RESEARCH · RESEARCH · #3718
MIT tenures Jacob Andreas; he researches models that explain themselves
Jacob Andreas, a newly tenured associate professor in MIT’s EECS and a member of CSAIL, develops techniques that enable deep learning models to communicate in natural language about what they learned from training data, aiming to improve interpretability, robustness, and human–AI collaboration. His group studies how machines acquire and express linguistic understanding and how to make training more natural and efficient.
KEY POINTS
- Jacob Andreas, a newly tenured associate professor in MIT’s EECS and a member of CSAIL, develops techniques that enable deep learning models to communicate in natural language about what they learned from training data, aiming to improve interpretability, robustness, and human–AI collaboration.
- His group studies how machines acquire and express linguistic understanding and how to make training more natural and efficient.
- If models can accurately describe their internal knowledge in human language, researchers gain a practical tool for debugging, auditing, and safer deployment of high‑stakes AI systems.
WHY IT MATTERS
If models can accurately describe their internal knowledge in human language, researchers gain a practical tool for debugging, auditing, and safer deployment of high‑stakes AI systems.