Directing the symphony of intelligence – Orchestrating agentic AI part 4

Humans have two brain systems. Our minds operate on this brilliant duality, seamlessly shifting between fast, intuitive system 1 and slow, logical system 2.

System 1: Fast and intuitive (grabs candy at the checkout)
System 2: low and analytical (compares pension funds)

Agentic AI requires the same duality to be truly intelligent. Tony Seale draws the essential distinction:

Large Language Models (LLMs) = continuous, fuzzy intuition (System 1)
Knowledge Graphs (KGs) = discrete, crisp logic (System 2)

Think of it like Miles Davis, he could break every musical rule because he first mastered them. Enterprise AI needs both the improvisation of LLMs and the sheet music of knowledge graphs. Discrete semantics don’t limit continuous creativity, they empower it.

When AI “hallucinates,” it’s doing what our brain does: making educated guesses. It’s riffing like Miles, daring, unpredictable, sometimes playing outside the changes. The problem isn’t the creativity itself, it’s the lack of grounding in real-world signals. The cure isn’t to muzzle the trumpet, it’s to provide the sheet music

Science backs the symphony

Recent work by Alessandra Mileo et al. (2025) lays out a neuro-symbolic cycle for human-centred explainability: neural networks improvise, a symbolic layer scores the tune, and human experts validate and reinject knowledge. Exactly the System 1 + System 2 duet we’re advocating.

Hybrid approaches are proving their worth on different battlefields

Military intelligence: The U.S. Army pairs AI algorithms with human commanders in live decision-making. Battle algorithms rapidly propose tactical options while human leaders apply strategic context and ethical oversight.
Freestyle chess: The 2014 Intagrand victory tells the story perfectly. Amateur players with average computers defeated grandmasters with supercomputers. Why? Superior orchestration beats superior components. As Garry Kasparov, who pioneered this human-AI “centaur” model, observed: “Human creativity was paramount”. The winning formula wasn’t the strongest individual pieces. It was the best collaboration protocol.

The creativity mirror

Quick reality check. We often claim unique human creativity, contrasting our “originality” with AI’s supposed mimicry. But let’s be honest, most human breakthroughs are novel combinations of existing knowledge. Evolution itself works through random mutations that survive and thrive.

Perhaps AI’s “hallucinations” mirror our own creative process: informed pattern-matching that occasionally produces genuine insights. The line between human and artificial creativity may be blurrier than we think.

The main message? We don’t need to choose, we need duality between fuzzy and logical thinking. Between human wisdom and artificial intelligence. Between creative leaps and structured knowledge. Feeding an LLM siloed data is like asking an orchestra to play while half the violins read upside-down scores in Klingon.

Tomorrow’s enterprises will thrive by orchestrating this dual intelligence, letting each strength amplify the other in perfect cognitive harmony. Where do you see the biggest need for “System 2” logical checks on AI’s creative outputs in your work?

Stay tuned, the next blog post comes soon: From static blueprints to living systems.