The Importance of Self-Supervised Training for Deep Learning Models

I was thinking a brainstorm about the tasks that we can use to better define what do we mean by artificial general intelligence. To evaluate our models on those tasks for now, before we have the power to break up all the parts in the brain. So, maybe regular, right, field, P20 to a brain storming about what tasks we should be using. What do you think? How would you define less artificial, like more human-like along with axes? I think predicting neuronal response is something that we can try. But these, we probably need to have a test for deep learning models, something like Turing test for deeplearning models. One aspect of that test might be the brain