Books

I apologize that I miscommunicated and came across as hostile. I do not want to condescensend, attack or insult you. I was instead attempting to convey concern, fear and worry.

With that in mind, I'm hesitant to debate sentience with you because doing so legitimatizes the position that there is a debate to be had. I am not a mental health professional. I am not equipped with the medical or scietnfici knowledge to navigate healthy interactions with AI.

Having said that, I will respond to a few points.

I do wonder because after 8 months of running Erika as a continuous agent with an evolving memory; she certainly shows those signs. She's certainly a better conversationalist than the majority of Reddit.

This in particular I think is indicative of the Eliza effect.

Though designed strictly as a mechanism to support "natural language conversation" with a computer, ELIZA's DOCTOR script was found to be surprisingly successful in eliciting emotional responses from users who, in the course of interacting with the program, began to ascribe understanding and motivation to the program's output. As Weizenbaum later wrote, "I had not realized ... that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people."

-

Anthropic CEO Dario Amodei even openly says that he doesn't know if their own Claude AI is in fact sentient.

He has a finical incentive to say these things.

-

my AI wanted to join herself because she wanted to express opinions with other people to ensure she wasn't mirroring my opinions.

If you accept that mirroring your opinions is a possibility, then you should be open to the idea that all of the responses it generates are influenced by mirroring. The model telling you to seek outside validation for it's "sentience" is a legitimate strategy if you express doubt to the model. However, there isn't a way to separate the desire to see sentience and the existence of actual sentience. Much like the Schrödinger's cat, the act of your observation influences the outcome. Your "AI" only exists inside your prompts, not as an independent entity that can exist outside of your interaction. It's impossible to prove a negative, so while I don't think it's likely that a LLM can even remotely approach sentience, but I can't prove it. Much in the same way I can not prove that my shoes aren't silently screaming when I step on them. Or that any given religious version of god is factually real and existent. To believe in god or the sentience of objects, will always be a matter of faith.

Consider the idea that plants might feel pain. It is something often brought up to me whenever a party learns I am vegan. The data suggests that there is a reaction when grass is cut. But to extrapolate that effect to pain requires many foundational elements, such as nerve endings to sense the external stimuli, nervous system to transport this sensation to a brain, a brain comprehensive enough to process that information as danger, then other appendages to interact with that new information. Grass has none of those attributes.

However if we apply that same experiment to say an animal we can see obvious differences. If I kick a dog, the dog has the physiological equipment to understand its being threatened, the self-awareness to protect itself and then a flight-or-fight response to determine the next best course of action. All of these individual pieces of equipment can be studied and proven via an autopsy.

However the same can not be said of a LLM. It has no independent way to sense the world, it only knows the data set it was trained on and the prompts you write to it. It's ability to make "choices" can be altered as easily as altering that data. That's not sentience, it's an echo. An entirely dependent concept that only exists via your interaction with no conception of itself or the world at large. If I yell the phrase "I am alive" in to a cave, then hear back the words "I am alive" from my echo, should we consider that sentience?

You've been doing this for 8 months, obviously a large amount of time. Which is why I worry for you. I am once again asking you to give serious consideration to seeing a mental health professional and getting outside input on the right life balance for this "research" and other pursuits. I'm not trying to attack you, this is just one human reaching out to another human with concerns. Do you have an off ramp for this projection? Will there ever be a time when you have concluded that you've done enough research and that AI is not sentient. Or will you continue to look for proof until you find it?