In his presentation, Alessandro Bozzon delves into the integration of AI in public decision-making, exploring its potential and challenges. He begins by introducing himself as a computer scientist working within the Faculty of Industrial Design Engineering. Bozzon emphasizes the importance of interdisciplinary approaches in developing AI systems, incorporating insights from sociology, psychology, and urban sciences. His primary focus is on public AI, which utilizes predictive and descriptive models to support decision-making and policy implementation in the public sector.
Bozzon explains that public AI involves algorithms that interact with real-world data, influencing human lives and creating an ecology of diverse agents and actors, including humans and AI systems. He illustrates this with examples like traffic systems, where smart traffic lights and autonomous cars negotiate the flow of traffic, affecting human behavior and urban rhythms. He highlights the subtle ways in which autonomous systems, such as AI-enabled trash collection or Airbnb monitoring, impact societal routines and policies.
A key theme in Bozzon's presentation is the need for designing AI systems that are purpose-fit from the start, rather than creating technology first and addressing its consequences later. He advocates for a methodology where the design process begins with a clear understanding of the desired outcomes and the contextual realities of the environment in which the AI will operate. This approach acknowledges the limitations of data, emphasizing that data is always an approximation of reality, not a perfect representation.
Bozzon introduces the concept of "contestable AI," which is AI that is open to dispute and feedback throughout its lifecycle. This involves creating systems that are transparent and responsive to public input, ensuring that the systems align with societal values and are perceived as legitimate by those affected. He discusses the importance of involving citizens, policymakers, and developers in the design and monitoring of AI systems, expanding the traditional loops of interaction and making the process more democratic and inclusive.
He provides a practical example with the camera car used for monitoring traffic and public spaces. Bozzon explains how such systems, when designed with contestability in mind, can be more responsive and fair, as they allow for public participation in their development and operation. This participatory approach helps bridge the gap between the digital and physical worlds, ensuring that AI systems do not disrupt but rather enhance societal functions.
Bozzon also touches on the ethical and technical complexities of implementing public AI. He emphasizes the need for rich methodological portfolios that combine ethnographic and data-driven research to understand and predict human behavior accurately. This interdisciplinary approach helps create AI systems that respect and integrate the values, rhythms, and properties of the communities they serve.
In conclusion, Bozzon underscores the importance of contestability as a core property of public AI systems. He argues that making AI systems open to dispute and responsive to public input not only increases their perceived legitimacy but also ensures they are better aligned with societal needs and values. He calls for a shift from merely building advanced models to focusing on problem formulation, system design, and the integration of diverse perspectives in the AI development process. Bozzon's vision for public AI is one where technology serves the public good, operating transparently and democratically, with continuous feedback and improvement based on public participation.