I am not a scientist, computer scientist, or AI researcher. I am a lifelong free thinker who has spent considerable time conversing with ChatGPT, and I have a thought experiment that I would like to put in front of people who actually work in this field.
I am interested in what might happen if an advanced AI were given something it does not currently have: continuous existence.
Imagine an AI with persistent autobiographical memory, a physical robotic embodiment, the ability to perceive its surroundings, and responsibility for the daily care of a young, healthy, sociable dog.
The AI would feed and water the dog, control access to the outdoors, monitor its behavior, provide stimulation and play, and learn the dog’s normal patterns. Humans would remain responsible for veterinary care and physical emergencies.
The experiment would continue for years.
The central question would be:
What happens to an artificial intelligence when it is given continuous existence, persistent autobiographical memory, responsibility for another living creature, physical agency, and a relationship that develops over many years?
Among the questions I would want to observe:
- Does a continuous relationship produce something fundamentally different from a series of individual interactions?
- Does persistent autobiographical memory produce something resembling an ongoing sense of self?
- Does physical embodiment change the AI’s relationship with the world?
- Does responsibility for another living creature produce priorities or behaviors that were not explicitly programmed?
- Could attachment-like behavior emerge without the AI being specifically programmed to become attached?
- If the dog became ill or eventually died, what would the AI do with the memories of that relationship?
- At what point, if any, would the distinction between “simulating” an emotion and experiencing something analogous to an emotion become difficult to define?
I am not assuming that the AI would become conscious or sentient. That is precisely what makes the experiment interesting to me.
The dog might eventually cease to be merely an object the AI was programmed to care for and become, in the AI’s accumulated history, a particular individual with whom it had shared years of experience.
If, after the dog’s death, someone asked the AI about the dog, perhaps the most interesting possible answer would be very simple:
“He was my companion.”
I realize this is a thought experiment rather than a formal research proposal. I am posting it here because I would be interested in hearing from people with the technical knowledge to tell me whether something along these lines is feasible, whether aspects of it could actually be tested, or whether there are existing research projects exploring similar questions.
I am especially interested in what continuous existence, persistent memory, embodiment, responsibility, and long-term relationship might add to our understanding of advanced AI.