"Europe, with GDPR, will be far behind"
All I can say is "lucky them" (the citizens of the EU, that is). The usefulness of any given AI is closely related to the quality of its training data, the degree to which hidden bias in that training data is removed/corrected, and the quality of the actual data being supplied to the trained system. If your AI/ML system has an extremely narrow focus with high quality training data and high quality input data it can be useful and probably fairly reliable.
In systems that are being fed data collected en masse (both voluntarily supplied and involuntarily mined) the spectrum of data quality is going to be so wide that training an AI that never produces a 'junk' result from real data is nigh on impossible.
You can't use AI to make decisions about people's lives on the basis of this kind of data. You can never be certain that the system you regard as well trained will not produce a stupid answer. Even if you catch the stupid answer you won't know how to prevent the same thing happening again. Statistical likelihood is not enough dealing with tens, and hundreds, of millions of people, each with thousands of data points, every day.
I think we've reached peak folly here. Coat, because much like gambling and financial instrument trading, AI/ML is something I'll never work on.