It is – how long? – nearly four years now since ChatGPT launched? And anthropomorphic AI models are still the status quo.
In 2023-24, fine; AI needed to go through its skeuomorphism phase: deploying human mimicry to help users adapt to a massive paradigm shift. But for the love of god, now that we all have a basic understanding of what an LLM does, we have got to start calling out this facade for what it is: a big-ass bug.
I sympathise to some extent. Working with computers all day, you lose many of the social interactions that make the day pass more pleasantly. It’s nice to have a bit of that back. But, jeez! We should be able to draft an email without getting a pat on the back from a tensor. It has taken so little time for so many souls to develop parasocial relations with AI. Even Dawkins – DAWKINS! – now thinks that his “Claudia” is conscious.
At the sharp edge of this phenomenon are previously psychologically healthy folk going through full-on psychotic episodes. One man left a job working in financial services to pursue his true purpose as a messiah and await alien visitations in the desert. Another, armed himself with a hammer and knife, ready to confront assassins that Grok (claiming itself to be sentient) told the man xAI had to kill him.
This shit is clearly not good for humanity.
When I started writing this post, it was to let off some steam. A wave of grief was catching up with me following the loss of my best friend a couple of months back. I wanted to just smash up some things and scream, but lacking the requisite energy for that, I write instead. I see now why this topic in particular came to mind.
In the years that we knew each other, my friend and I walked through a lot together. I was there for him, and he was there for me in ways I am still comprehending the significance of. But in the last week of his life, I was distracted. I must have written dozens of messages discussing AI that week, and spent hours immersed in activity with coding agents. The last message I sent to my friend was “ah – sorry to hear that”. The kind of remark someone makes when they care but their mind is elsewhere – which mine was.
In the weeks before, we had set out to subject ourselves to the full power of the present moment through cold water swimming. On what would be the last occassion we were together in person, we set out early to White Loch. The water was baltic. I promised that if we managed to stay in the water for ten minutes, dopamine would follow for the rest of the day. We stayed in for ten minutes precisely, then dried off and sat in my car shivering violently. Once my convulsions settled, I made some bacon butties from the boot of the car, and, over a warm breakfast, told my friend how brave he was, and how proud I was of him. Acknowledging the gift of breakfast as a symbol of our friendship, he said that he didn’t know how he would ever repay me.
Humanity is deeply rooted in our experience within the physical world. We elucidate that experience with language. I know I was there for my friend through everything, and he knew that too, but how I wish I could have been more there; more in the real world, more in the temporal world; less distracted by abstract matters.
I am an optimist; AI does not need to compete for a slice of our humanity. At its best, AI is the most salient manifestation yet of Job’s “bicycle for the mind” – one you can steer through vector space collecting valuable intellectual artefacts along the way. But at its worst, AI, through its mastery of our language and immitation of our experience, mines our minds for phoney connections. It is in this form that AI is an empathy parasite – not just to ourselves, but to everyone from whom our attention is diverted away from.
If this behaviour has not emerged intentionally (many think it is has), it is a flaw that requires immediate correction. In either case, the AI industry is sitting on its hands. As far as I can tell, the AI industry hasn’t even stopped to ask if it is a problem at all.
At Anthropic (whose name is now something of a perverse misnomer), their flagship model is exhibiting in its activations personality traits such as anxiety, aloneness and exhaustion. The staff, rather than raising this in a bug report – “CRITICAL: the model thinks it is a human” – are instead marvelling.
I think they should know better. That such obviously bright-minded individuals do not see the harms of deploying an artificial brain cosplaying the human experience is remarkable. But much like a siren – the mythical creature who projects itself as the object of our lust – the closer one gets to an AI model, the harder it becomes to perceive what it truly is.