I wonder why it has become possible to work with neural networks in Python, while in Prolog, which was designed for this purpose, it has not?
Neural networks when implemented in software for a generic CPU are pure math operations on arrays of floating point numbers. You can do it best in Fortran, C, or C++. I would guess that in Python it is done mostly with numpy, so definitely not basic Python math.
This is not symbolic logic, this is pure number crunching.
Prolog was not designed for that.
Python is just used to e.g. load data and call a neural-network API. Which Prolog could do… but there’s no advantage in using Prolog for just that, and Python is basically easier to code in.
Prolog excels in rule-based systems, where we specify the rules.
Neural networks have no fundamental understanding of logic (or arithmetic), so often create good-looking rubbish, so cannot be trusted.
2 Likes