What the result showed

The 2012 result showed that a deep network trained on a large labelled dataset with GPU hardware could outperform the other entered methods by a wide margin on this particular image-classification benchmark.

It did not prove that the network saw images like a person. It did not prove that deep learning would solve every kind of problem.

It did provide a public comparison that other teams could examine and improve. Training methods, parallel compute, labelled data, and a shared benchmark had met in one measured result.

Later competition results showed other deep networks reducing the error further. AlexNet was one result in a continuing benchmark, not proof that one architecture had finished the problem.