There are different AA qualities for methods to be judged on. Not least of which is 'can your algorithm be applied to your engine?' AFAIK MSAA does not work well with deferred rendering that is a preferred technique, so it might not even be an option for some games. Furthermore, you need perhaps at least 4xMSAA to approach the jaggy reduction of TAA, which can be excellent, and even then you only get geometry AA from it and not surface AA, whereas TAA samples the whole scene.
In short, all techniques have pros and cons. The best quality is supersampling, but the cost is prohibitive. I expect TAA is used because it's realistically the only option in a fair number of cases. What was the last AAA game to use MSAA? I feel it's unheard of since the PS3 era!
) on HDTVs in the PS3 era. Aliasing was of least concern on low-quality CRTs, but then they were blurry. CRT monitors with proper RGB input were crisp, and the jaggies clearly visible with plenty of scrawl. That's why we had AA options. The IHVs added MSAA, advertising up to 16x in their big cards, and Quincunx, and even supersampling, to try and eliminate the aliasing. Articles were written showing different dotty sampling patterns and their results. The average was 2xMSAA, maybe 4xMSAA if you were lucky. MLAA was a revelation as it provided the edge clarity of 16x supersampling without the ridiculous cost. I remember commenting at the time I wondered why no-one drew polygon edges with something like Wu's Algorithm.
