The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Creating Robust Language Representations with Jamie Macbeth - #477
7 snips
Apr 21, 2021 
Jamie Macbeth, an assistant professor at Smith College focusing on cognitive systems and natural language understanding, dives into his unique approach to language representation. He critiques misconceptions in AI while advocating for using handcrafted models to understand human intelligence. The conversation touches on the limitations of deep learning in grasping linguistic nuance and the need for innovative evaluation metrics. Jamie also explores how pre-linguistic structures contribute to common sense knowledge and discusses the future of AI in enhancing reasoning through episodic memories.
AI Snips
Chapters
Transcript
Episode notes
Cognitive Systems vs. AI
- Cognitive systems research aims to understand human intelligence through AI.
- This contrasts with AI research focused solely on benchmark performance.
Scripts Example
- Roger Schank's work on "scripts" in AI has influenced other fields like cognitive psychology.
- Social scientists often use the concept without knowing its AI origins.
NLU Needs Thought Representation
- Natural language understanding requires more than just processing words and grammar.
- Systems must represent, manipulate, and understand ideas like humans.