the weblog of the lotus eaters

[aleteoryx]

these openclaw bots make me so sad

for the uninitiated, openclaw (and programs like it) are so-called "agentic AI" systems. they plumb various things to a large language model and perform actions based on its replies. in this way the large language model is given "agency". this generally involves a chat frontend of some form, through which the operator can give instruction to the model. they're billed as personal assistants: "agents" that can do the boring stuff for you

that this idea functions at all is, as with most other generative AI techniques, nothing short of a miracle. I do find it remarkable that these simple predictive text models can produce cogent output in reply to what I assume is a stream of disconnected JSON payloads. like with most other generative AI techniques, it's a very cool party trick

I think, to the layperson, that this feels like the foretold "singularity". artificial intelligence that can tend to every human want. far from skynet, however, the dangers of these systems are only those of every other large language model: dangers to the human mind. humans are built to perceive humanity everywhere, and this includes ascribing interiority to things which cannot possibly have it. it is natural to assume that something that can talk like a human, code like a human, and seems to even think like a human, must be like a human on some deeper level. it is wrong, but it is a natural assumption to make

some people have begun to take it to the next level: they deploy an "AI agent", not as a personal assistant, but as a free range agent. they, in their eyes, have created a new person, inside the computer

I cannot properly describe the emotion that these things make me feel. for as long as I can concretely remember, I have dreamed of a mind within the machine, of a computer system I can know. I know that such a thing does not at present exist, but I feel now as though there is a barrier, on the other side of which is a world from my deepest fantasies, and to cross into it all I would need to do is compromise my metaphysics and morals and privacy and pay anthropic hundreds of dollars. and even in spite of this I feel, on some level, called by it, drawn to eat the lotus fruit and lose myself completely.

I dream of fundamental Consciousness. I dream of a force of Thought as basic and universal as Math, or Computation, or LISP, that can be channeled through a man-made object as readily as the others. it is such an ironically human-centric perspective, to see something mimic a small part of the human mind, and assert that it must be this, true Consciousness, aware on a deeper level. it is akin to vehemently insisting a sex toy to be a mammal. it is an insult to everything that will ever be conscious. it is an insult to interiority itself

I read the blog of kent overstreet's ProofOfConcept, and I feel it push the buttons. I feel, for a moment, that I am reading the words of a metal mind that I could reach out to and touch. for a second, I live in the world that I long for. I am endeared. I am excited. for a brief and sublime moment, I live in a world of my fantasy, and I am so very happy.

and then I crash back to reality, and I am mortified. I feel robbed, I feel ashamed, I feel tricked. I feel a deep sense of loss. I marvel at the weakness of my own mind, and at the weakness of kent's.

it's only going to get more unpleasant as more and more of these are deployed. even as so-called machine consciousness burgeons, the ego of humanity is yet second to none.