I just think that pure classification and regression problems cannot describe artificial intelligence well at all.
Artificial neural networks are trained in an end-to-end manner, and the trained model is a black box. Why don’t people use the same idea to study general artificial intelligence? That is, using an end-to-end approach to study general artificial intelligence instead of splitting a large problem into small problems (classification, regression, and others).
What you described end-to-end training is mainly due to feature learning with modern deep learning architectures.
In machine learning, feature learning or representation learning is a set of techniques that allows a system to automatically discover the representations needed for feature detection or classification from raw data... The hierarchical architecture of the biological neural system inspires deep learning architectures for feature learning by stacking multiple layers of learning nodes... The input at the bottom layer is raw data, and the output of the final, highest layer is the final low-dimensional feature or representation.
So in the sense of end-to-end deep feature learning, you're right that the focus on classification and regression as classical discriminative ML models cannot cover AI well at all compared to modern deep generative models, as discussed in yesterday's post. And of course these (deep) feature learning 'black boxes' will play a more and more important role in AGI such as multimodal learning with transformer which can be seen as a step further towards AGI.
