Machine learning vs Statistics

I once jokingly said that I finally began to understand the maths of AI when I started to think of it as machine learning v.s statistics (as opposed to the traditional?view that data science and machine learning are based on statistics)

But joking apart, there is some truth to this

To clarify, of course, statistical ideas are used in machine?learning and deep learning

But indeed the two practises differ fundamentally when it comes to inference as I shared before (statistical inference is not the same as machine learning inference)

when statisticians?say that machine learning and deep learning models are fundamentally unknowable (black box) it sounds like something bad

But it need not be because?ML and DL models can be evaluated by withholding part of the data thereby providing a measure of usefulness of the model?

However, ML and DL techniques lead to altogether new types of applications (for example based on image recognition) and lives have been saved using these techniques that do not(by definition) require the underlying distribution?to be known.

to summarise

Statistical Inference: Based on known probability distributions and simpler models, suitable for smaller datasets, and provides clear interpretability and measures of uncertainty.

Machine Learning Inference: Data-driven approach that uses a variety of algorithms to learn from data, requires larger datasets, and includes both interpretable and complex models.

Deep Learning Inference: A specialized subset of machine learning that uses deep neural networks to model complex patterns, requires very large datasets and computational power, and is less interpretable.

The future belongs to highly paramateriised?models i.e. less interpretablle?models and hence a deviation from the statistical mindset towards the deep learning mindset (although statistical techniques will continue to be used within the context of machine learning)

Image source - a famous rivalry - spy v.s. spy - image from MAD magazine spy vs spy