This summary explores a breakthrough conversation with Adrian Kosowski, a physicist, mathematician, and computer scientist from the startup Pathway. Kosowski discusses a newly discovered paradigm in Artificial Intelligence that aims to connect the biological functioning of the human brain with modern Transformer technology. This approach shifts the focus from simple machine learning to complex machine reasoning.

The Missing Link: Biology Meets Technology

Kosowski and his team have identified a mathematical model that serves as a bridge between biological intelligence and artificial neural networks. This “missing link” has two equivalent interpretations: it can be viewed as a biological model following neurobiological rules, or as a standard AI model compatible with today’s computer hardware. This dual perspective allows for the creation of systems that do not essentially replicate biology cell-by-cell but mimic the outcome of biological structures—specifically, the emergence of complex reasoning.

Shifting Focus from Learning to Reasoning

For decades, AI research has prioritized “learning”—how a system acquires data. Kosowski argues the new frontier is “reasoning”—the process of creating a chain of thought to reach a decision. Human reasoning involves two key components: abstraction (generalizing from past experiences) and logic (deduction). While current AI models (like LLMs) are improving at logic, they lack the continuous, lifelong learning and self-correction derived from experience that humans possess. The goal is to build AI that doesn’t just predict the next word but deliberates to find the correct course of action.

Autonomous Safety and Reliability

A critical issue with current AI is its unpredictability over long periods of autonomous operation. A model might work perfectly for five minutes but fail after six. The proposed new paradigm aims to solve this by creating systems that are provably safe and consistent. Kosowski draws a parallel to hiring an employee: just as we observe a human’s work for a week to trust their long-term performance, we need AI that creates trust through consistent, human-like reasoning processes, even in untrained scenarios.

The Future of Civilization

The discussion concludes with a profound look at the future. Kosowski predicts that within two years, we will have a definitive answer to whether non-biological systems can truly reason like humans. If the answer is yes, AI may cease to be viewed merely as a tool and instead be seen as a continuation of human civilization—a system capable of collaborating with humanity to solve problems previously thought impossible due to their scale and complexity.

Mentoring question

If we successfully create AI that reasons, learns from experience, and behaves indistinguishably from humans, should we continue to regard it as a mere tool, or must we accept it as a legitimate extension of our civilization?

Source: https://youtube.com/watch?v=v-odCCqBb74&is=5WD1mic09xbNQpKY