If you could solve the vision problem through deduction, then a good decision tree would be enough to implement artificial vision. The fact that it isn't suggests that the problem is much harder than it seems. https://en.wikipedia.org/wiki/Moravec%27s_paradox
