holds a Bachelor’s degree in Architecture from IIEST, Shibpur, and a M.Arch (Urban Design) from CEPT University, Ahmedabad.