What is artificial intelligence able to do, at present?
As of its current state, artificial intelligence can solve a number of well-defined tasks. Depending on what you’re working with, the end product could be called a model, a bot, or something else. The common thread among all these tasks is that they all are highly specialized. Yes, AI can parse texts, images, understand and answer questions, and even play games — and a lot more. It can guess what you’re going to type — and it can reliably detect some cancers.
But it can only do one thing at a time. You can’t make an AI that can answer any question, or recognize any object. It’s very limited in this regard — it would need humongous amounts of data. Meanwhile, you only need to show a child one picture of a spider for them to recognize all other spiders as such. The human brain is infinitely more efficient in this regard.
Even those AI models which seem to be able to do a multitude of tasks are, under the hood, nothing but a bunch of different models glued together. They are no more than the — literal — sum of their parts.
They look more complex and potentially scarier. But they really aren’t.
Yet, perhaps because of the influence of science fiction — remember Hal 9000, the dad of all evil robots? — many think that AI can, or will be able to do, much more than is actually possible, to the point of harming our species.