I attended Jeff Clune’s seminar at the Schwartz Reisman Institute on September 17, 2025. The title—“Open-ended and AI-generating algorithms in the era of foundation models”—was precise. What followed was unsettling in the way good ideas often are.
He described agents that don’t fix their goals. Instead, they wander. They switch targets. They explore dead ends, not to fail, but to discover. These agents solve problems that rigid optimisers never touch. I saw simulations of agents that only completed their objective after passing through three distinct local minima: first failing, then adapting, then inventing. The solution wasn’t a straight line. It was a detour that looked like collapse.
This is the “NEVER” insight: without space to wander, certain solutions are NEVER reachable. LLMs, he argued, now serve as our proxy for “human notions of interestingness.” They don’t chase efficiency. They chase novelty—what the OMNI framework calls the generative edge. They are not tools. They are co-evolutionary partners in problem spaces too vast for any single mind to map.
The uncomfortable question came next: What if maximising human quality of life requires passing through local minima that cause harm? Think of a cybersecurity breach that forces systemic hardening. Or an economic shock that restructures obsolete institutions. We call these “necessary suffering.” But we don’t design for them. We wait. Clune suggested simulation as the answer—a sandbox where we can let agents fail harmlessly. I am not convinced.
Simulations lack visceral weight. They cannot trick our biology into feeling the real cost. For a simulation to move society, it must be so immersive it becomes a functional reality—an echo of the world that feels authentic, even if it is not. This connects directly to my collaboration with Barry Wylant at the University of Calgary on design for resilient systems. Our work asks: how do we create innovations that build resilience before crisis? Not as reaction, but as rehearsal.
There is a parallel here in ecology. Hormesis teaches that organisms grow stronger under controlled stress. A forest needs fire. A immune system needs pathogens. Society needs innovation that stings. Chris Kennedy’s “The Malthus Enigma”—which I heard him launch weeks earlier—warned that the worst thing for humanity could be magically solving infinite energy and food. Without cultural resilience, we may not survive abundance.
This is the pivot. Innovators must stop asking “How do we maximise outcomes?” and start asking “How do we build adaptive capacity?” Not just better products. Not just faster exits. Better resilience. That is where the next tier of innovation lives: not in perfecting the path, but in expanding the terrain.
Related: