AGI Is the Wrong Question
Summary
Philip Rathel, CTO of Neo4j, argues that the 95% AI failure rate reflects healthy corporate checks and balances rather than technological failure. He explains that successful enterprise AI relies on composite, neurosymbolic systems that combine LLMs with deterministic graph databases to ensure accuracy and explainability. Rathel details how graph intelligence solves issues like hallucination and data silos in high-stakes fields such as healthcare, finance, and cybersecurity. The episode concludes with a framework for understanding different AI tools and the importance of human agency in dec…
Topics discussed
Intro: The 95% AI failure stat and guest Philip Rathel
Sponsor segments and show opening
Guest background: Neo4j and the 'left brain' of AI
Why AI projects fail: Checks, balances, and composite systems
LLM limitations: Black boxes, hallucinations, and lack of discernment
The AGI debate: Agency, sentience, and human control
Neurosymbolic AI: Combining probabilistic and rules-based systems
Deterministic use cases: Walmart, Uber, and medical paths
High-stakes domains: Zero tolerance for error in maintenance and fraud
Where LLMs struggle: Deep connectivity and multi-level calculations
Graph databases: Representing real-world networks and hierarchies
GQL standard and why models write better graph queries than SQL
Industry impact: Drug discovery, supply chain, and anti-fraud
Cybersecurity: Attackers think in graphs, defenders must too
Breaking data silos: Building an enterprise knowledge graph
The future of agentic memory and digital rights
Practical advice: Categorizing AI tools and maintaining human agency
Credits and sign-off
Listen ad-free on Castria