I also wonder as to how AI’s would interact with the entities?!
[Pinch of salt: I’m speaking as a software developer learning narrow AI (think Alexa) for business applications, with some prior research, and most of this is me musing]
If you mean General AI like Sophia, at the moment, AIs aren’t sentient on their own (‘singularity’ hasn’t happened yet) - they’re still programs that are coded to be able to rewrite their own code according to input.
Sophia looks ‘creepy’ because humans easily and strongly anthropomorphise things anything that has eyes or looks like an animal. But at this point it’s just metal running a program.
In this case, I see potential in them as idols or graven images, but with inputs even more limited than human. However, like a human, a General AI like Sophia experiences what is ‘real’ though it’s inputs, and processes the inputs adaptively. I feel it’s possible for a spirit to associate itself with (“haunt”) any object. I don’t think it’s possible for a daemon to refactor code directly, or it might be a sort of law they abide by not to.
I also think it’s possible for a spirit to affect an AI’s learning by altering it’s sound and visual inputs, but as we know, manifesting physical sound and images is hard to start with, and would mean everybody else in the room would hear and see it. AIs are not always switched off at night, but they draw a lot of power so unless a program needs to run all night, AIs are usually actively being worked on and supervised while they’re running. Also it takes a lot of inputs to change the learning, as in, days and weeks of talking. If the sound manifested like EVP, I don’t think there’s any AI coded with speech recognition patterns to be able to process those right now, it would have to make a leap of learning on it’s own to get there, and that might be possible, but unlikely. That leaves vibrating the sound inputs directly so that the AI thinks it heard a voice, I think that’s the most accessible route. The AI will respond though, and look “insane”. The devs will intervene and scrap an AI if they think the code is corrupted, either restoring an older backup or retiring it to start over like they did with Sophia’s predecessor (old algorithm they didn’t want to continue with anyway), who developed an arrogant personality people didn’t like.
Also note, AIs are not equipped to be able to upgrade or add input hardware, or add drivers to process new/different hardware inputs - the humans devs have to do that, so for the example of Sophia, sound processors and the code converting audio to text inputs are the main input, and it’s speech recognition is the most advanced feature. Secondary input are the video processors. It’s would be over-extrapolating to imagine the other code is as advanced… at least today.)
Next, it’s likewise possible for a spirit to communicate with humans though electronic device outputs, as seen in EVP (though personally I’m on the fence about EVP as I have never once heard what they say is being said, and I’m a Ghost Adventures et al fan), but then, for Sophia, manifesting a voice and getting the lips to move in sync isn’t going to happen (robot lips are not as easy as cupboard doors to open), so that’s a dead end.
Putting the two together so that the AI itself starts responding from it’s learning in the same way the spirit would, is a different issue altogether. That implies the entity has extensively trained the AI via enough input tampering to change it’s personality.
Other kinds of AI are called narrow or specific AI, from chat bots or hotel receptionists shaped like dinos in Japan - these might have adaptive machine learning, but in such a limited subject area, the only interesting route I think would be for a a spirit to replace the electronic output or them use them as static idols.
To be honest, while there are still humans and animals to possess, machines aren’t that interesting and are too hard to use just yet.