In this session, Professor Niu Xinyi from Tongji University will provide an in-depth interpretation of how generative artificial intelligence (GAI) injects new momentum into urban planning, exploring related foundational theories, cutting-edge perspectives, and the impact of educational reform in this field.
The Disruptive Nature and Applications of AIGC: Revealing How Generative AI Reshapes Planning Futures
In early 2024, the release of Sora sparked heated discussions, bringing generative artificial intelligence back into the public spotlight. In what aspects does the disruptiveness of generative AI manifest?
Niu Xinyi:Generative artificial intelligence technology has attracted widespread attention across the industry due to its unique disruptive nature. Although AI technology has a long history, it has only recently become a focal point of public discussion. This technology is termed disruptive because it fundamentally differs from traditional reasoning or discriminative AI technologies. Traditional AI technologies primarily focus on solving specific problems or optimizing tasks, while generative AI technologies can create entirely new content based on their intrinsic logic and knowledge base. This capability allows AI to produce knowledge and output content like humans, especially supported by large models such as ChatGPT and Sora, whose generated images, texts, and videos have reached a level that is difficult to distinguish from reality. Additionally, another disruptive feature of generative AI technology is its low barrier to entry. It enables people across society to easily register and use these models to quickly generate various content.This technology’s popularity has not only profoundly impacted the field of planning but has also sparked a revolution across all sectors of society. The disruptiveness of generative AI technology is reflected in its creativity, realism, and ease of use, changing our traditional perceptions of computer-generated content and demonstrating its immense potential and influence in various fields.
You just mentioned the enhancement of planning decisions through ChatGPT and Sora. Returning to the planning decisions themselves, how should the training of general generative AI algorithms integrate planning professional knowledge to form intelligent and scientific knowledge, enhancing their planning decision-making capabilities?
Niu Xinyi:Generative artificial intelligence technology, especially general AI models, has already shown its application potential across multiple fields. These models accumulate a vast amount of interdisciplinary knowledge through extensive pre-training, forming a general capability to handle various tasks. However, to further enhance the scientific and intelligent levels of planning decisions, we need to further develop specialized large models for the planning discipline based on these general models.
Currently, the industry widely agrees that the construction of specialized models should be based on general models, undergoing secondary training or specific forms of learning to meet the needs of specific professional fields. The key to this process lies in how to effectively integrate the specialized knowledge of the planning discipline into the general model, transforming it into a large model with specialized characteristics. Through this approach, generative AI can not only provide more precise and planning-specific decision support but also promote the intelligent and scientific nature of the planning decision-making process, thus playing a more significant role in the field of planning. This marks a new developmental stage for the application of AI technology in planning decision-making.
How should general generative AI combine with planning professional models to enhance the intelligence and scientific level of planning decisions?
Niu Xinyi:When discussing the combination of general generative AI and planning professional models, we first need to understand the two different modeling paths. Traditional planning discipline modeling is based on explicit planning theoretical knowledge, transformed into algorithms through principled formulas, thereby achieving the construction of computer programs. This top-down modeling approach emphasizes the leading role of planning theory and the quantification of knowledge. In contrast, generative AI technology adopts a data-driven modeling approach, which does not rely on pre-existing knowledge systems but instead extracts features and establishes internal models through machine learning from large amounts of data. This bottom-up modeling process does not require explicit planning principles but learns and generates knowledge through the data itself.
Combining these two modeling approaches presents both challenges and opportunities. Knowledge-driven models have clarity and interpretability in solving specific problems, while data-driven models exhibit robust capabilities in handling complex, nonlinear issues.Exploring the integration of both will help us open new research paths and enhance the intelligence and scientific nature of planning decisions.
Currently, although the industry has yet to form a mature integration model, this direction is undoubtedly worth exploring in depth. In the future, we may need to find a balance between knowledge-driven and data-driven modeling approaches, achieving complementary advantages between the two. This integration can not only promote the development of the planning discipline but also provide new perspectives and tools for solving complex issues in the real world.
