If these captcha images tests can’t be figured out by a robot, how do they know if you answer correctly? Does some human pre-answer all?

No, some human does NOT pre-answer all. Verizon DOES NOT PAY employees to monitor or moderate YA OR TO ANSWER questions on YA. Yahoo LAID OFF everyone. Verizon continues to lay off people in many locations.

The new security measures are NOT going to stop most of the human slammers who are PAID to spam YA. Some of them have been on YA for over a year.

YA is run by server computers and computer programs.

Here's my guess:

Their program simply has a list of images that match what they want you to find (i.e. fire hydrants). The only reason their computer recognizes the images is because it is told what each is. In order for a bot to complete the captcha images, it would have to be able to take low-quality images as input and churn out an output matching the given term. Unless you have either that or access to their own bot, bots can't figure out the images.

The capture images are a very simple human test machine cannot think for itself even artificial logic will not work and if they ever figure out a way to do it they'll simply reset the capture images

That's one approach. But they could do it without pre-answering at all.

Essentially they let a few humans answer these and just accept whatever they say. Then for later humans, they see if they match closely to the first ones. Over time they come up with a map of which squares contain the images and which don't.

I also have a feeling someone is using these results to feed a neural network to see if computers can do the same thing. So not only are they checking if you are human, they are using your "training" to teach computers and that ability can help self-driving cars detect buses, street lights, fire hydrants, etc.

It's similar to what Google was doing earlier with two-word captcha tests. One would be a known word that was digitally masked to make it hard for computers to read, but should be doable by a human. The second word would be actually a word from a book that they had scanned where their computer OCR (optical character recognition) had failed. They'd use our responses to do the recognition of words that the computers couldn't fully understand. But they wouldn't reject the captcha for the second unknown word, just the first known word.

You've probably seen a similar thing with house addresses which I think they used from Google Street View cars trying to read house addresses. Every so often they would include addresses their computers couldn't decipher to make us humans do the work for them.

Pretty sneaky, huh?