Yes, I'm thinking about exactly the same way to prove understanding of the rule, it's in the end of the post: >>> Generative NN architectures could be used as

Yes, I'm thinking about exactly the same way to prove understanding of the rule, it's in the end of the post: >>> Generative NN architectures could be used as well, like variational autoencoder (VAE), or generative adversarial networks (GAN). In this case, it would be “explanation by example”. [...] NN would generate new examples of images from Bongard problems, and if this generated images capture the concepts behind classification rule, it might be enough to show understanding of the problem.