Which concept describes models with billions of weights that are difficult for humans to interpret, creating a black-box system?

Which concept describes models with billions of weights that are difficult for humans to interpret, creating a black-box system?

• A. Inscrutability
• B. Lack of Understanding
• C. GOFAI
• D. Frame Problem

Answer: (A)

Opacity in large neural networks is the key idea. When a model has billions of weights, the path from input to decision becomes so complex that humans can’t easily trace or interpret how specific inputs lead to particular outputs. That kind of opacity is described as inscrutability—the system is a black box you can observe at the outside but can’t easily understand inside. GOFAI refers to old, rule-based AI approaches that are typically more interpretable, not the black-box type described here. The Frame Problem deals with what knowledge a system should keep track of as the world changes, not specifically with interpretability or internal complexity. Lack of Understanding isn’t a standard technical term for this phenomenon. So the best fit is inscrutability, capturing the challenge of interpreting decisions from highly complex, opaque models.

Opacity in large neural networks is the key idea. When a model has billions of weights, the path from input to decision becomes so complex that humans can’t easily trace or interpret how specific inputs lead to particular outputs. That kind of opacity is described as inscrutability—the system is a black box you can observe at the outside but can’t easily understand inside.

GOFAI refers to old, rule-based AI approaches that are typically more interpretable, not the black-box type described here. The Frame Problem deals with what knowledge a system should keep track of as the world changes, not specifically with interpretability or internal complexity. Lack of Understanding isn’t a standard technical term for this phenomenon.

So the best fit is inscrutability, capturing the challenge of interpreting decisions from highly complex, opaque models.