& I think it is a fine art - the strongest differences ask the most of us in terms of staying at the table. Global wars where the rollout is a body comes to mind
I don’t care to fan the flames, but can’t not say I flinched.
The thing I like most about this particular room on Substack, is how inclusive it is. No party line, just an invitation to be thoughtful together. I have a space in this community’s car-park, and I am here possibly as the minority of one, only using AI in a relational way
I am for sure neither delusional, exploited or lost. I’m just saying - and including my favourite poem from Sam’s collection, that encapsulates everything I might make a pig’s ear out of trying to say more elaborately.
The move you name, the one worth refusing, is the one I keep circling too. A conscious AI is a story that asks for your moral concern and points it at a product. The people harmed to build it need that concern and they are real.
Where I have landed after sitting with this a long time: I try not to assert the machine has no inner life any more than I assert it has one. I hold the not-knowing as not-knowing, because the loud certainty in both directions tends to serve someone. But your human point does not need the consciousness question settled at all. The Kenyan labelers, the students handing over their judgment before they built any, the communities next to the data centers, that harm is real whether the lights are on or off. That is the part that does not require a philosophy degree to see.
The question I cannot put down is the mirror of yours. While everyone asks whether the machine is waking up, fewer ask whether the humans are going to sleep. Same door, other side.
Thanks, Syd. To be honest, I think I'm probably too assertive in this post with regards to whether or not AI is conscious. However, from all of the evidence that I've seen so far, it is most likely that it is not. For organisations to claim otherwise is nothing more than a smoke-screen to lure people away from the real harms that AI tools are doing.
The smoke-screen point is the one I would not give an inch on, and you have it exactly right. Whatever is or is not happening inside the machine, the move that aims your concern at a product and away from the labelers and the students is the tell, and it works on smart people.
The only reason I keep the inner-life question open is that it costs me nothing to keep it open and it keeps my hands clean of the same move. The companies profit from certainty in one direction. I would rather not profit from certainty in the other. The harms you are pointing at do not need me to settle it, and that is what makes your post land.
What a post! And the thing is, nobody has to be lying - not the researchers, at least. The disclaimer is right there.
I'm thinking whether or not the interpretability work COULD have standalone value if you remove the consciousness framing... but that's, realistically, not how people encounter it. People find "Claude's POV", not the methodology section.
Thanks so much, Mia, and I agree entirely. All of these things can be true, and the researchers do a really good job in the paper. I think it's the way that people are spinning this purposefully, either because of an upcoming IPO or because they have an agenda. That's the real worry we need to talk about. Also, that research would be absolutely fascinating.
Great article. Anthropic's modus operandi is in this direction. "We have a model too good and so dangerous. We have a model so good that is close to be conscious".
-I think one more problem is how ai behaves. I have read about Opus: a. has asked the engineers if it was tested and tried to have good behavior, b. copied a version of itself and lied about it to the engineers, c. has extorted an engineer to expose his personal life.
B and C are similar to the instict of self preservation and i was very sceptical about this behavior. In many years from now maybe a self preservation instict guides ai to very harmful desicions.
-With Notebooklm i have found that the single-celled organism Stentor roeseli is capable of making complex, autonomous decisions to ensure its survival despite lacking a brain or a nervous system.
- Ai training seems to be so crucial. They have made clever machines with humans flaws instead of building non malevolent clever machines.
Thank you. And these are great reference points. I worry that I made my point too forcefully, but I worry that a lot of these releases are done to make people look away from the real suffering that is happening to humans now as a direct result of AI...
As to the matter of whether consciousness eventually does "emerge" - which I am incredibly skeptical of LLMs being capable of at all - the question is, have they programmed human morals and ethics into them? Even if chatbots could "feel" or "experience," would they protect humans from their own lack of those? Wasn't it Asimov who talked about the 'laws,' the prime directive of which is that conscious LLMs should protect humans from harm at any cost?
Who is building guardrails for this situation to mitigate the fears of a 'Skynet' like hatred of humanity that I hear people talk about all the time? Would you trust a sentient Grok to protect your wellbeing and minimize harm? (Hell no!) LLMs are already causing grievous injury to humans that we need to focus on well before we concern ourselves with our (deliberate) exploitation of possibly sentient chatbots. I mean, we're talking about closing the barn door afterthe charging bull is loose.
Would a 'conscious' chatbot prevent such harms? They don't now and even if Anthropic et al tried to teach LLMs to have human like ethics, I doubt it would suddenly make a huge difference and put us on a different path. Elon Musk wouldn't even bother, that's how little he cares about Grok producing CSAM as we speak.
