Guy Reams (00:00.686)
This is day 69, my personal touring test. Yesterday I decided to connect my own experiment. I wanted to see how far the platform we are building could replicate me. Not just my words, but my tone, my creative thinking, the way I construct an argument or tell a story. I wanted to know if AI tools could replicate me well enough to fool the people who know me the best. So I did my own touring test.
I had the AI engine write something in my style. Then I took it further. I created a video of me speaking using my voice and my image. I sent it to my family and friends to see if they could tell the difference. The results were interesting. It largely failed. They could tell. But even though it failed, it was remarkably close. Close enough to make me little uncomfortable. Close enough to make me think harder about what this actually means.
The original Turing test was designed by Alan Turing in 1950 as a way to measure machine intelligence. The idea was simple, put a human evaluator in a room and have them communicate with two other entities through text, one human and one machine. If the evaluator cannot reliably tell which is which, then the machine has passed the test. It was never meant to be a perfect measure of intelligence, but rather a practical benchmark. Can a machine fool a human into thinking it is also human?
That was the question. But my experiment was different. I was not trying to see if a machine could fool a stranger into thinking it was human. I was trying to see if a machine could fool people into thinking it was me. That is a much harder test. A stranger does not know my quirks, my patterns, the way I think or the way I express myself. They do not know the stories I tell or the metaphors that I use. They do not know the rhythm of my sentences or the way I build an argument.
But my family does, my friends do. The people who read my writing regularly do. They know me, they know my voice, and that is what I wanted to test. The classic Turing test asks if a human evaluator can tell whether the responses come from a machine or a human. Applied to my experiment, the question becomes more nuanced. If a reader who knows me well could immediately tell the blog was not me, then the AI fails.
Guy Reams (02:19.788)
If a neutral reader suspects it is AI, even they do not know me. Then the AI fails. If a neutral reader and someone who knows my writing both cannot confidently say it is artificial, then we are approaching a creative touring threshold. But here's the issue. The touring test was never designed for personalized creativity. It only measures surface level imitation. It does not measure style, depth, emotional authenticity, consistency with your worldview.
alignment with lived experience or temporal coherence in your writing history, the classic Turing test evaluates plausibility, not personal authenticity. So I need a better test. I need a test that measures not just whether the AI can fool a stranger, but whether it can fool the people that know me. I need a test that measures not just plausibility, but authenticity. I need a test that measures not just whether the AI can replicate my words, but whether it can replicate me.
That is much harder problem. That is the problem that I guess I'm trying to solve for. I propose a new test. I call it the personal touring test. It's an adapted version of the original, but designed specifically for personalized creativity. It asks three groups of people to read the piece and evaluate it. The first group is people who know my writing very well. They read my blog on a regular basis. They know my style, my tone, my patterns. They know the way I think and the way I express myself.
The question for them is simple. Can they tell it's not me? Where does it ring false exactly? What gives it away? The second group is people who know me personally, but not necessarily my writing. They know my personality, my worldview, my values. They know the way I talk and the way I think. The question for them is different. Does it sound like something I would say? Does it match my personality and worldview? Does it feel like me? The third group of people is those who neither
Know me nor my writing. They are neutral evaluators. They're like the control group. They do not have any context. The question for them is whether anything feels generically AI. Does it feel repetitive, over-smoothed, flat, lacking specificity? Does it feel like it was written by a machine? When all three groups cannot reliably detect AI, the piece has passed this personal touring test for authored work. That is the standard that I guess I'd be aiming for.
Guy Reams (04:47.512)
That is the benchmark I'm trying to reach and right now I am not there, not even close. I am somewhere around the 60 to 80 % pass rate I suppose. The AI got my tone, it got my phrasing, it used my thematic patterns, but it still lacked the texture of a human experience. It lacked the specificity, the lived detail, the personal anecdotes that make writing mine.
It lacked the mistakes, the tangents, the moments of vulnerability that make writing feel human. It was close, but it just wasn't me. The question is why? Why did it fail? What is missing? What is the difference between plausible and authentic? I think the answer lies in the details. The AI can replicate patterns, but it cannot replicate experience. It can mimic my style, but it cannot mimic my life.
It can generate sentences to sound like me, but it cannot generate sentences that come from me. It lacks the context, the history, the lived experience, shapes, the way I think and the way I write. It lacks the mistakes, the false starts, the moments of doubt that make writing feel real. It basically lacks the humanity. But here's the thing, it is getting better, surprisingly better. The gap is closing.
The AI is learning. It is getting closer to replicating not just my words, but my voice. And that raises a question. If the AI can replicate me well enough to fool most people, what does that mean? What does that mean for authenticity? What does that mean for creativity? What does that mean for me? I do not have any answers to that question yet, but I'm going to keep testing this out. I'm going to keep refining the personal touring test.
I'm gonna keep pushing the boundaries of what is possible and I'll document that along the way as I keep trying. And I'm going to keep asking the question, can AI replicate me? And if it can, what does that say about what it means to be me?
Guy Reams (06:51.48)
That sucks.