At the very least, it needs to be an agent in a world (which is the real world or close enough), have many senses which it is able to connect to coherent experience. chatbots only talking in text don't really come close because the words are meaningless to them, they don't connect to any other sense or relate to any model of the world. And then it needs to have an internal model of the world and an internal understanding of how its actions effect the world.

Then, it needs to be able to learn continuously, not just be pretrained. And it needs to be able to learn from few training materials, like humans. It needs a sense of time, and a sense of self.

And that's not nearly enough, but we're not even there yet.