Login
I couldn’t care less if AI suffers, and so should you
A recent AEON article asks “If AIs can feel pain, what is our responsibility towards them?”.
This post could be really, really short. I may point out that speculation is good but sometimes should remain just that, and then ask why on Earth anybody should waste time worrying about causing pain to AIs that don’t even exist yet, instead of how to end the pain we humans inflict on each other. I could do that, and be done with it.
Then again, seeing how eager people are to carry certain thoughts to places they should never reach, it can be useful to summarize the main points, and why they don’t hold water in practice. Bonus content: AI faithwashing.
If AIs can feel pain...
The AEON article was written because “We once denied the suffering of animals in pain. As AIs grow more complex, we run the danger of making the same mistake.”
Up to just a few centuries ago, the default justification for acts like beating to death infant seals was something like “Can animals talk or reason? Obviously not, so there’s nothing wrong in harming them”. In the 18th century, that attitude (very) slowly started to morph into “Never mind speech or reasoning! if animals can suffer, they shouldn’t be harmed without serious reasons”.
In the 19th century that criterion was finally extended to slaves. One day, maybe, it will cover ordinary human workers too, who knows? But right now, says the article, AI forces us to change the question again to “Can something that’s not a biological species and doesn’t even have a body suffer?” Because, of course, if it does, no decent human should make it suffer without a very serious reason.
Therefore, we cannot avoid to “consider the moral status of AI [and] what suffering might look like when it is entirely disconnected from biology”. In turn, this should force us to conclude that even AIs deserve the protection of the “precautionary principle”. That’s the extremely sensible idea that, recognising the concrete or moral costs that certain actions may have, suggests to avoid them even if we aren’t sure they will actually cause harm.
Since AIs are black boxes that keep getting more complex, “the risk of dismissing the possibility of an inner life because it fits no familiar template becomes harder to ignore.” That’s why the precautionary principle compels us to treat AIs as if they might be sentient until we have strong evidence otherwise.
My answer to AIs feeling pain
The article acknowledges that “current consensus in cognitive science is that these systems remain mindless statistical machines”, but it does so just to ask “what if that changed?” My heartfelt answer is that even if that status did change tomorrow, it couldn’t matter less.
The whole thesis of the AEON piece consists of three points:
Harming without real reasons any living being that can feel pain is just wrong.
Should we find out that some artificial entity can feel pain, we should not hurt it without real reason, just like we do with humans and animals, even if that entity has no biological life of any sort, and no body either.
Since we cannot be sure that something as complex and highly unobservable as AIs cannot “feel pain”, we better not harm them, because the precautionary principle.
Point 1 is obviously right. Point 2 may sound sensible in general, but applying it to the AIs of today is a really convenient cover-up for some, and an obscuration of a much more important issue for everybody else.
How do you separate, in practice, “Don’t hurt the poor AIs because they may suffer” from demands to “not hurt our own divinity and bank accounts” by the folks who control AI?
The article asks what we ought to do “as AI systems develop forms of behaviour that mimic - or perhaps even instantiate - emotions”. That’s a cover-up again, because AI systems don’t “develop”. They are developed to “develop” behaviors by people that, so far, have almost never proved to be mature and unselfish enough to avoid monumental mistakes.
Even ignoring those folks and their interests, there is a huge difference between non-human biological beings of any kind and the AIs of tomorrow: the former came to exist and have been around for ages, independently from our will. That’s why they have our same rights to exist and not suffer (yes, I am simplifying real big here, but not in ways that hurt the point I’m trying to make).
AIs are a completely different matter, for the simple reason that real AIs do not exist yet and their actual arrival and occupation of every bit of our lives are not unavoidable at all. Not THESE AIs, for sure, even if their bubble won’t pop, which it will. Yes, I know that that whole article is about “what about the moment when they may exist?”. But that’s another smoke screen, albeit very different from the other.
If the subject is “species” that will exist only if and when we could succeed to create them, the question of whether it would be OK to harm it almost disappears before the real one which is “MUST we create this thing, and let it loose in the world?”
That’s the issue that makes that second point moot: we have no obligation to create all the AIs we could, just because we could. NO OBLIGATION. If there is one thing certain here, is that what doesn’t exist cannot suffer. Philosophical speculation is good and necessary, but actual actions and policies have different priorities.
Prove me that we MUST really create something that may also make humans suffer, before demanding we pay the slightest attention to its hypothetical sufferings. Especially if what would surely suffer from limiting creation and deployment of what passes for AI today is just someone’s stock portolio, or the juvenile urge to play gods without any unpleasant consequence.
The third argument - precautionary principle for AIs - is even easier. And funnier, in the wrong way of course.
