The point of this change is to make things easier on 90% of humans -- the ones who have JavaScript and third-party cookies enabled now get to tick a checkbox and be on their merry way, instead of doing a useless captcha when we knew they were already humans. Recall that when ReCaptcha initially came out, the argument was "humans are wasting all of this time, let's turn it into useful work to digitize books".
If book-based or street view-based captchas go away, I suspect it will be because bots/spammers got better at solving them than humans, not because Google thinks that the machine learning spam detection approach is fail-proof.
Recall that "reading" captchas already pose an insurmountable barrier to users with conditions such as illiteracy, low vision, no vision, and dyslexia. To accommodate these users, audio captchas are also provided, but a 2011 paper suggests that audio captchas are either easy to defeat programmatically or are difficult for users themselves to understand: https://cdn.elie.net/publications/decaptcha-breaking-75-perc...