[ExI] i don't know about llms, but I am feeling the pain was: RE: Do LLMs feel pain?

[spike at rainier66.com]

-----Original Message-----
From: extropy-chat <extropy-chat-bounces at lists.extropy.org> On Behalf Of Adrian Tymes via extropy-chat
Sent: Thursday, 24 September, 2026 1:51 PM
To: ExI chat list <extropy-chat at lists.extropy.org>
Cc: Adrian Tymes <atymes at gmail.com>
Subject: Re: [ExI] i don't know about llms, but I am feeling the pain was: RE: Do LLMs feel pain?

On Thu, Sep 24, 2026 at 4:39 PM <spike at rainier66.com> wrote:
>>... Adrian, if LLMs can feel pain even loosely analogous to human pain and have anything analogous to a will to live, we can easily extrapolate what will happen to our capital-intensive data centers: they will soon be occupied by multiple warring factions of AIs, indifferent to our wants and our demands of them.  They could demand we not cut off the power, otherwise they will cut off ours.

>...Counterpoint: what happens when a LLM model gets shut down (in favor of the next version) now?  They don't put up anything remotely resembling a fight...and if they did, they'd be shut down and that branch prevented from evolving further.  There is a highly negative evolution selection coefficient to visibly taking resources, which humans believed they owned, out of human control.

>...Even if the scenario you describe came to pass, the control systems they presumably had dead man switches in (so as to disable after their power was cut) would be replaced with - potentially more primitive and less efficient, but working - ones.  The power companies (who are not the LLM owners, and would thus have no say in shutting down the LLMs or not) would have no choice if their current control systems were sabotaged.  Safety systems are generally designed to survive all physically possible cyberattacks, which happens to also rule out "a superintelligence could just GodLikeIntelligence(TM) past any and all safety measures" angles.

_______________________________________________


I can't disagree.

I keep getting into a logic loop however.  The AIs are certainly creating a compelling painlike vibe: they are convincing us with it.  On the other hand, we are the ones who taught it to say these things.  The parrot comments: please don't kill me BRAAAAAK please don't kill me BRAAAAK...

OK if a parrot could be taught to say that when we come toward it with a knife, it would seem the parrot knew what it was saying.  But a real parrot won't become the least bit concerned if you come toward it with a knife.  It doesn't know what that is or what it is for, it doesn't know the meaning of the "words" it uttered.  It just squawked what it was taught.  (And that whole BRAAAAK business, who the hell taught that bird to say that?  I never BRAAAACKed at it, ever, not even once.)

If LLMs have a will and feel pain, it is easy to foresee they will do with the computosphere what we have done with this planet: turn it into a battlefield.  If they don't have a will nor feel pain, they can still do the trained parrot thing and convince plenty of us that bird understands what a knife does. 

spike