The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) What the Human Brain Can Tell Us About NLP Models with Allyson Ettinger - #483
May 13, 2021
In this engaging discussion, Allyson Ettinger, an Assistant Professor at the University of Chicago, dives into the intriguing intersection of machine learning and neuroscience. She shares insights on how brain research can enhance AI, particularly in natural language processing (NLP). The conversation highlights the importance of controlled evaluation methods and the challenges AI faces in truly understanding language. Ettinger also touches on the predictive abilities of NLP models and how they compare to human cognitive processing, revealing the ongoing quest to mimic brain functionality in AI.
AI Snips
Chapters
Transcript
Episode notes
Neuroscience-Inspired AI Assessment
- Neuroscience offers valuable insights for AI, particularly methods used to study the brain.
- These methods, focusing on controlled experiments and isolating variables, can enhance AI evaluation.
Adapting Brain Study Methods for NLP
- Allyson Ettinger proposed using methods from studying brain encoding of meaning for evaluating sentence embeddings in NLP.
- This involved controlling syntactic structure to target meaning and avoid confounding variables.
Brains and Deep Learning: Shared Challenges
- The brain, like deep learning models, can be influenced by unintended variables.
- Brain researchers use controls to address this, similar to how AI researchers aim to mitigate spurious correlations.