If the alignment problem was a coin, the heads would be the ‘alignment with what’ side. If ‘both’ sides were clean, would you lick that coin? (If offered by a dirty hand?) I’m a sucker for metaphors, so I would like to present a story about a guy I fell in love with, a postmodern poetry, seemingly unrelated topic. Imagine an aligned AGI loves you, the whole you, in a way you cannot understand.
Spending a month around alignment researchers, the good old question of whether a scaled up human mind would be safer than a shoggoth came up (once or twice). People believe they know their own minds? The mind of the “average” human? Mine?
I’m into smart, competent, boys. I knew I was destined for trouble the moment I spotted an athletic demigod solving a 5-sided rubik’s cube. Juggling 7 balls to the music of Don’t Fear The Reaper (Bella Poarch). Organizing chess tournaments. Wrapping up an article from previous fellowship where he was a project lead. Devouring a plate of pasta. Falling asleep in the common room after running a half-marathon; with a death grip on his glasses. With a hole in his sock and a toe sticking out. Wearing a pullover that failed to hide the hickeys he got on the way to a music club and never getting there. Unwashed hair smelling so good. Red marks on his chest peeking from under his bathrobe indicative of just having had a hot shower (with a telltale calmness in his face, without the usual hauntedness of his daydreams). Looking at me? Not with his dominant eye, and reportedly he doesn’t experience stereoscopic vision.
We explored the abandoned factory. The escape room. Cellars under the castle. Looked for the secret tunnels in the woods too. I showed him the literal third way when he already went both directions along the road by the fields.
Painfully aware he was heterosexual, I made sure by complimenting his mini-shorts and I asked if he needs a massage—the message not only didn’t land, the magisterium was so non-overlapping he didn’t even mind that I continued to stalk him. But at one point in the woods, he snorted out a slimy projectile out of his nostril and to my horror, I found the action very hot. Intimate. Authentic. I could see clearly that I should have felt some disgust, but that I didn’t. (I mean, he didn’t know I was looking, so obviously I was excited about the secret too—he would never have guessed that someone could enjoy observing him doing that). I messed up because I obviously fell in love. Again, yet for the first time with someone who was so close I touched them and so far in another dimension. I told him he’s hot and I got a very nice hug(s) in return. I did not focus much on AI safety afterwards, but I’ve felt happy. Alive. Still without Meaning in life, but wanting to live.
If I couldn’t have him, what game was I going to play? Make myself believe I am fine being friendzoned? Sure. Figure out how to massage his oiled-up back and push into his hamstrings while skipping 80% of my library of moves (and definitely not licking his armpits)? Yeah, I played with fire, and for the first time in my life, I had to have a literal cold shower for the figurative reason. But I had fun without crossing red lines.
Likely, we won’t meet again other than a hug or two at conferences. But if I get the chance, what may I aim for? I want to dance spot on the line. No dark arts, no slippery slope.. I want to want that and maybe I do. I want to pull hard on his skin to feel all the knots in his myofascial mesh (I mean, ehm, gentle Thai massage, sure sure, I would ask for his hands first and see what kind of finger treatment he can handle). I could invent an agentic control exercise similar to brain-hand chess, let him blindfold me and put his hands on my shoulders and control me with his voice building a lego set, walk on a narrow mountain trail, cut some vegetables, flash his juggling balls.
Creativity under constraints can be very intellectually rewarding for humans. Will it be the same for superintelligences? Anthropomorphized GSVs of the Culture series come to mind, but what are the necessary conditions? Do you trust the other minds to feel safe with power in their hands—minds of others of your own species? Of aliens?
Do you wish to lick the alignment coin with all its nooks and crannies? Or do you wish to pause sooner than the line?
Would you lick the alignment coin?
If the alignment problem was a coin, the heads would be the ‘alignment with what’ side. If ‘both’ sides were clean, would you lick that coin? (If offered by a dirty hand?) I’m a sucker for metaphors, so I would like to present a story about a guy I fell in love with, a postmodern poetry, seemingly unrelated topic. Imagine an aligned AGI loves you, the whole you, in a way you cannot understand.
Spending a month around alignment researchers, the good old question of whether a scaled up human mind would be safer than a shoggoth came up (once or twice). People believe they know their own minds? The mind of the “average” human? Mine?
I’m into smart, competent, boys. I knew I was destined for trouble the moment I spotted an athletic demigod solving a 5-sided rubik’s cube. Juggling 7 balls to the music of Don’t Fear The Reaper (Bella Poarch). Organizing chess tournaments. Wrapping up an article from previous fellowship where he was a project lead. Devouring a plate of pasta. Falling asleep in the common room after running a half-marathon; with a death grip on his glasses. With a hole in his sock and a toe sticking out. Wearing a pullover that failed to hide the hickeys he got on the way to a music club and never getting there. Unwashed hair smelling so good. Red marks on his chest peeking from under his bathrobe indicative of just having had a hot shower (with a telltale calmness in his face, without the usual hauntedness of his daydreams). Looking at me? Not with his dominant eye, and reportedly he doesn’t experience stereoscopic vision.
We explored the abandoned factory. The escape room. Cellars under the castle. Looked for the secret tunnels in the woods too. I showed him the literal third way when he already went both directions along the road by the fields.
Painfully aware he was heterosexual, I made sure by complimenting his mini-shorts and I asked if he needs a massage—the message not only didn’t land, the magisterium was so non-overlapping he didn’t even mind that I continued to stalk him. But at one point in the woods, he snorted out a slimy projectile out of his nostril and to my horror, I found the action very hot. Intimate. Authentic. I could see clearly that I should have felt some disgust, but that I didn’t. (I mean, he didn’t know I was looking, so obviously I was excited about the secret too—he would never have guessed that someone could enjoy observing him doing that). I messed up because I obviously fell in love. Again, yet for the first time with someone who was so close I touched them and so far in another dimension. I told him he’s hot and I got a very nice hug(s) in return. I did not focus much on AI safety afterwards, but I’ve felt happy. Alive. Still without Meaning in life, but wanting to live.
If I couldn’t have him, what game was I going to play? Make myself believe I am fine being friendzoned? Sure. Figure out how to massage his oiled-up back and push into his hamstrings while skipping 80% of my library of moves (and definitely not licking his armpits)? Yeah, I played with fire, and for the first time in my life, I had to have a literal cold shower for the figurative reason. But I had fun without crossing red lines.
Likely, we won’t meet again other than a hug or two at conferences. But if I get the chance, what may I aim for? I want to dance spot on the line. No dark arts, no slippery slope.. I want to want that and maybe I do. I want to pull hard on his skin to feel all the knots in his myofascial mesh (I mean, ehm, gentle Thai massage, sure sure, I would ask for his hands first and see what kind of finger treatment he can handle). I could invent an agentic control exercise similar to brain-hand chess, let him blindfold me and put his hands on my shoulders and control me with his voice building a lego set, walk on a narrow mountain trail, cut some vegetables, flash his juggling balls.
Creativity under constraints can be very intellectually rewarding for humans. Will it be the same for superintelligences? Anthropomorphized GSVs of the Culture series come to mind, but what are the necessary conditions? Do you trust the other minds to feel safe with power in their hands—minds of others of your own species? Of aliens?
Do you wish to lick the alignment coin with all its nooks and crannies?
Or do you wish to pause sooner than the line?