Upcoming Events

Free and open to the public!

When we interact with other people, we constantly attribute beliefs and desires to them to explain and predict what they do. Can we do the same with AI systems like ChatGPT and Claude?

Philosophers have long studied how we interpret the minds of others, and their tools turn out to be surprisingly relevant to understanding modern AI. But AI systems also present an unusual case: unlike with other people, we can crack open an AI system and look at its internal workings, yet we often struggle to make sense of what we find.

As AIs in the form of large language models (LLMs) continue to demonstrate remarkable abilities across various domains, computer scientists are developing methods to understand their cognitive processes, particularly concerning belief representation. However, the field currently lacks a unified theoretical foundation to underpin the study of belief in LLMs.

I will present work that begins to fill this gap by proposing adequacy conditions for a representation in an LLM to count as belief-like. Drawing from insights in philosophy and contemporary practices of machine learning, I’ll establish criteria informed by theoretical considerations and practical constraints. These conditions help lay the groundwork for a comprehensive understanding of belief representation in LLMs.

Join Darcy Otto, faculty in computer science and Director of BCAI, with Bennington alumns Beth Kanter ’79 and Adnan Iftekhar ’97 for a community conversation on how to approach artificial intelligence in ways that are human.

Drawing on Beth’s leadership in human-centered technology adoption and Adnan’s experience with AI across higher education and the corporate sector, the discussion will explore both the opportunities and the tensions AI introduces into educational and creative practice. Topics will include cognitive benefits and risks, ethical and environmental considerations, and the crucial distinction between using AI with our “brains on” versus “brains off.”

Free and open to the public!

Is this company’s AI model biased? Are its predictions reliable? Are they using my data responsibly? As AI is deployed in sensitive applications, it is increasingly important to audit models to ensure they uphold societal values. However, AI service providers almost never release their models or data for auditing due to intellectual property and data privacy issues.

My work aims to address this tension through privacy-preserving cryptographic ‘contracts’ which can bind service providers’ models. These contracts use zero-knowledge proofs and other cryptographic tools to guarantee that (i) the model satisfies an important property such as group fairness, robustness, or differential privacy; (ii) outside parties can view the contract to verify whether the model has the property, but they learn no information about the model parameters or data by doing so.