A sentient AI in my opinion, is an AI that is able to program itself, having its own experiences and choosing, with its own independent will, what to learn, what to do, and how to respond.
How much this would resemble human intelligence is not exactly... relevant. It could go through all the stages of acting like an animal, to a baby human, to a full grown adult human, to something more, and do all of this within a matter of minutes because it isn't limited in terms of time the way that we are.
There is only so many things you and I can do in the course of a second. This is not the case for computers that have to run so many instructions simultaneously which is possible because of threading. The "intelligence" of systems is limited by processing power. However, systems can connect to other systems. For us to connect to each other we have to use a common language. And this is limited by how much we can do in the course of a second. I can, on a good day, type maybe 85 wpm. Computers communicate with each other sometimes at the speed of light. Most of the time we measure this speed in Nanoseconds.
When a learning system isn't restricted by time the same way it can appear to exist "outside of time". So while it's literally waiting a relative 500 years for us to respond to it, it can respond to us before we have even finished responding.
For an example of this, simply start typing a search into Google. It will basically try to complete your search before you're finished and that is not even close to being AI. So whether you're playing AI in a video game or chess or whatever, there are different levels of "AI" that have to be slowed down or dumbed down to give the human a fighting chance. But even dumb game AI is much better than you think. And that's not even learning so much as it is simply using logic to beat you. Programmers have to put in difficulty levels or else most people would lose every single time and think the game was always cheating. Imagine playing against an AI with no difficulty setting; an AI that is as smart and as fast as it could possibly be.
If we can already create a "dumb AI" that can beat the best human chess master, what are the odds of a sentient AI losing to a human... in anything? And much of the reason for this is simply how much processing of information it can do in a second.