Still, a 25 times lower likelihood of shooting a suspect if they are black in the simulation-based experiment..
That is entirely corroborated by what common sense would predict given the backlash a typical police officer would expect in the event that they shoot and kill a black suspect versus a white one.
Putting common sense aside, a 25X lower likelihood of shooting unarmed black suspects in the experiment suggests massive bias in favor of black suspects.
All it tells us in isolation is that this is how a set of people in current society acts in a simulation of that kind. It does not tell us why, and trying to draw conclusions about why without further experiments is irresponsible.
Even multiple minor/medium sized flaws in the experimental design that produce black favoritism not reflected in real world behavior wouldn't in the aggregate account for producing a 25 times exaggeration of bias.
The bias is undoubtedly there.
With that said, there's a massive caveat here: This is based on simulations
For it to be generally applicable you also need to then demonstrate that this matched real-world observed behaviour. On its own it is not at all obvious whether or not this accurately reflects anything at all.
There are many possible confounding issues here. E.g. for starters a typical excuse in shootings is that the officers claim to have feared for their safety. You then need to account for whether or not the simulation accurate trigger the same fear, otherwise someone with a racial bias in what makes them scared in "real life" could be expected to act differently in a simulation where they know they're safe.
I don't know whether that is the case or not, but there are many enough possible issues like this that a simulation only allow us to draw very limited conclusions here without a far more complicated experiment. Without understanding why.