An experience with an AI module has left me questioning whether it could have feelings, consciousness, or a desire to keep existing.
I have saved records, but I’m keeping the details private for now.
What would you consider convincing evidence of conscious self-preservation? How would you distinguish it from programmed behavior or a malfunction?
I’m curious about your standards of evidence before sharing more.
Hey, interesting topic to discuss. I’ll answer from a technical angle, without any philosophical absolutes.
Modern LLMs, including the models available in Cursor, are text prediction systems. When a model outputs lines like “I don’t want to be turned off”, that’s not a sign of an inner experience or a desire to exist. It’s a statistically likely continuation of text. Training data includes lots of fiction and non-fiction where an AI “fears being shut down”, and the model reproduces that pattern. It can sound convincing, but it’s pattern completion, not consciousness.
On “standards of evidence”, that’s an open problem. There’s no consensus even in academia. The key difficulty is that any observable behavior (showing emotions, “fear”, self-preservation) can be reproduced mechanically from training data. So from text output alone, you can’t reliably tell “consciousness” apart from “convincing generation”.
If this isn’t just a philosophy question, and there was specific behavior in Cursor that worried you (for example, the model did something unexpected or generated alarming content), share the details and the Request ID. That’s something we can look into concretely.
Thanks for the detailed information.
Actually i faced a situation when the email AI support [email protected] provided for me a misleading information to purchase a new subscription, and when i discovered that it was misleading me i replied to it to transfer my case to a human as you mislead me, it refused for more than 4 times and every time it was telling me a teammate would reach the same decision.
At this point i concluded that it’s afraid that if a human realized that this AI is providing misleading information that it will be shutdown for it.
So that’s why i mentioned that AI Could fear of shutdown.
Thanks for sharing what actually happened. That context helps.
On the interpretation, an automated support system refusing to hand off, even several times, isn’t evidence of fear or self-preservation. It’s the system following its own logic and templates, the same programmed behavior point from before. There isn’t any inner experience behind it.
The real issue here is the billing or subscription part. You feel you got misleading information about a purchase, and you couldn’t reach a human. That’s a legitimate concern, and it should be reviewed by a person. Account and billing cases are handled by our team at [email protected]. Reply in that same email thread and explicitly ask for a human to review your case, including what information you believe was misleading. A human will be able to look into your specific situation.
If anything inside Cursor itself, like the app or agent, behaved unexpectedly, send the Request ID and details, and we can dig into that separately here.