Planning Decisions and Urban Governance: AI Assistance, A New Chapter in Smart Planning
With the assistance of generative artificial intelligence, what are the key nodes for achieving full-process decision support in the planning governance of megacities like Beijing and Shanghai?
Niu Xinyi:In exploring the application of generative artificial intelligence in the planning governance of megacities, we first need to clarify the knowledge mastered by the general model and the supplements required by the professional model. The reason AI can creatively generate content stems from its deep learning and mastery of knowledge in specific fields. In the professional field of urban planning, the specificity and depth of knowledge are particularly crucial.
One key node is how to effectively extract, express, and convert the unique knowledge system of urban planning into a form that models can accept. This requires a profound understanding and precise grasp of specialized knowledge in the field to allow the general model to integrate and utilize this knowledge through retraining or specific training.
Another key point is to stimulate existing planning-related knowledge within the general model. Although this knowledge may only account for a small portion in large-scale training, specific learning in the professional field can prompt the model to recognize and utilize this knowledge, thereby better serving urban planning governance.
In other words, integrating specialized knowledge and stimulating the model’s potential relevant knowledge in the field are key exploration directions for achieving full-process decision support in megacity planning governance. This requires not only technological innovation but also a deep understanding and application of the planning discipline.
You just mentioned the full process of planning decision-making assisted by generative artificial intelligence, and public participation is an important part of planning. Can you elaborate on how large language models like ChatGPT can assist public participation in planning?
Niu Xinyi:Large language models like ChatGPT have significant potential in public participation in urban planning. They can not only understand and respond to professional planning questions, but also provide an interaction experience that is closer to human communication through their natural language processing capabilities. These models can accept questions in natural language and provide humanized responses, making them ideal tools for popularizing planning knowledge and introducing planning processes.
Through professional training, large language models can be deployed in public domains, such as government planning service windows, providing efficient consulting services and guiding individuals through relevant processes and steps. Additionally, they can also collect and analyze opinions and suggestions from residents or participants in urban planning, helping planners and decision-makers better understand public needs and reactions. For instance, large models can quickly organize targeted meeting minutes during planning discussions, clarifying consensus and points of divergence. This capability greatly enhances the efficiency of information collection and feedback, contributing to improved public participation outcomes.
You just discussed how generative artificial intelligence can assist public participation in planning from both top-down and bottom-up approaches. How can generative AI large models empower social forces in urban governance, such as in the renovation of urban villages, and promote communication between urban planners and communities, thus advancing a more participatory planning process?
Niu Xinyi:Generative AI large models hold significant empowering potential in urban governance, especially in projects like urban village renovations. Through “participatory design” or “participatory updates,” large models can facilitate communication and collaboration among multiple stakeholders, achieving a more democratic and transparent planning process.
In this process, large models can process and transform language expressions and visions from participants of diverse backgrounds, particularly converting the vague needs of non-professionals into professional planning terminology. Through professional training, large models can generate images or videos depicting the updated urban space based on input professional vocabulary or needs, providing residents and planners with an intuitive, discussable view of the updated effects. Moreover, the application of large models can accelerate the visualization process of update plans, allowing the effects of small-scale area updates to be quickly presented, thereby promoting consensus formation among all parties regarding the update proposals. The application of this technology not only improves planning efficiency but also makes the planning process more participatory and inclusive. Generative AI large models build a bridge for communication between planners and social forces, fostering a more participatory planning process and offering new ideas and tools for urban renewal and governance.
The Reform of AIGC in Planning Education: Cultivating Intelligent Planning Talent for the New Era
You’ve discussed a lot about how generative artificial intelligence empowers urban planning practitioners. As a university professor, how do you think the future education of urban planning should reform itself in conjunction with generative AI technology? Or how should we train students using generative AI?
