Bliss Reading Group - May 11

Join us for our third session of the Bliss Reading Group! ✨ This week, we're delving into the fascinating topic of representational alignment, shifting focus from AI safety to how different information processing systems create similar internal representations of the world.

We'll be discussing the pivotal paper Getting Aligned on Representational Alignment (Sucholutsky et al., 2023), presented by Tom Neuhäuser. This insightful Perspective paper features contributions from over 30 authors across cognitive science, neuroscience, and machine learning.

Key questions we will address include:

• How can we measure whether a neural network and a human brain (or two neural networks, or two brains) view the same stimulus similarly?
• Does similarity in representation equate to similar behavior?
• Is it possible to adjust one system's representation to better align with another's?

Let's explore what we mean by wanting AI representations to be "aligned" with human perspectives. Is representational similarity the right goal, or is behavioral equivalence sufficient? How does this concept of alignment connect to the safety-oriented discussions from our previous sessions?

🌟 Join us for what promises to be a lively and insightful discussion!

Location: Merantix AI Campus
Tickets: Get your ticket here!