Sorting doesn't require intelligence. You have X number of attributes and using them you do your best to get as many things in to "Like" buckets as possible.
Discovering new attributes is the start of intelligence. Discovering new buckets could be intelligence.
If the machine gets 100k pictures and can sort them in to Cat vs No Cat, and then suggests a new set of classification "Indoors vs outdoors" using attributes the developer didn't pre-identify that would be intelligence.
It was given all the possible traits. It didn't find a new category it had a collection of things with known traits. Google knew it would be able to find cats using the traits it was given. That's not intelligence on the computer's side.
If I tell you that there are things with lungs, things with gills, things with legs and things with arms, things with hair, I don't have to tell you there are mermaids and salamanders, you will figure it out when not everything fits in the bucket of Cats and goldfish.
For example ,a machine learning algorithm devised a new internet protocol that doubles the network capacity over best known methods.Another famous machine learning system understood own it's own how to classify images what belongs to the bucket of "cat" and many other object types.
Why isn't this enough to be considered intelligent?