The Secret is Out: How we're Building AGI

2025-05-01 · Source: Peter’s Substack · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Advanced, medium

Summary

A team led by Peter Voss, who co-coined the term "AGI" in 2002, is pursuing a "Cognitive AI" approach to achieve adult-level Artificial General Intelligence. This method contrasts with large language models (LLMs), which the team believes inherently lack key abilities for true intelligence, such as conceptual knowledge validation and integrated learning, memory, and reasoning. Their Integrated Neuro-Symbolic Architecture (INSA) emphasizes real-time, autonomous learning monitored by metacognition. After generating over $100 million in enterprise revenue from commercializing core aspects, the 12-person team shifted back to pure AGI development in late 2024. They are training their system like a child, starting with 3-4 year-old cognitive abilities, with a goal to reach adult-level intelligence within 18-24 months, requiring approximately 50 additional personnel.

Key takeaway

For research scientists evaluating AGI development paths, you should consider approaches that prioritize integrated neuro-symbolic architectures and autonomous conceptual learning over purely statistical models. This perspective suggests focusing on systems capable of incremental, real-time knowledge validation and deep integration of cognitive functions, rather than scaling current LLM paradigms, to achieve robust, adult-level general intelligence.

Key insights

True AGI requires conceptual knowledge validation and integrated neuro-symbolic architecture, unlike statistical LLMs.

Principles

• AGI must learn and adapt autonomously.
• Integrate learning, memory, and reasoning deeply.
• Simple mechanisms drive complex intelligence.

Method

Develop AGI by incrementally teaching a neuro-symbolic system, starting from child-level cognition and progressing to autonomous adult-level learning, guided by "AI Psychologists" and iterative engineering.

In practice

• Strip hand-crafted rules for AGI development.
• Teach AI with grounded examples like a child.
• Validate milestones with cognitive tests.

Topics

• Artificial General Intelligence
• Cognitive AI
• Neuro-symbolic Architecture
• Autonomous Learning
• AI Development Strategy

Best for: Research Scientist, AI Researcher, AI Scientist, AI Engineer

Related on AIssential

• Position: Hippocampal Explicit Memory Is the Cornerstone for AGI — Integrating explicit memory into LLMs is essential for achieving Artificial General Intelligence by enabling higher-order cognitive functions.
• Uneven Evolution of Cognition Across Generations of Generative AI Models — Generative AI models exhibit uneven cognitive development, excelling in language but struggling with visual-perceptual reasoning.
• System 3 AI: No Humans Needed — System 3 AI integrates formal logic and verifiable code into LLMs for autonomous, complex reasoning beyond human feedback.
• Live Q&A With the Hosts: Measuring Progress Toward AGI - Cognitive Abilities Hackathon — Robust AGI evaluation requires targeted benchmarks isolating specific cognitive faculties, moving beyond general problem-solving.
• My research: a computational cognitive neuroscience perspective on alignment — The first takeover-capable AI will likely be augmented LLMs, requiring alignment strategies that account for emergent human-like cognitive capacities.
• Continuous Learning for Real Intelligence — Human-like adaptive learning, especially via language, is crucial for practical AGI in dynamic real-world settings.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Peter’s Substack.