Open-endedness · abandoning the objective
Two searches evolve agents to reach the gold exit of the same deceptive maze. The left one rewards getting closer to the exit, the obvious plan. The right one ignores the exit entirely and just rewards doing something it has not done before. Watch which one actually gets there. To make progress in this maze you have to move away from the exit first, and a plan that always heads toward the goal cannot see that.
Reward = get closer to the exit. The sensible plan, and the one that gets trapped.
Reward = reach somewhere new. No goal in mind, yet it collects the stepping stones that lead there.
What you are seeing. The dim dots are every place a search has managed to end up. Objective search rewards only closeness to the exit, so its population piles against the wall nearest the exit and stops, because every legal next move points away from the goal. Novelty search never looks at the exit. It rewards ending somewhere new, so it keeps filling the maze, and the exit is simply one more new place it eventually finds. This is the core open-endedness result: collecting novelty gathers stepping stones that a direct plan skips right over.
But even this is only half of open-endedness. Novelty search still explores a fixed space, the set of positions in one unchanging maze, exactly as Conway's Game of Life rearranges a fixed grid under fixed rules. The deepest open-endedness does something these cannot: it expands its own space. Gordon Pask's electrochemical devices grew a sensitivity to sound they were never built to have, adding a whole new dimension of things they could respond to. That is the difference between finding new points in a given space and inventing a new axis for the space itself. A truly open-ended system does the second.