Question: Analyze why machine learning currently relies on informal definitions for task difficulty and explain how neural network architecture (shallow vs. deep) is used as an informal proxy to distinguish between easy and hard tasks.

Sample answer: Machine learning currently lacks a robust formal definition of task difficulty. Therefore, practitioners rely on informal definitions tied to computational complexity and deep learning architecture. A task is considered "easy" if it can be solved with fewer computational steps, which corresponds to a shallow neural network. In contrast, a task is deemed "hard" if it requires significantly more computational steps, necessitating a deeper neural network with multiple layers.

Key points:

• Machine learning lacks a good formal definition of task difficulty.
• Informally, task difficulty is linked to the number of computation steps required.
• Easy tasks correspond to fewer steps and shallow neural networks.
• Hard tasks correspond to more steps and deeper neural networks.

Rubric: The response should acknowledge the absence of a formal definition, define an easy task in terms of fewer computation steps and shallow networks, and define a hard task in terms of more computation steps and deeper networks.