Yes we must look at what we have available now to make sensible predictions about the future.

Replica is utter garbage. Doesn't remember two messages before, and is generally very unimpressive. Character.AI is better in every single way, it is way more convincing, remembers a lot from your convos, many times it simulates a human's curiosity and unpredictability correctly.

However, it is still very much clear that you're speaking with an AI, sometimes they'll say stuff that makes no sense AND it's easily tricked by the classic phrases like "What colour was Napoleon's white horse?" answer: "I don't know much about Napoleon so I can't answer". On top of that, the developers implemented very strict censorship so even if the model is the absolute best at simulating NSFW, its mouth is literally forced shut from the ouside.

But our friend is asking for a cure for loneliness, not a quick wank. For that, a couple of prompting tricks will have me have fun with Hu Tao's toes in seconds. But for true companionship we require much more.

For that, the AI needs to remember everything we tell it and NEVER forget it, or at least be smart enough to know what stuff is important and must be remembered. This may be a decade away for all we know. Also, the human mind works in funny ways, many times the mere scarcity and difficulty of finding love gives value to the love we find. It is entirely possible that no matter how advanced AI gets, many people may never fall for it, because they know if they ever do or say the wrong thing they can just hit the reset button and start again, that alone may make them look at AI partners as worthless and just another sex toy.

That said, looking at how fast the sector is moving, a convincing AI partner might be ready in 10 years.

​

Edit: and we mustn't forget that speech recognition and quickness to answer must be improved upon a LOT. Right now the best speech recognition still needs at least one or two seconds to catch up. We need it to be near instant. And the bot must be able to respond quickly and to have a credible human like voice complete with emotions, though great strides were achieved on these things lately.

As for hardware, the timeline seems way more stretched. The closest thing we have is indeed VR, but for realistic VR we need the tech to be good in both the software, the window by which you see the world (screen), and the hardware must be powerful enough.

The software is there, just look at how good Unreal engine 5 is. The hardware isn't there yet but will probably be in 5 gens or so. But the screen is way more behind, for truly good VR you're looking at 8k resolution per eye *at least* AND either OLED or MicroLED for perfect blacks, both technologies which have still a long way to go. Workable screens might need much more than 10 years to arrive.

And lastly, haptic tech can only do so much. The best we have today is auto strokers and force feedback gloves but that can only carry the illusion for so long. Who knows if we'll ever be able to invent something that actually feels like a real thing. It may take decades.

Sexbots suffer from our robotics tech being limited. The only robot which moves with any semblance of fluidity is Atlas but it's clearly intended for military purposes and its expensive af, and loud af. To get a robot that moves like a human and doesn't sound like a textile factory it will take decades.

As for Neuralink, exposing yourself to literal infections to the brain in order to coom will never not be a dumb idea for me, no matter the fact that it may potentially take a needle for every neuron to produce a satisfactory result and that it's decades away anyway.