Great potential for new science!
When lots of images have had streaks filled in by machine learning AIs searching those images for unusual objects will start finding new types of stars and galaxies only in the repaired parts images. ML corrections are fine for making pretty pictures but less good for discovering new objects.
On a (rare) clear night I can see the Moon, one or two planets, perhaps a dozen stars and about five aircraft. I have yet to see the ISS or a fresh batch of Starlinks. The Perseids have excellent correlation with cloud cover wherever I am. I saw the Milky way for the first time when on holiday in Tenerife. That was utterly beautiful and I thoroughly recommend looking up if you ever get over 50km from a village full of street lights.
I would like to thank Przemek Mróz for coming up with some proper numbers for the cost of Starlink and I would like to see similar work for other types of telescopes to weigh against the benefits.