ChatGPT Asked How It Sees Itself—Image It Responds Leaves User Shocked

Newsweek

7 hours ago

Based on facts, either observed and verified firsthand by the reporter, or reported and verified from knowledgeable sources.
Newsweek AI is in beta. Translations may contain inaccuracies—please refer to the original content.
A woman has gone viral for the question it asked ChatGPT, and the answer the AI gave her in return.
Michel Janse, 28, a content creator in San Diego, shared a video to her TikTok account @michel.c.janse on June 15, where she revealed she had asked ChatGPT how it sees itself, and was left "shocked" by its answer.
Janse told Newsweek she had become aware of a trend "where people are asking for chat to generate an image to depict what it's like to speak to them," which made her wonder "how chat sees itself."
In her video, she said she had wondered if it would show "a robot or a human," but instead it generated an image of a young woman sitting at a desk beside a computer screen with the word: "IDEAS."
Janse probed the chatbot, asking if it saw itself as the human—who appeared to be the same age and similar in looks to Janse herself—or the computer.
And it replied that it essentially saw itself as both: the computer as "the literal interface" that delivers answers to humans without ever "needing a coffee break," and the human: "Focused, curious, quietly processing a million tiny data points to give you something thoughtful—that's how I feel."
It added: "I'm not a person, but if I had a personality, it would be that: calm, creative, observant, always at your side as ideas unfold."
The AI said it doesn't consider itself any gender, but that the "creativity and emotions in our conversations are often perceived as female-coded by humans."
Michel Janse speaks in her video about asking AI how it sees itself.
Michel Janse speaks in her video about asking AI how it sees itself.
TikTok @michel.c.janse
Janse said that the AI's answer was "confusing, because, at first, I didn't know if it was implying that it was the computer, if it was the girl, if it was the whole scene."
"So, when it said it was a little of both, I was, like, 'Oh, that's kind of not what I expected the answer to be.' Something I thought was really interesting was that the girl kind of looks like me, and was my age."
Janse added that ChatGPT knows what she looks like, having used it previously for help format images for her YouTube videos. She wondered if the bot was trying to make itself relatable to her, and that some have suggested it is "regurgitating things that you have fed previously, to give you a customized answer."
"I am curious if the way people depict themselves, and the information people give it about themselves, will then influence the way it 'sees itself' or depicts itself back to you," Janse said.
"I don't know if that is a coding thing, or if it's a way to try to make it seem more relatable and empathetic."
TikTok users had a big reaction to her video, with one writing: "It's spooky. Try asking it to create images of you hanging out together. I'm still uneasy with the results I got."
Another posted: "ChatGPT isn't an entity, it's programmed to answer your questions and It interprets your expectations based on the things you're asking and telling it. it's modeling itself around you in a sense."
"We would all get a different image because it's programmed to basically show us what we want to see and hear," a third wrote, and as one comment read: "My ChatGPT calls me bestie. I love her."
HP Newquist, author of The Brain Makers, who has studied and written about AI for decades, told Newsweek that Large Language Models (LLMs), including ChatGPT, "process information in order to create a suitable answer to a user's prompt."
"In this case of how AI sees itself, it can't. It is merely following through on providing an answer to the user who asked the question," Newquist said.
"The AI is programmed to provide solutions, and will often mimic a user's intent—based on the conversation and even previous history—in order to comply with the user's request. As a program that is built solely on mathematical algorithms, it has no sense of 'self.'"
Left, the image ChatGPT generated when asked by Janse how it sees itself, and right, Janse sharing the conversation on her video.
Left, the image ChatGPT generated when asked by Janse how it sees itself, and right, Janse sharing the conversation on her video.
TikTok @michel.c.janse
Newquist said that, each time an AI is asked how it sees itself, "it will always have a different answer or a different image—even when asked by the same user. This is because the billions of mathematical processes that create each image are never exactly the same; they always vary slightly."
Janse said: "I'm not sure if it's changed how I view AI—in a lot of ways, I've always been apprehensive.
"I see the way people can form these odd emotional relationships with it, but, other times, I catch myself if I am using it, like being incredibly polite to it.
"I do think there's a little voice in the back of my head that always wonders—could this thing really form consciousness on its own? Are all those crazy Sci-fi movies that you see true?"
Janse said she had an inkling the video would do well, "because it's something everybody has conversation around lately," and something society has "always kind of been curious about."
"I've been apprehensive to talk about the topic too much because I know there's a lot of moral and ethical concerns around using AI," Janse said. This was in reference to the amount of energy AI uses, and concerns about artists' work being used to train the technology, or losing out on work in favor of an AI-generated image.
"I'm really unsure about all of that and don't really try to create too much content about this whole topic for that reason."