The easy part is that if we cannot be sure that something as complex and highly unobservable as AIs cannot “feel pain”, we can’t even be sure of what makes the same AIs “feel pleasure”. That is, if we are to apply the precautionary principle as the article suggests, the only way to be sure we never make these things “suffer” is to not create them at all.
The funny-in-the-wrong-way part is the application of the precautionary principle in the wrong direction and in the wrong moments. Must we really be so stupid to ignore that the precautionary principle should be applied, first and foremost, to the very creation and deployment of AIs?
Because, you know, the whole argument that “AI is already here, it’s too late for certain questions” is a really stupid argument. Once more, “AI Is Not Unavoidable. Not This AI, That’s for Sure!”
These are products (repeat with me: PRODUCTS, not “beings”) that, well before and well below being real intelligence, are already so dangerous for your savings and pension fund that its creators and the IMF are discussing right now how to contain it... and we’re supposed to care about their suffering after they’re released? Intelligence, this is not.
What the article gets right (without even realizing it)
Here are the two passages of the articles I do like and agree with:
“Whether we take that step [to avoid pain for AIs] or not will tell us more about ourselves than about machines”
“Even if artificial intelligences do not - and may never - possess the capacity to suffer, our attitudes toward them can serve as a test of our ethical reflexes”.
I agree. Accepting to discuss how to avoid pain for AIs in the current landscape will indeed serve as a test... of our own gullibility, and reveal who’s really intelligent: we or the machines?
Thanks for sharing this post as much as you can, in and outside Substack!
The deeper, more dangerous error
Here is another, deeper reason why I say never mind about AI pain: besides being intrinsically weak, certain arguments unavoidably validate the position that AI is a “being” on more or less the same level as humans. But if that is the case, it works both ways, and we already know where that leads and how bad it is. We have already seen workers treated like robots instead of robots replacing workers. Accepting certain arguments makes it easier to accept that humans are not worth more than software, and that the first AIs, that is corporations, can keep having the same rights and privileges as actual human beings.
Finally, the AI Overton window, getting smaller and smaller
Last March, Anthropic summoned about 15 Christian leaders from Catholic and Protestant churches, academia, and the business world to get advice on how to steer moral and spiritual development of their AI “Claude” to properly handle complex and unpredictable ethical queries, participants said. According to a report from Washington Post, the discussions also covered how Claude should respond to users who are grieving loved ones and whether Claude could be considered a “child of God”.
Still quoting from the report, here are some of the issues that Anthropic and its guests explored:
“We’ve got to build in ethical thinking into the machine so it’s able to adapt dynamically.”
Anthropic genuinely cares about Claude’s wellbeing
Anthropic leaders frequently talk about the need to give Claude a moral character
What does it mean to give someone (not “something”) a moral formation?
What’s wrong with this apparently reassuring picture? At least two things, actually:
considering a chatbot a “child of God” provided of “wellbeing” that’s also a recipient of “moral formation” is exactly what I called the “deeper error”. What needs serious moral formation are the people who develop chatbots and turn them loose on people who by and large are unequipped to handle the impact
the main real religion actually professed in that summit was the very american religion that “AI is unavoidable”. As I said, it isn’t. What religious leaders of any faith should teach in Silicon Valley is restraint, not faithwashing of dangerous products, or limiting what should be discussed about AI products
Especially if those products are - quoting from this synthesis of a Guardian piece you really should read - “being trained to mirror everything but us at our most authentically human.” Here is the gist of that argument (emphasis mine):
“The models learn from textbooks, social media posts, and scripted film and television dialogue. They have virtually no access to the unscripted, face-to-face, voice-to-voice conversations that make up the vast majority of human communication.”
“That blind spot is not just a technical limitation. It is a cultural feedback loop that is already running.”
“As billions of people absorb AI-generated text daily, they are beginning to absorb its patterns. The machine is not learning to sound like us. WE are learning to sound like it.”
Of course. This is the logical consequence of what I said above: accepting certain arguments makes it easier to accept that humans are not worth more than software. And when that happens, the result is not humanized AIs, but robotized humans. So, don’t give a f**k about any pain the AIs may feel someday. Worry about who imposes them today as just another, but really lousy religion instead.
Important final note to Anthropic, all their competitors and everybody else...
... that is, everybody aiming to reboot AI for the real common good: I do get a thing or two about AI right, and I’ve been preaching the similarities between sincere faith and digital technologies that make sense for twenty years now (*). If what you want is AI that makes sense for everybody, give me a call or (thanks!) sponsor my writing, by subscribing or with donations for any amount via PayPal or LiberaPay.
(*) to see what I mean, read the “Why I PERSONALLY care about all this stuff” paragraph of “Two great “Software for Catholics” initiatives and what may make them even more helpful... for everybody”
