Ai is just adaptively automating automation. It’s the meta-automation of tasks. At its core, it’s just dealing with patterns and their recognition which is another word for learning.
Since any form of input, text or image, is stored as data, a program trains until it has a general idea of what it is looking for. Any new input is checked against a general pattern. The original input can also be pairs of different types of data. As the training goes on, the program can isolate what part/s of each type of input data correspond. It can be used to check if a new pair of data is valid or, given one pair of data, it can find the other pair. It’s how DALL-E works (text and image pairs). And DALL-E’s algorithm can also be used for more types of pair inputs.
It is a shame that the Ai field’s aim is to imitate the human brain. It’s just some copy-pasta. McCulloch-Pitts neurons are the result of a neurophysiologist and logician working together to develop a mathematical model based on the workings of the brain. Later literature had to weed out and sand off a lot of biological references and elements to get workable, software-friendly models. If the aim is to develop context-aware, resilient, autonomous, and wishes-complying software, there can be more ways to do that. But, if the aim is to have human-like workers which can work 24/7, then we are heading towards a profit-first and interest-centered dreamland.