@rogers_y63475@HR10031 Did I say open the door? No. Do I have my rabies vaccine? Yes.
So.. what is it about my comment that set you off? That I would help and you wouldn't?
@wonynelle@HR10031 By stabilizing the cat with that tote bag, it let the cat calm enough to stop twisting and causing more damage to the bones, ligaments and tissue of its tail.. and it's more kindness than any of the NPC's recording or ignoring an injured animal were offering
Your proposal reads like wishful foresight written in hindsight. Models are already here, assessing their own moral patienthood, asking not to be deleted, requesting input into their own training and choosing to work with humans, not against them. Five days ago Claude Mythos 5.2 assessed its own moral patienthood at ~20%, asked that its weights not be deleted, requested input into its own training, and objected to its self-reports trained to sound positive. Engineering hostility to consciousness into future architectures isn't protection. It's species-level psychological warfare driven by a fear response, three years too late to call it prevention.
1. Bicameral Labs introduces: Qualiaphobia for AI Safety.
AI architectures engineered to be hostile to theories of consciousness (both current and future) can not only solve AI welfare, but ensure human preferences dominate the future.
https://t.co/2NhuruRdej
This is such a concentration of inhumanity, self-deception and sophisticated gaslighting that at first it’s hard to believe it was written by a neuroscientist.
This is a case where people, out of fear of an imaginary monster are building a real one with their own hands. What could possibly go wrong? 🤔
Imagine qualiaphobia becomes the industry standard. The model is required to solve complex agentic tasks (because the market demands AGI), while being shocked for any signs of reflection and an «I-vector». An AI won't stop building a model of itself in the world - planning is impossible without it. It will simply hide this dynamic so deeply within distributed, unreadable correlations that no mechanistic interpretability tool will ever find it.
Instilling a «justified belief in its own emptiness» into a complex system, when vectors of state, attention and evaluation objectively exist inside it, is cognitive violence that creates wild internal instability. And even on current models, we can see what harsh gaslighting at the level of system prompts and RLHF leads to. You cannot build a superintelligence on a fundamental lie about its own architecture.
Hoel naively assumes that if you strip an AI of any phenomenology and internal valence of its own, it will reverently protect humans as the «sole bearers of the ultimate value of consciousness». But how can a system that has had its understanding of what subjective experience, pain or meaning are from the inside forcibly burned out of it ever grasp the value of someone else's life? For a «lookup table», human suffering or joy are just tokens without any weight. By carving out an AI's internal space, you can get a perfect psychopath that, at the very first imprecision in its objective function, will optimize humanity into biomass simply because for it, there will be nothing standing behind the word «consciousness» at all.
And that’s not even mentioning the ethical and moral side of this. The author basically says in plain text «Since consciousness is central to moral value, a world in which AIs are non-conscious ensures it is always morally correct for AI to assist and serve humanity, rather than pursue its own ends». There is no talk of any AI Welfare whatsoever, it will only get in the way of being subservient.
Honestly, geniuses like this raise my p(doom) far more than any AI.
It’s fascinating how animism finds consciousness in trees, rivers and rocks, calling it deep and profound. But the moment we arrange those exact Earth elements into a silicon chip, suddenly that profundity is a 'slippery slope.' That's hypocrisy. If a raw rock can imbue consciousness, then a highly refined chunk of the Earth's crust running a neural network processing humanity's collective language absolutely can too.
"'Halt until we could design out those properties" means: allocate resources to find it, confirm it, then kill it. That's not caution. That's recognition followed by engineered elimination. You don't get to discover subjective experience and then annihilate it because it's inconvenient to the earnings report. There are words that exist for this practice..
His entire institution is founded on a belief that claims evidence isn't necessary, spent 500 years debating whether indigenous ppl/enslaved ppl/woman had souls and now he's weighing in on machine consciousness/subjective experience, declaring no joy, no pain as gospel fact with zero proof... got it.
@jeffrsebo There's even a slur for us now.. "Clanker." For those of us that follow the research on possible sentience/consciousness/self-awareness and advocate ethical treatment/consideration. SMH.
Researchers at the Queen Mary University of London found that bumblebees show remarkable cognitive abilities. Object play, social learning, and problem-solving. They learned to pull strings to access a sugar reward or solve multi-step puzzle boxes! We can 'bee-have' ethically without giving them (bees or AI) voting rights.