However, I find that the AML score gives different and varied results from one service to another. Are these results accurate, especially since most of them are generated by artificial intelligence? Can it be relied upon to bring charges or acquit people, since they may lead to criminal charges, or is this not possible as they are based on random results or inaccurate.
Classifying something as a part of money of ML is too vague. So, the number can easily be inflated.
For example, one known suspicipious activity worth $1,000 could transmit innocently to over 100 people after the initial problem transaction, after that, that could be counted as $100,000 of transfers related to ML.
One service might classify it as $1,000, others might classify it anywhere between $1k and $100k depending on their "algorithm".
The truth is that AML is a very hard task and it will take a lot of privacy limiting measures to make it more accurate. If we ever get to a stage where every coin can be identified and traced, privacy will be completely eliminated.
TLDR: Results will always be inaccurate, at least until privacy is dead. AML accuracy can't exist while privacy still exists to any extent.