Niu Xinyi:The integration of generative artificial intelligence technology provides new perspectives and methods for reforming urban planning education. In higher education, this technology has been explored for applications in facilitating teaching and improving learning efficiency. For example, Tsinghua University has utilized AI as a teaching assistant, enhancing the effectiveness of eight courses. In actual teaching processes, students practice content generation using large models, exploring and deepening their understanding of issues based on keyword input, thus providing new pathways for knowledge transmission.
In the future, simple repetitive tasks can be handled by AI teachers, while human teachers should focus on inspiring students’ creative thinking and guiding them to explore unknown fields, creating new knowledge that AI has not yet touched. Given that the capabilities of generative AI are currently limited to learning and integrating existing knowledge, educators need to consider how to cultivate students’ innovative abilities in the AI era.
The application of AI in fields like artistic creation demonstrates its ability to imitate and integrate, but true breakthrough innovations still require human involvement. Therefore, urban planning education should leverage general AI tools to enhance the efficiency and effectiveness of knowledge transmission while emphasizing the cultivation of students’ innovative thinking and problem-solving abilities. This exploration of educational reform will soon unfold across educational institutions at all levels, heralding a significant transformation in educational models.
As a professor in the planning discipline, what advice do you have for future students or students in the planning profession in the context of the rapid development of generative artificial intelligence?
Niu Xinyi:We are on the brink of the second digital transformation of the planning discipline, and the rise of generative artificial intelligence heralds the arrival of a new era. From computer-aided design (CAD) in the 1990s to modern AI technology, the planning industry continues to undergo technological innovations. Currently, students are beginning to explore using AI models to enhance efficiency in design, writing, and programming, which not only changes their working methods but also poses new requirements for education and industry practice.
For students in the planning profession, the first step is to actively embrace new technologies. AI is not a tool to replace planners but a partner to enhance work efficiency, optimize design, and analysis. Students should understand and master the applications of AI technology in the planning field through learning and practice, as this will become a fundamental skill in their future careers. Secondly, students need to cultivate critical thinking abilities. In the information and suggestions provided by AI, they must learn to filter, evaluate, and distill genuinely valuable content to assist in planning decisions and design innovations.Finally, and importantly, is to maintain a focus on humanistic care. The core of urban planning is to serve people, and the development of technology should always revolve around improving people’s quality of life. Students should leverage their professional knowledge and skills to address community needs and create more livable and human-centered urban spaces. We look forward to the development prospects of intelligent planning while recognizing its potentially significant role in academic research and theoretical methodological systems. In the wave of new technologies, let us embrace change together and work hand in hand to create a better urban future.
City Wall Cup Message: Looking Ahead, Technology Leading the Path of Planning Innovation
The finals of this year’s City Wall Cup Competition have concluded. From the perspectives of industry and discipline development, in the future, in which key directions and technologies can the competition expand and extend?
Niu Xinyi:The City Wall Cup Competition has successfully held its eighth edition, making significant contributions to promoting digitalization and informatization in the field of urban planning. This year, we witnessed the rapid integration of new technologies and methods in the entries, particularly the application of generative AI large models, demonstrating participants’ immense enthusiasm for technological exploration and innovation. Looking ahead, the expansion and extension of the City Wall Cup Competition should focus on two main directions: 1) Practicality and Applicability: The competition should emphasize the practicality and applicability of planning decision support models, closely integrating with industry needs. In the context of land space planning reform, smart planning has become a key direction for industry development, and the competition should encourage works that solve actual planning problems, innovating based on demand; 2) Technological Exploratory Nature: With the rapid development of AI technology, the competition should encourage participants to delve into and apply cutting-edge technologies, such as language large models, image large models, and graph neural networks, to promote theoretical development and technological advancement in the urban planning discipline.
The City Wall Cup Competition should continue to play its leading role in the industry, promoting the development of models with practical application value through close integration with national policies and industry needs, while exploring the application potential of new technologies in urban planning.We look forward to the future competition further strengthening its characteristics, promoting technological progress in the industry, and making greater contributions to the scientific development of urban planning.
This article is the “online premiere” of the journal “Beijing Planning and Construction”.
