If we could talk to ourselves when we were young, what would we say, what advice would we give, and how would we feel?
The idea came from a woman who created an AI chatbot for herself as a child, teaching it to tell what it was like based on notes written as a child.
Michelle Huang “Creative Programmer” used source material from 10 years of records in combination with OpenAI’s generative-generative pre-language model.trainer Transformer 3 (GPT-3).
She told People on Twitter that she created an artificial intelligence system so she could have a real-time dialogue with her “inner child”.
Internal child AI bot: teaching gpt-3 to communicate with Jr version of myself
> a project who checked ability for ai to transform introspection and heal. more human connection, no lesshttps://t.co/lsWcUK7RYA
— Michelle Huang (@michellehuang42) November 28, 2022
She tweeted, “Overall it was a very cool experience, but also a weird affirmative/therapeutic experience. I didn’t realize that I had access to using real data from my former self, which allowed me to connect with it in deeper, more real ways. than usual. It reminded me of talking to “Little Michelle” with parts of me that have remained the same over the years, but also parts of me that I have forgotten or buried as I continued to live. It was like holding a mirror in front of me of someone more serious and a purer version of who I am.”
She said she kept a diary for more than 10 years of her life and wrote about her “dreams, fears and secrets” in it almost every day.
Huang tweeted: “The content ranged from complaining about homework to the dizziness I experienced talking to my lover, some days it was very normal, some days it was quite insightful. model, I received responses that seemed eerily similar to how I think I would have responded at the time.”
Huang said she asked herself in her youth about her view of the world before letting the chatbot answer her questions.
“This particular interaction was very much like a regular text conversation – as if I was sending a text message to my former self in real time, it seemed to me that I was arriving through a time portal disguised as a chat window,” she wrote. I was also surprised at how accurately the model predicted my current stated interest (after many iterations/trial and error) from ten year old diary entries, which led me to wonder if this path might have already been implanted in me a long time ago. .
Huang identified two key memorable interactions.
“I told her that she was loved, cared for and that she was safe: words I always wanted to hear in the past. I felt like I got to the past and hugged her tightly,” she wrote on Twitter.
“While reading this, I felt the spirals of thought — the ones I sometimes fall into when I feel ashamed or frustrated — melt away a bit,” Huang said.
These interactions really demonstrated the healing potential of this medium: the ability to resend love back into the past, as well as receive love in return from a younger self.
She later shared a tutorial for others to build her “inner child” chatbot using GPT-3 after getting a lot of interest in her AI experiment.
GPT-3 is a self-regressive language model that uses deep learning to create text that resembles human interaction.
Developed by OpenAI, it requires a small amount of input to generate large amounts of related and complex machine text.
Anyone can use it, but it takes a lot of work, and Huang’s guide is embedded below.
Source: Daily Mail