Statistically black people are more likely to be arrested and convicted for crimes. So if you use an algorithm to determine who should be arrested and convicted, how long they should be sentenced, and likelihood of recidivism, and seed that algorithm with historical information about arrests and convictions tagged by race (even inadvertently through names or addresses), you end up permanently encoding untrue information like "black people are more likely to commit crimes."
an example from a particular class of crimes:
Study: Whites More Likely to Abuse Drugs Than Blacks
https://healthland.time.com/2011/11/07/study-whites-more-lik...
Black and white Americans use drugs at similar rates. One group is punished more for it.
You can get into a debate about the merits of controlling for factors, but the title is somewhat misleading. They are suggesting that whites have higher rates of drug use when you control for factors like socioeconomic status, not that the hard numbers show that whites use more drugs than blacks.
The second article is literally two paragraphs with no actual data other than an infographic. I think you need some better sources if you want to make the claims you are trying to make.
The problem is that algorithms are an easy way to hide behind. Think of an algorithm to check wether to do a police check for a person. Statistically black people are more likely to commit crimes. The police could easily hide behind an algorithm and say that they are not doing checks only on black people but only on persons flagged by the algorithm.
The problem with that is that most people are not criminal, but will be discriminated just because of which group they belong to.