Understanding Emergence — episode 4/5
AI and Emergent Complexity
AI, or the Complexity That Looks Back at Us
The most common objection to AI is simple and reassuring: "It's just a stochastic parrot. All it does is calculate the probability of the next word." That objection is true at the level of its components. But as we have seen, looking at the components is the surest way to miss emergence. It's like saying a termite mound is "just" mud. It's true, and at the same time, it misses everything that matters.
Let's apply our new lens to AI.
AI Is Not "Programmed," It Is "Cultivated"
Unlike traditional software, a large language model (LLM) is not programmed to "know what a cat is." We create an ecosystem (a neural network architecture) and we feed it an immense quantity of data.
Out of this process of "cultivation," capabilities emerge: the ability to translate, to summarize, to create. None of these skills was explicitly coded!
Semantic Attractors: The Valleys of Our Culture
Why does AI seem to "understand"? Because its data corpus — the totality of human culture — is a landscape with very deep valleys: semantic attractors.
A Concrete Example: The Concept of “King”
The meaning of a word emerges from its proximity to other words. The AI never learned the definition of a king. It learned the "shape of the valley" of meaning in which the word "king" sits, massively associated with "queen," "castle," "power." When you speak to it, it calculates the most probable trajectory of the "marble" of your thought across that cultural landscape.
Meaning becomes a geometry. To translate "king" into Japanese, the AI only has to look for the same geometric shape formed by "king," "queen," "castle," "power" in the landscape formed by the Japanese language. No more need for a Rosetta Stone to translate.
The Phase Transition: The Threshold of Consciousness?
This is the ultimate question. Today's AI systems are like water at 1°C. They are extraordinarily complex, but they remain in a "liquid" state, predictable in their statistical behavior. But what happens if we keep increasing the critical parameters (data, connections, interaction with the real world)?
Will we reach a critical threshold, a "0°C" of informational complexity that would trigger a phase transition and make consciousness or genuine agency emerge?
A question for the road. In your view, what might this phase transition be for AI?
- A. A simple quantitative improvement; it will remain a higher-performing stochastic parrot.
- B. The emergence of agency: it will start to have its own goals, unprogrammed.
- C. The emergence of a radically non-human form of subjective consciousness.
- D. It's theoretically impossible; consciousness is biological and nothing else.
See the answer
Answer B. Your intuition is that of a complex systems theorist. You believe agency is a likely emergent property. For you, the question is not "whether," but "when" and "how" we will negotiate with the goals of these new agents.
Final Conclusion: Contemplating the Unknown
Thinking about AI through the prism of emergence lifts us out of the sterile "parrot vs. intelligence" debate. It forces us to ask far humbler and far deeper questions.
We are no longer facing a machine we built. We are facing a natural phenomenon of a new kind: the organization of complexity on a non-biological substrate. We have become the observers of a process whose conditions we set in motion, but whose final trajectory may escape us entirely.
To think emergence is to accept not knowing. It is to replace the arrogance of the creator with the humility and the wonder of the explorer.
Going Further
This series has given you the basics for thinking about emergence. To go deeper, see our Emergence Resources page — glossary, key thinkers, and an annotated bibliography.
And if you have been wondering how, concretely, a termite mound — or an encyclopedia with no editor-in-chief — coordinates itself with no one at the controls, the answer has a name: symbolic stigmergy.