The Gradient: Perspectives on AI

Subbarao Kambhampati: Planning, Reasoning, and Interpretability in the Age of LLMs

Feb 8, 2024

Subbarao Kambhampati, Professor of computer science at Arizona State University, discusses planning, reasoning, and interpretability in the age of LLMs. Topics include explanation in AI, thinking and language, scalability in planning, computational complexity in LLMs, and concerns about misinformation generated by LLMs.

01:59:03

Creator website

Episode guests

Subbarao Kambhampati

AI Summary

Highlights

AI Chapters

Episode notes

Podcast summary created with Snipd AI

Quick takeaways

The reasoning ability of large language models is better described as approximate retrieval rather than robust thinking, highlighting the limitations of language models in planning and reasoning tasks.

Understanding the role of explanations and mental modeling in AI systems is crucial for making their reasoning more understandable and interactive.

AI systems should focus on making their reasoning understandable through symbolic explanations, while external critiques and verification help improve plan accuracy and bridge the gap between machine reasoning and human understanding.

The scalability of planning problems depends on factors such as determinism, stochasticity, dynamism, and partial observability, and can be addressed through external critiques, domain-specific improvements, and efficient back-promoting with an external verifier.

Deep dives

Large language models and reasoning capabilities

Many claims have been made about the capabilities of large language models to reason and plan. However, studies have shown that the reasoning ability of these models is better described as approximate retrieval, rather than robust thinking. Suburo Kanbambati's group has demonstrated similar results, highlighting the importance of understanding the limitations of language models in planning and reasoning tasks.

Leveraging LLMs as Idea Generators for Planning Problems

01:20

Evolution of LLMs and Tacit Knowledge Tasks

01:20

Balancing Innovation and Familiarity in Language Development

01:22

Efficiency in Testing Methods and LLM Performance

01:20

Introduction

2min

Suburo Kanbambati's Background and Entry into AI

2min

The Role of Explanation in Human-AI Interaction

26min

Language, Thinking, and Communication

12min

Planning and Its Relationship with Scalability

11min

Computational Complexity and Planning in LLMs

28min

Syntax vs. Semantics in Language Models (LLMs)

28min

The Use of LLMs as Assistants and Concerns about Misinformation

4min

Implications of AI's Turn to Natural Science for Computer Science at Large

5min

In episode 110 of The Gradient Podcast, Daniel Bashir speaks to Professor Subbarao Kambhampati.

Professor Kambhampati is a professor of computer science at Arizona State University. He studies fundamental problems in planning and decision making, motivated by the challenges of human-aware AI systems. He is a fellow of the Association for the Advancement of Artificial Intelligence, American Association for the Advancement of Science, and Association for Computing machinery, and was an NSF Young Investigator. He was the president of the Association for the Advancement of Artificial Intelligence, trustee of the International Joint Conference on Artificial Intelligence, and a founding board member of Partnership on AI.

Have suggestions for future podcast guests (or other feedback)? Let us know here or reach us at editor@thegradient.pub

Subscribe to The Gradient Podcast: Apple Podcasts | Spotify | Pocket Casts | RSSFollow The Gradient on Twitter

Outline:

* (00:00) Intro

* (02:11) Professor Kambhampati’s background

* (06:07) Explanation in AI

* (18:08) What people want from explanations—vocabulary and symbolic explanations

* (21:23) The realization of new concepts in explanation—analogy and grounding

* (30:36) Thinking and language

* (31:48) Conscious and subconscious mental activity

* (36:58) Tacit and explicit knowledge

* (42:09) The development of planning as a research area

* (46:12) RL and planning

* (47:47) What makes a planning problem hard?

* (51:23) Scalability in planning

* (54:48) LLMs do not perform reasoning

* (56:51) How to show LLMs aren’t reasoning

* (59:38) External verifiers and backprompting LLMs

* (1:07:51) LLMs as cognitive orthotics, language and representations

* (1:16:45) Finding out what kinds of representations an AI system uses

* (1:31:08) “Compiling” system 2 knowledge into system 1 knowledge in LLMs

* (1:39:53) The Generative AI Paradox, reasoning and retrieval

* (1:43:48) AI as an ersatz natural science

* (1:44:03) Why AI is straying away from its engineering roots, and what constitutes engineering

* (1:58:33) Outro

Links:

* Professor Kambhampati’s Twitter and homepage

* Research and Writing — Planning and Human-Aware AI Systems

* A Validation-structure-based theory of plan modification and reuse (1990)

* Challenges of Human-Aware AI Systems (2020)

* Polanyi vs. Planning (2021)

* LLMs and Planning

* Can LLMs Really Reason and Plan? (2023)

* On the Planning Abilities of LLMs (2023)

* Other

* Changing the nature of AI research

Get full access to The Gradient at thegradientpub.substack.com/subscribe

Remember Everything You Learn from Podcasts

Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.

The Gradient: Perspectives on AI

Subbarao Kambhampati: Planning, Reasoning, and Interpretability in the Age of LLMs

Episode guests

Podcast summary created with Snipd AI

Quick takeaways

Deep dives

Large language models and reasoning capabilities

Suburo Kanbambati's research on planning and language models

The relationship between explicit and tacit knowledge tasks

The challenges of scalability in planning

The importance of external verification for language models

Using language models with human experts

Understanding the difference between human cognitive capabilities and language models

Implications of AI as an ersatz natural science for computer science

Leveraging LLMs as Idea Generators for Planning Problems

Evolution of LLMs and Tacit Knowledge Tasks

Balancing Innovation and Familiarity in Language Development

Efficiency in Testing Methods and LLM Performance

Remember Everything You Learn from Podcasts