This Open-Source Launch Symposium presents the Thousand Brains Project, with Jeff Hawkins providing an overview of how distributed sensorimotor learning and reference frames enable intelligence in AI and robotics. It features talks from Viviane Clay and Niels Leadholm on the concrete implementation of the system, a clear development roadmap, opportunities for community contributions, and a focused Q&A session addressing technical and practical questions.
00:00 Welcome
00:29 Agenda
00:59 Administration
01:33 Introduction: Jeff Hawkins
02:30 Introduction: Viviane Clay
03:02 Introduction: Niels Leadholm
03:32 The Team
04:22 Jeff Gives an Overview of the Thousand Brains Theory
06:33 Sensorimotor Learning
07:06 Don't Focus on Output
08:24 Focus on How Knowledge is Represented
08:42 Brains are Sensorimotor Learning Systems
09:42 Location is Key
11:17 Reference Frames in a Single Column
13:35 The Organ of Intelligence
15:22 Thousands of Brains
16:46 Columns Vote
19:07 Any Kind of Sensor
20:24 The Future of AI
22:39 Monty: Our First Implementation of a Thousand Brains System
24:42 What is a Thousand Brains System
25:52 Key Principles of Thousand Brains Systems
29:01 Cortical Messaging Protocol
34:50 Sensor Modules and Learning Modules
43:23 Voting for Faster, More Robust Inference
44:09 Components in Thousand Brains Systems
53:03 Motor Policies
58:19 Model-Free vs. Model-Based Policies
01:09:58 Current Capabilities & Experiments
01:18:41 Future Capabilities
01:28:44 Applications
01:35:48 Community Engagement
01:46:34 Roadmap
01:49:07 Request for Comments (RFCs)
01:52:07 The Future
01:53:55 Questions and Answers
01:54:37 Integration of Abstract Concepts That Lack Direct Sensory Motor Grounding
01:56:20 How does TBT Differ From HTM?
01:59:10 Will Scaling These Systems Surpass Human Intelligence
02:01:12 Concepts Cut From Continuous Stream of Features at Locations
02:04:14 Is Language Important to the Monty system?
02:07:48 How Does Monty Deal With Time Series Data
02:12:13 When Learning Modules Vote They use Their own Independent IDs
02:16:32 What is the Learning Algorithm Being Used?
02:18:18 Extracting other Features such as Planar Intersections
02:20:10 Does Monty Require Neuromorphic Hardware for Better Performance?
02:21:42 How Does Monty Represent Itself in the World?
02:23:33 Implications of Bringing This Technology to the World
