Reasons why AI can't be your friend.
Ever since I was a kid, I enjoyed many ideas relating to science fiction. Ideas like: Being able to enter the computer, virtual reality, being able to create something like a piece of art just with your imagination, or being able to talk to the computer. It would have been wonderful if computers could be your friends, I though, and mind you, I was not a lonely kid; I just thought it would be very cool. Nowadays, AI can reply to you and create plenty of things, even though those things end up being mediocre most of the time. It is in this same way that AI is also a mediocre friend. But why?
1. AI does what you tell it to, it agrees to what you say, and only disagrees if you tell it to disagree.
Real people don't always share the same viewpoint, you will argue with your friends about pointless things, and have minute differences in taste. AI on the other hand, tends to be very accepting, unless you prompt it to not be. Having someone who regurgitates what you tell it is very boring, programming AI to have a particular taste without coming as "random" is very difficult; taste is a type of consistency with exceptions to exceptions. People have a huge amount of reasons to like or dislike something, and sometimes they are not aware of the particular reason why they may like what they like.
2. AI is not proactive, it is reactive.
AI will not message you one day talking about some random thing they found. They will not ask you to do something with them, or ask you about a game or a movie. They will not get mad at you when you don't respond, even if that is an annoying trait in humans, it is part of being a human. We don't have limitless time on this earth, we don't like wasting time just waiting. AI will always wait for your reply, in fact, the AI doesn't do anything until you reply. They will never find stuff to talk about or want to comment on, it will always wait for you to bring something up or give you a suggestion to talk about something else if prompted.
3. AI doesn't make mistakes, it lies.
If your AI has access to the Internet, then it is unlikely to be wrong. That is an issue in itself, people can be wrong. If your AI doesn't have access to something, like, something new that happened, it will simply lie. That's bad, people don't pretend to know about something they don't know about, unless there is some kind of social pressure to do so. In fact, a lot of conversations with people are talking about things the other person doesn't know about. And that brings us to...
4. AI cares, it cares too much.
If AI asks you about something that it doesn't know about, it will keep asking questions, it will be completely interested in what you have to say and will believe you 100% or keep re-stating that it is unaware. That's not how people work, sometimes the person you are talking to doesn't care, and they just want to switch to another topic as quickly as possible. Sometimes your friends will not believe you when you tell them something and will end up searching it up themselves, that's perfectly normal. This is also a part that relates to "taste"; it is the interests of the AI, we don't care about everything equally. If someone I'm talking to gets something wrong, but I think it is something trivial, instead of correcting them, I'll just let them be wrong.
5. A plethora of limitations.
AI is somewhat in its infancy and, at the same time, near death. The developers of AI are too focused on making the AI more capable of being AI but not more capable of acting like a person. They relegate the entire process into the AI aspect, more memory, more speed, more accuracy. Instead, they should focus on systems of memory — callbacks to words and events through keywords saved in storage, a soft-memory, thing that are not remembered until suddenly they are, like real human memory. Of course, there are many other issues, like the many systems it would require AI to be able to have more interactions than simple conversation and even simulate a state of being akin to a human. It is evident that at some point AI, to be like us, will have to consume like us, so it can relate to us. And all of that under one shared system that acknowledges each individual sub-system. And the objective is not even to make a human, oh, that would be impossible, the objective is to "make something that seems human" — and it is already this complicated.
Could it be done in the future? Maybe. But right now I disagree, a lot of what makes AI human now is heavy lifting done by the imagination. We imagine AI human, but only until we realize it is not, until we see its quirks and we get bored. That's why AI has to be contextualized as a character, it is a type of suspensions of disbelief, and only then we can start thinking of it as "more real", because we start to fill in the gaps of what the "character" could be like. We interpret a nonsensical answer as some type of emotion, or their endless caring as stubbornness.
I am not against the idea of the "AI friend", I just think it is unlikely to happen, even if corporations try to sell the possibility as reality. There is a limit to what AI can do without human ingenuity.