Anyone who’s written code to accomplish what I and many others call “doing AI” are familiar with the desire to sidestep the daunting task of establishing constraints and heuristics. Let an algorithm crawl through a search space with no plan whatsoever, except to do what it thinks best at each step. This is to conduct a “blind” or “greedy” search. This is usually done as an exercise to discover what obstacles must be overcome, rather than with an expectation of success. But seeing where this algorithm fails is very instructive.
The canonical failure scenario for the troglodyte is when it encounters a local maximum. The blind climber searching for the peak gets stuck when he reaches a false peak. The strategy of just climbing the steepest wall available at any given time breaks down. And if the climber is blind and stubborn, this will mark the end of his ascent. Local maxima can be devilish traps when all one knows how to do is go up.