We are very far from self awareness. Even models like GPT3 are just transformers very good at predicting the next words in a chain of text, while keeping track of the context. So it is just a very good prediction algorithm, not a sentient being. My guess is problems could get real when they will combine different AI. This is how our brains work. We have a language center, an image recognition center... So, maybe in a decade or so?
Note that the concept of combining multiples AI already exist. The GAN's (Generative adversarial network) for example. Those are the models most likely used to generate all those images of hot chicks I have seen in other posts on this forum. The GAN's are in fact 2 AI working again each other: one that creates images, and one that tries to detect bad images. But there is still a long way between that and a metal skeleton wielding a machine gun... At least I hope so.
|