View source for AI and design: Sources of “truth”?

[[File:CBC-Logo.png|right|200px|link=CBC_2023]]
[[Category:CBC]]
[[Category:CBC2023]]
[[Category:CBC Proceedings]]
[[Category:Conferences]]

[[CBC 2023]] presentation by [[Sean Hanna]]. https://doi.org/10.47330/CBC.2023.KPGV9387 | Watch [[File:VideoRecord-Icon.png |Left|22px|link=https://www.youtube.com/watch?v=4TXLpXjYV68]] | [[File:Slides-Icon.png |Left|30px|link=https://www.dropbox.com/]]

=Abstract=
Sean Hanna's keynote at CBC 2023 explores the intersection of Artificial Intelligence (AI) and Building Information Modeling (BIM) in the context of design, focusing on the concept of managing uncertainty across various stages of a project. Hanna draws parallels between AI and BIM's objective to refine and rely on increasingly accurate models or "single sources of truth" to reduce errors. He advocates for a decentralized approach in utilizing AI in design, suggesting it as a method to manage the inherent uncertainties prevalent in the early stages of design processes.

Hanna references the Mlini Curve to discuss how early design phases carry significant uncertainty and potential for influence, alongside rising costs as projects progress. He presents examples spanning from pre-design to operational stages, demonstrating AI's capacity to tackle uncertainty throughout a project's lifecycle. For instance, he cites the use of neural networks in predicting building energy performance and computational fluid dynamics to model wind flow around structures more efficiently than traditional simulations.

A pivotal aspect of Hanna's argument is the distinction between risks, which are quantifiable uncertainties, and "true uncertainty," where probabilities cannot be scientifically determined. He points out that design frequently encounters true uncertainties or "wicked problems," which are not readily addressed by optimizing models or machine learning that seeks to fit patterns within existing data.

To illustrate the challenge of addressing true uncertainty with AI, Hanna discusses the variability in solving problems like protein folding versus capturing the first image of a black hole. The former, despite its complexity, allows for experimental validation, whereas the latter involves making inferences from irreplicable and distant phenomena, highlighting difficulties in validating AI-generated outcomes against a "ground truth."

Hanna also examines paradigm shifts in design, where a change in representation or model is required to address new challenges effectively. He uses examples from his work and others', emphasizing that certain creative leaps in design cannot be achieved through linear optimization or model fitting but require a fresh conceptual framework.

Central to Hanna's proposition is the use of decentralized systems like Speckle, which facilitates data transfer across disparate models without necessitating their reconciliation. This approach aligns with managing the multifaceted uncertainties in early design phases, suggesting a shift from centralized to distributed models as projects move from high uncertainty to greater certainty.

Concluding his keynote, Hanna underscores the need for further research into decentralized methods of handling design uncertainties, leveraging AI and technologies like blockchain. He suggests that such approaches could enable more fluid negotiation and management of uncertainties, supporting a more adaptive and responsive design process.

=Keywords=
[[AI in Design]], [[Decentralization in BIM]], [[Managing Uncertainty]], [[True Uncertainty vs. Risk]], [[Speckle for Data Transfer]].