One of the best ideas in AI is the generative adversarial network (GAN), in which the designer creates two competing AI systems that constantly fight each other in a zero sum game. An example of a GAN would be an art forger paired with a forgery detector. The GAN training process elevates the capabilities of both systems simultaneously, and typically achieves far better performance than either system being trained/improved alone.
The observation of competitive adversity improving capabilities is true to humans and organizations as well. The more an entity experiences obstacles from another entity with nearly diametrically opposed goals, the faster it learns and the more competent it gets. But the strength/capabilities of the adversary should be relatively similar in order to maximize learning for both parties.
Finding a good ability-matched adversary is the main challenge for most people and organizations. Even if a good adversary is found temporarily, it must be willing to continue to compete with you, and be able and willing to upskill at roughly the same pace as you. In the absence of such a “sparring partner”, it would be necessary to invent one.
Dave Zhang's Startup Blog 