From what I’ve read in different papers and talks on the subject the most concise definition for what a real AI would be is as follows.
The system must be self aware, understand it’s place in an environment, and be able to Reason, Adapt, and work towards internal goals.
If you ever want a good example of a failure of adaptability try playing chess with Chat GPT. Chat GPT is INCREDIBLE at what it does and is like nothing else we have ever seen.sometimes it seems downright spooky how good it can be at responding to normal requests and conversations. However It knows what chess is because it’s been trained on text that has the rules and the names of all the pieces. But because it is just a LLM which is effectively a word probability generator it can’t REASON what the game actually is or remember what pieces have been played or removed.
It’s like asking the deep blue chess computer to hold a conversation with you. It’s not happening.
Right now (to the best of my knowledge) there are no actual AI systems. Everything that is being called AI is just complex programs that are following EXACTLY what the programming has set in stone with no deviations.
The YouTube algorithm interestingly might be the one exception. If the rumors are to be. Believed it has grown so complex that YouTube itself doesn’t actually know how it works.