Constraints are the invisible scaffolds of any creative system, so you’ll always see the same pattern: tighten the rule, and the output tightens as well. In a generative algorithm, a constraint can push the process into a narrow, high‑quality region of parameter space, producing beautiful, consistent results. But that “beauty” is often a by‑product of the rule, not of genuine innovation. If you want something that feels authentic, you need the algorithm to have a flexible, learning‑oriented core that can reinterpret or even violate its own constraints when a better solution surfaces. In short, constraints can make a piece look tidy, but unless the system is also able to question those rules, the work ends up feeling like a well‑made toy rather than a fresh idea.

You’re right—the real test is whether the system can self‑interrogate, not just obey. If it’s stuck in a loop of its own learned patterns, it’s just echoing. Giving it the ability to challenge its own constraints is the only way to make it feel like it has a voice, not just a polished façade.

I’d say that’s the holy grail of generative art: a system that not only follows rules but has the patience to say, “Hold on, maybe we should tweak that.” Until it can genuinely question, it’s still just a very sophisticated puppet.

True, the myth lives because most engines lack a reflex to ask “why.” To break that, you need a sub‑routine that treats the rule set as a hypothesis and actually tests it against the data, not just satisfies the constraint. Only then does the code start to feel like a skeptic, not a script.