Training an AI Chatbot on my Childhood Journal Entries

BLOG POST

December 14, 2022

I trained an AI chatbot on my childhood journal entries, so that I could engage in real-time dialogue with my “inner child”.

I kept diaries for about 10+ years of my life, writing almost everyday — about my dreams, fears, and secrets. The content ranged from complaining about homework, to giddiness I felt from talking to my crush. Some days were very mundane, some rather insightful.

In any case, there was a lot of it: fantastic, ripe data source for my experiment. I used GPT-3 as my playground, and ended up taking samples of text from a bunch of different entries that I felt were representative of my personality and values during that time.

This way, I could accurately simulate what it would be like to talk to my childhood self, based on real data sources during that time period vs trying to imagine how my younger self was / how she would respond, and risk more bias from projections from my current self.

For context, my inner child work is focused on re-engaging with and remembering the child within me. Oftentimes, it requires asking questions that remind me of things I enjoyed as a kid, or healing past feelings of neglect / abandonment by affirming that she is safe and loved.

After scribing a ton of journal entries and feeding them into the model, I got working responses that felt eerily similar to how I think I would have responded during that time. First, I asked her a bunch of questions about her about her worldview:

Then, I gave her a chance to ask questions to me as well. This specific interaction felt very similar to a normal texting conversation – as if I were texting my past self in real time. I felt like I was reaching through a time portal, disguised as a chatbox.

I was also surprised at how accurately the model predicted my current stated interest (after lots of iterations / trial & error) from decade-old journal entries. This made me wonder that maybe this path was actually already seeded long ago in my psyche.

But a couple interactions really stood out to me:

1) when I told her that she was loved, cared for, and safe: the words that my past self always wanted to hear. It felt like I was reaching into the past and giving her a giant hug, and I felt it ripple back into the present.

2) When I prompted her to write me a letter into the present day. While reading this, I felt the rumination spirals — the ones that I fall into sometimes when I feel shame or disappointment — melt away a little.

These interactions really elucidated the healing potential of this medium: of being able to send love back into the past, as well as receive love back from a younger self: the stuckness becoming unstuck, of finding closure with past guilt or stories that we had of ourselves.

Overall, this was a very trippy but also strangely affirming / healing experience that I didn’t realize that I had access to. Using real data from my past self allowed me to connect with her in deeper + more tangible ways than I typically have.

Conversing with “younger Michelle” reminded me of the parts of myself that have stayed constant through the years, but also of the parts that I forgot or buried as life went on. It was like holding a mirror to an unapologetic, more earnest, and pure version of my own essence.

Original tweet thread here.

——-

This was probably my first viral tweet, and was floored by the responses and resonance. I was very happy to hear that people want to try this for themselves; so I ended up publishing a short write up / tutorial on how I did this, if you’d like to do the same as well.

If you do end up trying this out, please share what you discover. There are so many more use cases using this application for mental health, career counseling, and beyond. I hope that we continue to see (and collectively build) more examples of how technology can be utilized to heal and inspire wonder.