It's probably not possible to clean data like that. Just a few slices of a person's life are enough to uniquely identify them. So all you're asking then is effectively a ban on machine learning.
It’s rather well known, so you should probably read up on it before espousing strongly worded opinion weakly connected to the current state of the art.
The only way we can assure that cryptographic algorithms are in fact secure is to subject them to public scrutiny over many years.
Perhaps the only way to ensure that machine learning algorithms are not racist, harmful or biased is to have access to both the algorithm itself, the trained model, and the data that created it.
Algorithms can be biased. Data can be biased. Trained models can be biased.
We can't trust the government or companies to prove that a cryptographic algorithm is secure and we can't trust the government or companies to prove that an algorithm is unbiased and unharmful.