I find myself wondering if the type of inference required to make this claim could be automated. Given the right observations, couldn't dark matter be detected (or at least hypothesized) algorithmically? Couldn't this be used to create a "dark matter scope" by which the dark matter in the universe can be "seen" and visualized?
I'm guessing here, but they probably detected this automatically. There are millions of millions of galaxies and you can't put a even graduate student to look at each one.
My guess is that they were preparing a boring paper, like "Analysis of Normal/Dark Matter ratio in WhAtEvEr type galaxies near SoMeWhErE". They put a telescope, some processing and then they transfer the data to Excel to make a nice graphic. Then got some outliers, and with more analysis they were discarded as error. But they got one nasty outlier that were not easy to kill. They measure it again, and again, and probably made another team double check it.
(Perhaps they were looking for faint galaxies, and it was not too much luck.)
Anyway, probably most of the calculation was automated, and probably now some other teams will try to find similar objects.
This was definitely a high-priority, by-hand, single object reduction. They put 33 hours of Keck time into it--you don't do that unless you're certain you have something very interesting.
(For reference, the capitalized value of one night (~8 hours) of 10-m telescope time is roughly $100k.)
Not everything can be automated, but there is certainly at least some machine learning in all of the larger surveys these days. As to dark matter surveys explicitly, I found this one: http://kids.strw.leidenuniv.nl/pr_july2015.php
