The paper, co-authored by Ofir Glassman, Orly Cohen-Moas, Achituv Cohen, Noam Teshuva, and Jonathan Dortheimer, investigates a question that sits at the heart of our research: if we build AI tools for urban design, do students actually use them?
We studied how third-year architecture students engaged with a web-based machine learning tool that predicted pedestrian volumes as a proxy for walkability. Despite being introduced to the tool in a design studio setting, students rarely integrated it into their iterative design process. Through a usage survey and 12 semi-structured interviews, we identified four key barriers to adoption:
These findings point to a broader sociotechnical challenge: making AI tools genuinely useful in design education requires more than technical capability. It demands attention to model transparency, pedagogical scaffolding, and the realities of studio culture.
We look forward to sharing this work at eCAADe 2026 and to the conversations it will spark about the future of AI in architectural education.
]]>Her paper, “Architectural Diagrams: Mapping, Classification, and Digital Evolution,” examines how architectural diagrams have changed across four decades of published architectural graphics.
Diagrams have long been central to architectural design as tools for reasoning, communication, and visual synthesis. While their historical and conceptual roles are well established, the effect of digitization on their actual function and representational conventions remains underexplored.
The study systematically analyzes 245 diagrams drawn from 45 architectural graphics publications between 1980 and 2025. Each diagram is classified by design stage, representation technique, production method, and primary purpose.
The findings document a significant transformation in published diagram practices. In the 1980s and 1990s, diagrams were predominantly hand-produced and exploratory, concentrated in early design stages. From the 2000s onward, digital production becomes near-universal, yet representational conventions remain largely two-dimensional. By the 2020s, published diagrams are increasingly oriented toward external communication and later-stage coordination; conceptual and exploratory uses decline sharply.
The paper interprets this shift as not only technical but disciplinary. In architectural publishing, diagrams appear to function less as instruments of design thinking and more as curated visual statements for professional audiences. The study argues for a renewed focus on digital-native diagram practices and opens directions for studying visual communication as a research object in its own right.
]]>Cities face growing pressure to address sustainable growth, climate resilience, energy efficiency, mobility, and social inclusion — often with limited capacity to evaluate the consequences of decisions before they are made.
Urban digital twins can in principle connect real-time data, historical records, and predictive AI models into a shared environment for urban analysis and scenario testing. In practice, however, their deployment remains limited. Urban data is fragmented across agencies and formats, AI models require heterogeneous and often incompatible inputs, and cities lack the workflow orchestration and benchmarking infrastructure needed to integrate these components into coherent planning workflows.
METROPOLIS addresses these barriers directly. The framework contributes a data interoperability engine, a modular interface for deploying urban AI models, pipelines for orchestrating multi-model scenarios, and benchmarks for comparing model performance across cities and planning contexts. Decision-support interfaces are designed for use by planning practitioners, not only researchers.
The project will be validated through a pilot deployment with the City of Tel Aviv, assessing technical performance, usability for planning workflows, and replicability for other municipal contexts. METROPOLIS does not aim to produce a single predictive tool — it aims to build the infrastructure that makes urban AI research and practice more cumulative and interoperable.
]]>The ISF is Israel’s central competitive funding body for basic research. The grant recognizes the scientific potential of studying how designers and AI systems can communicate through text, diagrams, sketches, and other visual modalities.
Our project, “Towards Human-AI Collaboration: An Investigation into Effective Design Communication Modalities Within Chat Interfaces”, studies how designers, architects, and everyday users interact with AI when design problems require both conversation and visual reasoning. At its heart, the project examines the gap between human creativity and AI’s emerging capabilities, particularly in tackling design problems. Design problems are complex, iterative, and involve multiple stakeholders—clients, professionals, and communities. Solutions are often subjective and require nuanced conversation, negotiation, and reflection. Traditional computer approaches can fall short in addressing these richly contextual, visual, and interactive exchanges.
By combining conversational AI and visual communication techniques, our research seeks to make design dialogue more intuitive, accessible, and effective. The project will examine how architects, city planners, and community members can discuss, sketch, and critique design concepts with AI inside familiar chat interfaces.
The coming years promise exciting developments as we share findings, tools, and resources with the academic community, design practitioners, and the public. Stay tuned—there’s much to come!
]]>At AAIRL, we’re excited to spotlight a compelling step forward in architectural design, led by a collaboration between Gal Guz, Nikolas Martelaro, Gerhard Schubert, and myself. Our new research paper, “Quantifying Architectural Experience Using Visual Language Models: Does AI Dream of Rendered Spaces?”, asks a fascinating question: Can advanced AI models predict how humans will emotionally respond to a space, even before it’s built?
Architecture has long depended on measurements: dimensions, materials, light. But translating the emotional impact of a space (“Is this cozy? Stimulating? Overwhelming?”) into numbers has been elusive. Each person feels space differently, which makes universal design both crucial and challenging.
