"Artificial intelligence" is a buzzword that does not exist in reality. Everything that today is touted as "AI" is actually something called machine learning (ML). They are statistical models that are trained on a lot of data (text, photos, etc.) to recognize patterns (image recognition, automated medical scans, etc.) and, depending on the model, output similar but novel data (text generation, voice synthesis, etc.).
For example, image recognition works by feeding the ML model tons of photos of named objects. Photos of cats labeled as "cat", photos of boats labeled as "boat", and so on. With enough of this training data, the algorithms are capable of being fed a photo of a cat they had not been trained on and are able to recognize it as a photo of a cat. There is no real intelligence or thinking going on.
The same holds true for even our most advanced text generation models, such as GPT-3, that are able to generate content that is virtually indistinguishable from human-written texts. It does so by having been fed terabytes of text taken from novels, websites, Wikipedia, etc. When you then prompt the model by inputting a sentence of text, it will look at all the data it has been fed to predict what words follow. It's basically your phone's autocomplete, only much more advanced. There is no thinking involved. The model does not hold any values, does not have a personality, and is not self-aware.
To answer your question, ML will definitely replace human interaction for a whole lot of people. It would essentially be social media echo chambers, the models being fine-tuned by their very nature to be the perfect reflection of its user. Also, as @GentleSoul said, for a lot of people you don't even need anything as advanced as an ML model to replace human connection.