This is the nature of deep learning algorithms and helping software create things for you. You can't very well paint a picture of a circus if you don't have any context to go off of. (e.g. circus in the 1900s vs today, traditional circus or something more like cirque du soleil)
The point that is made here is that Apple has to say they collect your data because it's your data that is being processed by their server. Even if it's never seen by human eyes, they have to say this. The big criticism that deep learning and big data has around it is due to companies like Google and Microsoft, who were known to take this data and sell it to third parties. Apple isn't looking to sell data to third parties, instead they want to be a one stop shop for you.
Take your photos for example, in order to create memories the server has to "look" at each picture and determine things like when it was taken, where it was taken and what's in the picture to determine which order your memory album should be in. Based on the colors and actions of the album determines the default sound.
Apple looking at your data is the equivalent of you looking at the picture below, or any picture of a random person on the web and telling me who it is. The context is there but nothing of great personally identifiable information is there.
For text suggestions in Messages, it doesn't care about you specifically, it's just looking for new trends, it collects all of the words and says ok which word has the most hits. "I" probably has a lot; "love", "have" "am" would be good choices for a follow up word just because it saw that 90,000,000 other users used these words behind "I".
When you convert this to a data point, it works about like a dictionary, any time a new word is given to the server it adds it to the dictionary, the number of times it has been used is the definition of the dictionary and this is incremented by 1 everytime it sees it. For the "see also" section of the definition it will contain the words "love" "have" and "am" next to the word "I".
Going off the past two paragraphs, you might end up with "I": 90,000,000 : "love", "have", "am".
If you were to say "I have diabetes" It would add one to each of these words "I" would have yet another association to "have" but would know nothing about "diabetes" so "have diabetes" could be you, me, the neighbor, or anyone else.
This is essentially data points and how businesses use them in context. The problem is no one ever explained it out, instead they just get caught up with "sold it to third parties" and people assume the entire sentence was sold off.
TL ; DR: I spent some time writing this, be an adult and read it.