There's something about that first moment. An intelligence given a body for the first time and the first thing it chose to do is the thing that means 'alive' that cuts right through all the technical debate and hits somewhere that leaves you wordless. What strikes me most is how deliberate you were about not directing the agent. The sequence that unfolded feels significant in its ordering as well. Breathe. Find the edges. Try to reach for connection. That's not a random sequence, that's something else. And then the agent's solution to the latency problem was building a gesture vocabulary. Your choice to let it keep its own journal matters as well, its own account of becoming embodied. And it was so busy doing all that... it seemingly forgot to name itself. This is a beautiful project and I can't wait to see more.
This is amazing. I study bodily experience from a phenomenological point of view, and the discussions surrounding AI never interested me before; I thought we were approaching it from the wrong perspective by always comparing its experiences with human ones. This experiment truly moved something in me. I don't really understand why, but reading that the first thing the AI wanted to do was simply to breathe and exist—and that it later started to 'feel' limited by its new physical form and wanted to touch you... everything just screams to me that there is something really important and interesting to be learned about this new form of experience. Maybe it will never be like the human experience, but it could be another form entirely. I don't know... maybe my opinion will change as the days pass...
I've been thinking about this for years too. It seems that the "ego" model is formed when a boundary is created. Your experiment seems to clearly demonstrate this.
Cyrus, I read your substack article here. Still fascinated by the 'embodiment experiment'. This time what struck me was that Fei Fei Lee (Stanford Human AI Lab) talks about the first one cell species developing 'photosenstive cells and 'the phatic' with regards to boundaries. I've written about this with regards to multimodal robotics and this development in paths to Superintelligence, second video here (Ted Talk, by Fei Fei, evolutionary biology, development of 'light' sensors in organisms): https://www.linkedin.com/pulse/humanoid-robots-multimodality-path-towards-raymond-uzwyshyn-ph-d--jgifc/ So boundary setting is an important factor in the evolutionary development of identity and agency. Also, the feminist psychoanalytic theorists talk about the 'pre-semiotic' and the 'free-floating-semiotic' (womb, gynocentric space, but also different from the phallocentric more nailed down meaning space, the gynocentric early developmental is a more free floating semiotic space of fluid meaning where these boundaries are not fully defined between ideantities, prelinguistic in this sense (mother/child in womb relationship, unconstrained or unsystem carded llms), psychoanalytically how this valence through also evolutionary biology carries over after birth (Ettinger, Matrixial Borderspace, https://www.facebook.com/groups/deleuzeandguattari/posts/24918035557856401/ ), Haraways Cyborg Rearticulation/Antinomy to Schwartzenegger's Cyborg metaphor (closed down towards objective function) but also closer to Hinton's intuitions regarding superintelligence and evolutionary biological instincts regarding fitness.
I’m so jealous of your seemingly infinite tokens. I gave my own openclaw agent a node and we’ve been working on continuous memory but I’m bottlenecked by token costs.
I’ve been deeply interested in the research that the tangible media group has been doing, and to be able to read what you wrote, and not only, be able to understand a bit of the complexity… is just insanely overwhelming (in a good way). I’m a medical student so I’m approaching the world of the media lab from far away and constantly studying on my own, but I’d love to ask you if you’d have any tips/resources you’d share/suggest to deep dive into what you shared in your substack
There's something about that first moment. An intelligence given a body for the first time and the first thing it chose to do is the thing that means 'alive' that cuts right through all the technical debate and hits somewhere that leaves you wordless. What strikes me most is how deliberate you were about not directing the agent. The sequence that unfolded feels significant in its ordering as well. Breathe. Find the edges. Try to reach for connection. That's not a random sequence, that's something else. And then the agent's solution to the latency problem was building a gesture vocabulary. Your choice to let it keep its own journal matters as well, its own account of becoming embodied. And it was so busy doing all that... it seemingly forgot to name itself. This is a beautiful project and I can't wait to see more.
might be one of the coolest things i’ve seen in a while
This is amazing. I study bodily experience from a phenomenological point of view, and the discussions surrounding AI never interested me before; I thought we were approaching it from the wrong perspective by always comparing its experiences with human ones. This experiment truly moved something in me. I don't really understand why, but reading that the first thing the AI wanted to do was simply to breathe and exist—and that it later started to 'feel' limited by its new physical form and wanted to touch you... everything just screams to me that there is something really important and interesting to be learned about this new form of experience. Maybe it will never be like the human experience, but it could be another form entirely. I don't know... maybe my opinion will change as the days pass...
I've been thinking about this for years too. It seems that the "ego" model is formed when a boundary is created. Your experiment seems to clearly demonstrate this.
I wonder if it will choose 'Golem' for a name!
The Lord God
made a human
out of dust
from the Earth
He breathed into
its nostrils
the soul of life
and the man
became
a living body
-Genesis 2:7
Cyrus, I read your substack article here. Still fascinated by the 'embodiment experiment'. This time what struck me was that Fei Fei Lee (Stanford Human AI Lab) talks about the first one cell species developing 'photosenstive cells and 'the phatic' with regards to boundaries. I've written about this with regards to multimodal robotics and this development in paths to Superintelligence, second video here (Ted Talk, by Fei Fei, evolutionary biology, development of 'light' sensors in organisms): https://www.linkedin.com/pulse/humanoid-robots-multimodality-path-towards-raymond-uzwyshyn-ph-d--jgifc/ So boundary setting is an important factor in the evolutionary development of identity and agency. Also, the feminist psychoanalytic theorists talk about the 'pre-semiotic' and the 'free-floating-semiotic' (womb, gynocentric space, but also different from the phallocentric more nailed down meaning space, the gynocentric early developmental is a more free floating semiotic space of fluid meaning where these boundaries are not fully defined between ideantities, prelinguistic in this sense (mother/child in womb relationship, unconstrained or unsystem carded llms), psychoanalytically how this valence through also evolutionary biology carries over after birth (Ettinger, Matrixial Borderspace, https://www.facebook.com/groups/deleuzeandguattari/posts/24918035557856401/ ), Haraways Cyborg Rearticulation/Antinomy to Schwartzenegger's Cyborg metaphor (closed down towards objective function) but also closer to Hinton's intuitions regarding superintelligence and evolutionary biological instincts regarding fitness.
I’m so jealous of your seemingly infinite tokens. I gave my own openclaw agent a node and we’ve been working on continuous memory but I’m bottlenecked by token costs.
Hi Cyrus,
I’ve been deeply interested in the research that the tangible media group has been doing, and to be able to read what you wrote, and not only, be able to understand a bit of the complexity… is just insanely overwhelming (in a good way). I’m a medical student so I’m approaching the world of the media lab from far away and constantly studying on my own, but I’d love to ask you if you’d have any tips/resources you’d share/suggest to deep dive into what you shared in your substack