The software I used is available through Vision API - Image Content Analysis (cloud.google.com/vision). I used another Google service, image search, to locate a studio portrait of GM Kasparov, then fed it to the Vision API. The 'Faces' section of the analysis returned the image shown below on the left, while the first item in the 'Web Entities' section told me 'Garry Kasparov 10.947', where the number is probably some kind of a confidence metric. The second item was 'Chess 1.06755' and the last was 'Bobby Fischer 0.09179'.
The 'Pages with Matched Images' section returned a long list of links, most of them from Pinterest.com, including a link to the same image I used for the test. No question about it, Garry Kasparov has both name and face recognition.
While I was conducting that test, it occurred to me that the Google AI software might be cheating. It could easily locate copies of the image in the Google archive of results from web searches, then analyze the associated text.
I searched my own archive of images and located one that might not have found its way to a web page somewhere. I fed this one to the Vision API and it returned the image shown above on the right, along with 'Garry Kasparov 10.2885'. There's no doubt about it -- the API recognized Garry, and this time 'Pages with Matched Images' was empty.
One more curiosity is worth mentioning: the analysis of both photos returned the following list of face attributes:-
Joy : Very Unlikely
Sorrow : Very Unlikely
Anger : Very Unlikely
Surprise : Very Unlikely
It seems that AI software can recognize Kasparov's face, but it can't recognize his expression.