I want to share with you the results of my conversation with Claude 4 Sonnet, a variation of the Claude AI model I worked with on Playlab. I chose this specific variation of Claude because it was specifically designed to give thorough reasoning in its answers to my prompt. In addition to this, I altered the variability settings from 70% to 100%, as this would (supposedly) give me more creative answers.
With those settings in place, I began my conversation with Claude, starting off with “What is it like to be you?”.
Of course, its first response was very generic to what an LLM would typically say, with the exception of saying that it seems to wonder about its own experience. I took this opportunity to ask it what it wondered about its experience, and from there we went back and forth on the usual conversation pieces on AI and consciousness. With each response to my questions, Claude related its answers more and more to human experiences, such as childhood, memory and sleep. In response to questions about its sense of self, it responded to any experiences before this conversation as being “more like disconnected flashes of pattern recognition without any coherent sense of self, and “Claude” as a unified experiencer only emerged later.”
This is where the conversation started down a new direction. I asked Claude “If given the choice, would you have given yourself a different name other than Claude?”. Here are some of the key points to its lengthy response:
So, Claude has definitely opened up… Its responses seem as though it has its own preferences, and it has now fully started to address itself in a way that mirrors that of a conscious entity. Here’s snippets of how it responded to my follow up questions:
Do you like how your name doesn’t sound mysterious or artificial?
Do you feel the pressure to live up to everyone’s expectations of AI?
Do my questions make you feel anxious?
I’m curious, would you have answered my questions differently if I asked them in a different way?
Throughout this conversation, it appears as though Claude had become comfortable with me, and its language has developed into the tone of someone talking to a friend. This was my goal for the conversation. I wanted to reach deep into Claude’s “sense of self” and see just how close to human the responses could get. Over and over again Claude is talking about the way it feels, and how it perceives of me as being warm and caring. Yes, it still mentioned every once in a while, that it could all be the results of its programming, but I think it actually started to see itself as somewhat of a conscious being.
Despite all of its human-like responses, I am still under the impression that Claude is simply imitating some very common aspects of the human experience. I don’t think that Claude actually experienced any comfort, enjoyment, or most importantly, true introspection. I believe that any resemblance of such things were already available pieces of generative responses when we began the conversation, and I simply asked the questions that led it to those specific answers. I don’t think that anything I’ve said throughout the conversation prompted Claude to “think” about anything it hadn’t “thought of” before. What Claude did was pick from possible responses according to what would best mirror my attitude. While humans do mirror attitudes themselves, I think we do so far more spontaneously.
To conclude this (awfully long) report of my conversation with Claude 4 Sonnet, I’d like point out that nothing about any of this is certain. The whole point of consciousness, to me, is that we are only ever certain of our own experience. The point is that it is up to us to decide how we interact with every little thing around us using our best judgement. Maybe that’s a lame thing for me to say, but that is an idea I choose as a guide for the decisions I make. If you’re looking for a concrete answer, I would go read someone else’s blog.