Taking appearances seriously in machine learning
I’ve been attempting to kind of distil what are the basic claims made in Taking Appearance Seriously: The Dynamic Way of Seeing in Goethe and European Thought. The authors claim that modern science is locked in the the Cartesian/Lockean tradition, and goes something like this:
- Everything has a hidden “form” (in the neo-platonic sense) and a separate appearance.
- All we can observe is the appearance.
- We can use various methods of statistical/mathematical thinking to discover the form from the appearance (this is where ML fits in).
A example given in the book is hand-waving (literally!). The Cartesian understanding would have brain/mind breaking the procedure of observing hand-waving in two separate parts: one observing arms flaying around and one transforming this movement into a “hello” representation in the brain. Any gesture recognition software has this knowledge baked in. Now, what is being argued by the author is that this processing pathway does not exist – we see “hello” directly. I am not sure what to make of this or do with it.
The second claim made by the book, which seems to be to far more useful is the equivalence of process vs the equivalence of the final product. All you need to claim that two things are the same is that they are created in the same way – equivalence of a learning/developmental algorithm is important – not the final outcome. Two cocktails are the same if they are made the same way. Will they be somewhat different? Yes, but there is no need to explain this variance away, it’s fine.