Recent developments in large language models (LLMs) and visual language models (VLMs), such as GPT-4o, raise a new research possibility: using AI systems to approximate selected aspects of human response to architectural images. This does not replace human evaluation, but it may provide an additional source of early-stage design feedback.
The Experiment: AI vs. Human Ratings: We evaluated GPT-4o’s image-to-emotion ratings against human responses. We gathered 10 images of different spaces and asked both human participants and the AI model to complete emotional ratings using the well-known PANAS scale, which measures 20 affective states.
We found out:
This work suggests several possible directions for future design research:
This methodology could pave the way for large-scale studies on emotional response to the built environment—including cultural or demographic adjustment.
This research suggests that AI may help designers measure and, eventually, optimize aspects of emotional response to space. If “form follows function,” perhaps design evaluation can also begin to account more systematically for feeling.
We at AAIRL are eager to see how these tools empower more human-centric architecture—making the intangible, finally, a little more measurable.
]]>Two papers from AAIRL have been accepted for presentation at CAADRIA 2025 in Tokyo. The papers reflect two active strands of the lab’s work: AI-supported urban renewal and computational methods for architectural design.
Authored by Dr. Haya Brama, Dr. Jonathan Dortheimer, Prof. Tal Grinshpoun, and our industry partner, Oded Landau, this paper introduces a decision-support approach for urban renewal planning. The team has developed a system that uses algorithmic and machine-learning methods to help municipal stakeholders explore alternative parcel combinations beyond the constraints of existing urban layouts.
The core of this system is its modular plugin architecture, which supports flexibility and transparency in decision-making. Custom-designed algorithmic solutions within these plugins address urban renewal’s specific and nuanced requirements. Moreover, deep learning techniques are employed to predict the potential of future projects based on historical data, a feature particularly relevant for peripheral regions with low-profit margins. Successful renewal in these areas can catalyze further growth and development, making this model a useful tool for urban planners.
The second paper, “Addressing Religious Architectural Restrictions with Computer Code: A Genetic Algorithm Approach,” by Gaya Bitman and Dr. Jonathan Dortheimer tackles the unique architectural challenge of designing sukkah balconies in high-rise multi-residential buildings. Jewish law requires each balcony to have an unobstructed view of the sky, a stipulation that complicates architectural design and limits building heights in densely populated areas.
This study models the design problem as a single-objective optimization task using a Genetic Algorithm (GA), a popular method in parametric architecture. However, the research identifies significant limitations in the GA’s application to design, such as rapid convergence that restricts solution exploration and difficulties in meeting specific minimum requirements. To address these challenges, the authors propose an enhanced approach that involves executing multiple parallel, more minor GAs to improve random sampling. Additionally, they introduce an evaluation function distinct from the fitness function, allowing for selecting solutions that meet critical minimum criteria. This refined method offers a more effective strategy for designing sukkah balconies and demonstrates an improved application of GA in architectural design.
See you in CAADRIA 2025 at Tokyo University!
]]>Urban renewal in Israel is a complex challenge that requires creative solutions to improve national resilience, improve citizens’ lives, and preserve natural open spaces through urban densification. Green Group will support AAIRL research on deep learning methods for identifying promising locations for urban renewal interventions. The project aims to help governments, cities, architects, and developers make more informed, data-driven decisions by analyzing urban data and estimating where renewal projects may catalyze broader investment and development.
Green Group has long been at the forefront of urban renewal in Israel, known for its bottom-up, community-driven approach to creating modern, sustainable urban environments. Green Group reaffirms its commitment to Israeli innovation and its vision of leading the market in urban renewal by investing in urban science research.
Contact us to learn more about research opportunities.
]]>The project team comprises distinguished experts, including Prof. Tal Grinshpun from the Department of Industrial Engineering and Management at Ariel. As the project’s co-researcher, Prof. Grinshpun brings a wealth of expertise in artificial intelligence algorithms and optimization problems. Dr. Haya Berma, a seasoned researcher with a rich background in psychology and machine learning algorithms, will spearhead the project, providing invaluable insights into the social aspects of urban life.
The project will leverage data from local authorities, including Tel Aviv, Hadera, and Rehovot, to develop this innovative model. It is anticipated to bring about a paradigm shift in Israel’s urban planning and development strategies.
]]>The project seeks to make a diagnosis and propose projects on a regional scale within the Probogotas region’s Metropolitan Vision 2051 framework. This project aims to be a voice for the people of Bogota and let them be a part of the development of their communities.
Our Arc. Orly Moas-Cohen, Gilad Chalfon, and Ela Moyal will present and discuss each of their projects:
Orly’s project investigates how natural disasters can be an opportunity for urban renewal, using places like Hiroshima, Japan, and New Orleans, Louisiana, as case studies.
Ela’s research gives a new perspective regarding urban revitalization in Tel Aviv, looking into and assessing the implementation, implications, and prospects of facade renovation projects.
Finally, Gilad will present a study investigating how “Smart” parking applications can be tools in tech-based privatization and part of a broader control mechanism.
We are excited to see all the scholars at the conference and discuss our research!
]]>