Tensor Logic "Unifies" AI Paradigms [Pedro Domingos]

Pedro Domingos, author of the bestselling book "The Master Algorithm," introduces his latest work: Tensor Logic - a new programming language he believes could become the fundamental language for artificial intelligence.

Think of it like this: Physics found its language in calculus. Circuit design found its language in Boolean logic. Pedro argues that AI has been missing its language - until now.

*SPONSOR MESSAGES START*
—
Build your ideas with AI Studio from Google - http://ai.studio/build
—
Prolific - Quality data. From real people. For faster breakthroughs.
https://www.prolific.com/?utm_source=...
—
cyber•Fund https://cyber.fund/?utm_source=mlst is a founder-led investment firm accelerating the cybernetic economy
Hiring a SF VC Principal: https://talent.cyber.fund/companies/c...
Submit investment deck: https://cyber.fund/contact?utm_source...
—
*END*

Current AI is split between two worlds that don't play well together:

Deep Learning (neural networks, transformers, ChatGPT) - great at learning from data, terrible at logical reasoning
Symbolic AI (logic programming, expert systems) - great at logical reasoning, terrible at learning from messy real-world data

Tensor Logic unifies both. It's a single language where you can:
Write logical rules that the system can actually learn and modify
Do transparent, verifiable reasoning (no hallucinations)
Mix "fuzzy" analogical thinking with rock-solid deduction

The Killer Feature: The Temperature Knob

Why Should You Care?

Pedro makes a provocative claim: 01:24:50 → 01:27:47 “We've wasted trillions of dollars” on brute-force compute because we're ignoring 40 years of AI research. Companies are "reinventing reasoning" when they could just read a textbook and save billions.

INTERACTIVE TRANSCRIPT:
https://app.rescript.info/public/shar...

TOC:
00:00:00 - Introduction
00:04:41 - What is Tensor Logic?
00:09:59 - Tensor Logic vs PyTorch & Einsum
00:17:50 - The Master Algorithm Connection
00:20:41 - Predicate Invention & Learning New Concepts
00:31:22 - Symmetries in AI & Physics
00:35:30 - Computational Reducibility & The Universe
00:43:34 - Technical Details: RNN Implementation
00:45:35 - Turing Completeness Debate
00:56:45 - Transformers vs Turing Machines
01:02:32 - Reasoning in Embedding Space
01:11:46 - Solving Hallucination with Deductive Modes
01:16:17 - Adoption Strategy & Migration Path
01:21:50 - AI Education & Abstraction
01:24:50 - The Trillion-Dollar Waste

REFS
Tensor Logic: The Language of AI [Pedro Domingos]
https://arxiv.org/abs/2510.12269
The Master Algorithm [Pedro Domingos]
https://www.amazon.co.uk/Master-Algor...
Einsum is All you Need (TIM ROCKTÄSCHEL)
https://rockt.ai/2018/04/30/einsum
   • Open-Ended AI: The Key to Superhuman Intel...  
More Is Different [P. W. Anderson] (not "Ross" we misremembered name in interview)
https://www.tkm.kit.edu/downloads/TKM...
Autoregressive Large Language Models are Computationally Universal (Dale Schuurmans et al - GDM)
https://arxiv.org/abs/2410.03170
Memory Augmented Large Language Models are Computationally Universal [Dale Schuurmans]
https://arxiv.org/pdf/2301.04589
On the computational power of NNs [95/Siegelmann]
https://binds.cs.umass.edu/papers/199...
Sebastian Bubeck
  / openai_researcher_sebastian_bubeck_falsely...  
I am a strange loop - Hofstadter
https://www.amazon.co.uk/Am-Strange-L...
Stephen Wolfram
   • Mystery of Entropy FINALLY Solved After 50...  
The Complex World: An Introduction to the Foundations of Complexity Science [David C. Krakauer]
https://www.amazon.co.uk/Complex-Worl...
Geometric Deep Learning
   • GEOMETRIC DEEP LEARNING BLUEPRINT  
Andrew Wilson (NYU)
   • The Real Reason Huge AI Models Actually Wo...  
Yi Ma
  / yi-ma-scientific-141953348  
Roger Penrose - road to reality
https://www.amazon.co.uk/Road-Reality...
Artificial Intelligence: A Modern Approach [Russel and Norvig]
https://www.amazon.co.uk/Artificial-I...

Best Moments:
01:01:50 → 01:02:15 [The Universal Induction Machine] - Pedro's quest for the "Turing Machine of Learning"
00:20:41 → 00:24:37 [Predicate Invention] - How the system learns to see "objects" instead of pixels, like humans do
01:12:55 → 01:13:30 [Why This Matters Now] - The hallucination problem explained