I used to have an artificial intelligence project that I worked on, (which I never completed), and one insight that I had was that there are certain things that

I used to have an artificial intelligence project that I worked on, (which I never completed), and one insight that I had was that there are certain things that a neural net can learn, but there are certain things it can’t. You have to program in time invariance as an assumption, that the same object can have the same meaning when you see it a second time. And, you have to make position invariance and rotation invariance as an explicit assumption too in your image recognition model. My guess is that the human brain has some dedicated hardware somewhere that rotates every object it sees and matches it in different ways, and my guess is that in dyslexic people that hardware is a bit different. I hesitate to say wrong, because that’s a value judgement, but different.