This can happen to a product that goes through a long and gradual evolution, where the changes may seem subtle, and subsequently are not properly vetted out.
If you look at the old Gmail's icons, there was much better distinction between Back and Reply buttons.
This is how the "Back" Button looked like in 2011 Gmail.
And this is how the "Reply" icon looked like in 2011 Gmail.
As you can see, the distinction between the Back and Reply button was clear.
Things were even more clear in 2009, when they actually spelled out "Back to Inbox".
Anytime there is any change, however benign it may seem, we should take a few steps back, and re-assess. Because a series of benign changes can add up to a major problem down the road. And that's exactly what happened in later 2011 when Gmail completely eliminated text-based labels for all their buttons and used only icons. The changes were net well received.
I've personally worked on several applications that gradually evolved over 10 years, and have witnessed the same type of problem. In a large organization, it is extremely difficult to have a process in place that ensures periodic re-assessment, because it is hard to justify the cost due to the fact that the process itself doesn't yield direct financial benefits. And the process can also be rife with politics.