Speaker: Belén Saldías (Massachusetts Institute of Technology)

Title: Designing Values-Aligned, Community-Centered Online Spaces with Language Models

The seminar will be available for in-person and Zoom participation. To participate online, please email inquiry-cs@barnard.edu to receive the Zoom link.

As of 2023, internet users spent, on average, over six hours daily engaged in online activities, highlighting the pervasive influence of digital environments on human relationships and interactions. Analogous to physical environments that shape human behavior, online spaces exert a profound impact on how individuals engage with one another. However, unlike physical spaces, where individuals retain a degree of agency over their surroundings, digital experiences are largely dictated by opaque, centralized platforms, often resulting in adverse outcomes such as anxiety, polarization, and diminished trust.

This talk examines these challenges and explores opportunities to design online spaces that foster healthier, more constructive interactions. By leveraging advancements in AI, particularly large language models (LLMs), alongside the intentional design of user-centric tools and decentralized online communities, this work aims to shift decision-making power from platform designers to communities, advancing scholarship at the intersection of human-computer interaction and computational social sciences.

Specifically, this talk focuses on the design considerations and key findings from Odessa, a novel Decentralized Social Systems Application. Odessa, a speech-based social network, functions as an open-source sandbox for experimenting with algorithms for content moderation and ranking and is currently deployed with LLMs as the core of distributed governance. The talk concludes with strategies and early findings on leveraging large language models to elucidate differences among communities and bridge their divides. Finally, it envisions a future of values-aligned, community-centered online spaces and their transformative potential for how we connect and interact in digital environments.

Belén Saldías is a final-year PhD candidate at MIT, working at the MIT Center for Constructive Communication and the MIT Media Lab. Her research focuses on the human-centered design and evaluation of machine learning systems and human-AI collaboration. Her dissertation explores how language models can be used to reimagine decentralized governance and enhance user agency online. Her overarching goal is to make the internet a safer and more empowering space for children and historically marginalized communities, bridging machine learning and natural language processing with the social sciences.

In her TEDx Talk, “Does AI Hold the Same Values We Do?”, Belén examines the cultural biases embedded in tools like ChatGPT and reflects on the challenges of deploying AI at scale.

Before joining MIT, Belén served as an Adjunct Professor at Pontificia Universidad Católica de Chile, where she taught data science and computer science courses. She completed her master’s thesis on variational inference methods while working as a visiting research assistant at the Institute for Applied Computational Science at Harvard University. Beyond academia, she spearheaded the first machine learning engineering initiatives at one of Latin America’s largest retailers and continues to collaborate with AI-focused startups.