Recently, I’ve been reflecting on my journey into the AI safety community. Perhaps like many, my introduction didn’t include much fanfare, but it did include a healthy dose of magic.
In December 2022, a friend introduced me to ChatGPT. This was my first tangible interface with AI and I wasn’t sure what to think. Initially, I was skeptical about whether the outputs were actually unique. My cynicism faded and my curiosity grew when I learned that they were unique and that the computer was predicting the output one small chunk at a time. As I played with it over and over, it was difficult not to feel like this magical technology would change the world.
Over the next few days, I shared this new technology with anyone who would listen. Every demo felt like I was handing them a small piece of the future. And in it all, the feeling of magic kept coming back. I think a part of it was due to my not knowing how this machine worked, and I wanted to learn more.
To help bridge this epistemic gap, I bought Superintelligence, which I slowly read cover-to-cover. In the process of reading the book, what had first appeared to be a fun party trick had quickly morphed into something far more ominous. While it was still magical, it was a powerful technology that, according to the book, would only get more powerful. Following the trend lines, I could see how the world I inhabited was about to change forever.
It’s tough to overstate how transformational these early days of AI were for me. I was experiencing something I couldn’t fully explain, something that felt like it belonged in a movie. Except it wasn’t a movie, it was real life, happening to me, and I was a character in the story whether I liked it or not.
And since those early days, I have made an effort to form an appreciation for being alive during this moment. I knew that if this technology was going to change the world, I would want to help make it go well. And perhaps at the time, I could only dimly sense the gravity and responsibility that would come with having that desire. It’s a responsibility that many people working on AI safety embody beautifully.
It’s been almost four years since that introduction, and the feeling of magic hasn’t completely worn off. It’s only gotten more complicated.