From "Steering" to "Speaking": The 5 Breakthroughs of Intelligence

The Evolutionary Stack

Welcome to the seventh installment of our series. As we explore the future of artificial intelligence we must understand the precise sequence of events that created biological cognition. Technology leaders often view the human mind as a single monolithic entity. However evolutionary psychology reveals a different reality. Our intelligence is actually a cumulative stack of five distinct software updates.

Max Bennett details these chronological updates in his book A Brief History of Intelligence. By mapping these five breakthroughs we can identify exactly what modern data science has achieved and what it is still fundamentally missing.

The Five Biological Breakthroughs

Evolution did not design the human brain all at once. It layered new capabilities on top of older systems over hundreds of millions of years.

Breakthrough One is Steering. Over five hundred million years ago early bilaterians developed the first centralized nervous systems. Their only goal was physical navigation. They learned to move toward food and away from danger in a three dimensional space.

Breakthrough Two is Reinforcing. Early vertebrates like fish developed the dopamine system. This allowed them to learn from trial and error. Instead of just reacting to the immediate present they could associate actions with delayed rewards. This is the biological origin of reinforcement learning.

Breakthrough Three is Simulating. Mammals evolved the neocortex. This monumental upgrade gave them an internal world model. They could simulate potential futures in their minds before taking physical action allowing them to plan and avoid fatal mistakes.

Breakthrough Four is Mentalizing. Primates developed Theory of Mind. They learned to model the internal emotional and cognitive states of other members of their group. This psychological leap allowed for complex social hierarchies deep collaboration and strategic deception.

Breakthrough Five is Speaking. Only very recently humans developed the capacity for syntactic language. This allowed us to share our internal world models with others and accumulate abstract knowledge across generations.

The Artificial Intelligence Shortcut

When we look at this evolutionary stack from a data science perspective we uncover a startling philosophical truth. Computer scientists did not build artificial intelligence in the same order that nature built biological intelligence.

We essentially skipped the first four steps and jumped straight to step five. Large Language Models are phenomenal speaking machines. They have mastered the statistical rules of human syntax and vocabulary. However they lack the foundational layers of steering reinforcing simulating and mentalizing.

This architectural shortcut explains the paradox of modern AI. It explains why a language model can write a brilliant Python script but cannot understand that a physical object drops when you let it go. It has no physical world model. It also explains why chatbots struggle with empathy and alignment. They have no biological framework for mentalizing or understanding the unstated social desires of human beings. We built the roof of the cognitive house before we poured the physical foundation.

Implications for Technology Leaders

For CEOs and CTOs mapping out enterprise AI strategies this timeline is your ultimate roadmap. It tells you exactly where current technology will excel and where it will inevitably fail.

You can confidently deploy generative models for tasks that rely entirely on language document processing and pattern recognition. However you must realize that true Artificial General Intelligence requires the entire evolutionary stack. The companies that eventually solve robotics and autonomous reasoning will not just scale up language models. They will be the ones who successfully reverse engineer the first four biological breakthroughs.

Takeaway

Biological intelligence evolved through five distinct leaps. These leaps are steering reinforcing simulating mentalizing and speaking. Modern machine learning has achieved the final leap of speaking but completely skipped the foundational physical and social layers. To build robust and safe artificial agents data scientists must look backward and engineer the foundational cognitive abilities that evolution perfected millions of years ago.

Next

We have mapped the architecture and the timeline of intelligence. But what actually drives an intelligent agent to explore the unknown? In our final article Curiosity as an Algorithm we will explore the mathematics of intrinsic motivation. We will discover how evolution programmed mammals to seek out novelty and how data scientists are copying this exact mechanism to build artificial intelligence that teaches itself.

Series Parts

Series: The Evolutionary Blueprint of Artificial Intelligence

Theme 1: The Architecture of Intelligence

• 1. The "World Model" Gap: What ChatGPT Is Missing
• 2. Generative AI is Older Than You Think: The Brain as a Prediction Machine
• 3. Why Robots Can't Load the Dishwasher (Yet)

Theme 2: Learning Algorithms & Data

• 4. Dopamine is a Teaching Signal: The Biology of Reinforcement Learning
• 5. The Problem of "Catastrophic Forgetting"

Theme 3: The Future & Ethics of AI

• 6. The "Paper Clip Problem" and Theory of Mind
• 7. From "Steering" to "Speaking": The 5 Breakthroughs of Intelligence
• 8. Curiosity as an Algorithm; [next]