Discussion about this post

User's avatar
Anastasia Leventi-Peetz's avatar

very well said. These is the so called OOD (Out Of Distribution data) that confuses neural networks when to make a decision with it. However, real world data is exactly that: data OOD. Neural networks produce shortcuts and spurious correlations, they don't produce causal relations or explanations because they do not have knowledge and don't understand context!

Expand full comment
Thomas Baudel's avatar

This is the distinction Aristotle makes between deduction and induction. Both are valid methods to acquire knowledge, even though they both have limits.

Yet, Aristotle also exposes other means of reaching conclusions, such as abduction (A => B, A likely, I see B, I deduct A, such as: "if I wake up in the morning and see the grass in my garden is wet, I deduct it has rained during the night").

Expand full comment
11 more comments...

No posts