I love that there is a whole track here at South By titled “2050.” That’s the year I will turn 100 years old, so why not imagine myself looking back to this year’s SXSW to recall if any of the smart people on this panel got it right about AI?
The panelists are Allison Duettmann, Michael Graziano of Princeton, Judd Rosenblatt, and Anil Seth of University of Sussex. I am taking notes with my Vision Pro headset. The battery is at 41 percent. I have positioned my Substack editing screen to my left, so I can see the speakers sitting in their comfortable chairs on the stage. I’m typing on an Apple Magic Keyboard. I’m not sure what the advantage is of taking notes at a session with VP compared with a laptop. It’s as if I brought my Apple Studio Display and set it up between my row and the row in front of me. I’m straddling the “real” world of the room and the VR world in the headset. Spatial computing, baby. I have not seen anyone else wearing Vision Pro so far. I have to imagine it would look strange and kind of scary from the stage if everyone in the audience, or even half of us, were wearing headsets and typing during the discussion. But maybe by 2050 the panelists would all be wearing headsets as well, except they would look like normal eyeglasses and cost as much as a smoothie.
I’m not normally a fan of existential angst about AI. I just like playing with the LLMs I’ve saved as favorites on my iPhone and computer. Perplexity is my favorite, because of how clean and friendly the UI seems. Copilot and Gemini seem solid, and ChatGPT seems like the past. But I came to South By this year curious about the future, and nothing says Future these days like a panel of experts discussing Conscious AI. This makes me wonder about my own consciousness, which varies from meditative stillness while sitting on my Zafu to mushy distractedness scrolling X or watching “Suits.”
I updated the software on Vision Pro this morning at my hotel. It included improvements in the persona which is the image of myself that shows up when I Zoom or FaceTime wearing the headset. I think the new one is a little better, a little less creepy, but it’s not there yet.
And the discussion begins. Judd starts with a question: Is Anthropic’s Claude AI Conscious?
Anil Seth says we project consciousness into things that don’t have it. “It’s not conscious, but I might be wrong,” he says. Allison is from the Bay Area. She says yes and no to the question of whether Claude is conscious. We need more science about consciousness in order to answer the question. Fair point. Mike from Princeton for the past 12 years has been researching the question of consciousness—why it evolved, and can we build it. “It seems very unlikely that Claude is conscious,” Mike offers. The AI lacks the kind of self models that we have, bundles of information, constantly updated that describe the self. Sorry, Claude.
We live in a sea of perceived consciousness around us, Mike says. There is a key social aspect to our beliefs about consciousness.
Judd is CEO of a company. He lost me as he described what it does. He uses words like “impactful” and “model.” He asks why everyone is freaking out about AI these days. Does consciousness have anything to do with it?
My battery is down to 26 percent. I have a spare, but when this one runs out I’m going to take off the headset and shift to taking notes on my Kindle Scribe.
The conversation has gone deep into the meaning of consciousness. Anil Seth suggests consciousness is many things, not one big scary thing. We can chip away at it. Is it something only living systems can have? A computer simulation of the weather never gets wet or windy. Likewise an AI trained on consciousness won’t be conscious.
Mike from Princeton reminds us he’s been working on this for 12 years. Everything you think you know about yourself derives from information about the self. If you have a control system, it works better if it has a working model of the thing it controls—and ancient engineering principle. A brain better have a good working model of itself. No matter if evolution selected for it. Now, Mike says, we can build this into machines.
Only 18 percent batter left, so I’m going to post this before I lose it.