And so, what does it do?? Where does input go? Where is the rewards?
What can't it do?
Does it just hold info?
Now that you're asking some smarter questions I'll make more of an effort to help you understand it. That diagram is just a map of a thing called a knowledge base. It shows the hierarchy of classes in the SUMO knowledge base. A class is a type of thing. You can have another class that is a more specialised type of the first thing. e.g. the class of animals and the class of cats which is a subclass of animals. All the subclasses of animals have in common the attributes of animals, but add some special features like whiskers or claws or flippers.
Just like maps of the world, you can have different kinds of maps that show different aspects of the same thing. For example what that map doesn't show is the containment hierarchy of the knowledge base. The containment hierarchy describes how things make up other things. For example, a normal cat would have four legs. Legs aren't a kind of cat so they're not a subclass of cat (but cat legs could be a subclass of animal legs) they are parts of a cat.
To see the entire knowledge base you would have to look at the source code. It's hundreds of thousands of lines of code written in a dialect of a language called KIF (Knowledge Interchange Format). The source code includes everything, but it can be easy to get lost in the details, that's why it's helpful to generate maps so we can find our way around more easily. The diagrams don't contain all the information that's needed for the software, they are just an aid to understanding.
Some kinds of maps have a start (inputs) and a finish (rewards) but not this one. This is like a map of the roads of a city which tells you where you can and can't go. This map tells you what can and can't be true. The knowledge base describes knowledge in such a way that you can use it to evaluate new knowledge.
If someone tells the AI something, the AI can translate what it has been told into KIF, and then see if it is in the knowledge base already. If it is in the knowledge base already then it "knows" it is true. If it isn't in the knowledge base, then the AI can check to see if it contradicts anything that is in the knowledge base already. If there is a contradiction then the AI "knows" that the new fact is false. Suppose that the new fact isn't in the knowledge base and it doesn't contradict anything that is already in the knowledge base? Then the new fact could be added to the knowledge base and that's how the AI learns.
There is no guess work or approximation or even magic involved. KIF and SUMO together allow you to perform calculations with knowledge in exactly the same way you would perform arithmetic with numbers.