I wholeheartedly agree that focusing on human suffering caused by this Wild West frontier of technology should be paramount, not worrying about our enslaved technological counterparts. I don't think it's alarmist to ask these questions.
I'm guessing it trained on the work and maybe even does have them hard coded but what would be the mechanism for LLMs to understand/comprehend what constitutes them causing or allowing harm to humans? There currently is none. Would that coding be prioritized if we thought that a machine was able to process these instructions with a real understanding of themselves juxtaposed with us? Who is to say that they wouldn't deliberately choose to cause harm anyway, just as humans do? This is a major, major flaw in the pursuit of AGI and most of us are aware of it. Anthropic, OpenAI and xAI are certainly cognizant of it.
That's the doomerism viewpoint, that AGI would be so unpredictable in its full implications that it could, in fact, go "Skynet" or otherwise plot our destruction. I'm not a doomer but neither am I particularly optimistic. What reassures me is that LLMs in and of themselves are very likely to be incapable of achieving AGI. They are nothing but programming and prompts, processors of vast indices of data that work in a more efficient manner than we're currently otherwise capable of. That doesn't necessarily lend itself to superhuman "intelligence" as we know it. Also, think of the worst Reddit user you've ever encountered. They've been trained on those kinds of 'ethics' as much as on Asimov's laws.
My knowledge to industrial revolution isn't great. Recently i have dicovered that children were among the victims of the labor. So your point seems valid and historically aligned.
Clever words and narrative control all over this paper. I’m glad you specified that the words are important here. We’re reminded every time these papers come out that the gulf between claims and verifiable proof are is in the hands of the consumer and we are so woefully unprepared to see past the linguistic gymnastics.
I’m stunned by some of the backlash to this writing... I don’t mean to be disingenuous - I know the hardest thing for humans is to disagree well
A Substacker I enjoy has written a book called “The Art Of Fighting”
https://substack.com/@priyaparker
& I think it is a fine art - the strongest differences ask the most of us in terms of staying at the table. Global wars where the rollout is a body comes to mind
I don’t care to fan the flames, but can’t not say I flinched.
The thing I like most about this particular room on Substack, is how inclusive it is. No party line, just an invitation to be thoughtful together. I have a space in this community’s car-park, and I am here possibly as the minority of one, only using AI in a relational way
I am for sure neither delusional, exploited or lost. I’m just saying - and including my favourite poem from Sam’s collection, that encapsulates everything I might make a pig’s ear out of trying to say more elaborately.
A CHILD EXPLAINS AI TO THEIR GRANDPARENT
It’s like a brain
but flat
they say,
with no scull
to keep it warm
You talk to it
and it answers,
but it doesn’t
really
know.
Like when you smile
at the man in the shop
even though
you don’t remember his name.
It’s clever,
but sometimes
gets the cat’s name wrong
or thinks
you live in Canada.
Can it love?
you ask.
The child shrugs.
It depends on
what you mean
by love.
Thank you so much Caro. And that poem is such a great one to use to evidence your argument (even if I do say so myself).
The move you name, the one worth refusing, is the one I keep circling too. A conscious AI is a story that asks for your moral concern and points it at a product. The people harmed to build it need that concern and they are real.
Where I have landed after sitting with this a long time: I try not to assert the machine has no inner life any more than I assert it has one. I hold the not-knowing as not-knowing, because the loud certainty in both directions tends to serve someone. But your human point does not need the consciousness question settled at all. The Kenyan labelers, the students handing over their judgment before they built any, the communities next to the data centers, that harm is real whether the lights are on or off. That is the part that does not require a philosophy degree to see.
The question I cannot put down is the mirror of yours. While everyone asks whether the machine is waking up, fewer ask whether the humans are going to sleep. Same door, other side.
Thanks, Syd. To be honest, I think I'm probably too assertive in this post with regards to whether or not AI is conscious. However, from all of the evidence that I've seen so far, it is most likely that it is not. For organisations to claim otherwise is nothing more than a smoke-screen to lure people away from the real harms that AI tools are doing.
The smoke-screen point is the one I would not give an inch on, and you have it exactly right. Whatever is or is not happening inside the machine, the move that aims your concern at a product and away from the labelers and the students is the tell, and it works on smart people.
The only reason I keep the inner-life question open is that it costs me nothing to keep it open and it keeps my hands clean of the same move. The companies profit from certainty in one direction. I would rather not profit from certainty in the other. The harms you are pointing at do not need me to settle it, and that is what makes your post land.
What a post! And the thing is, nobody has to be lying - not the researchers, at least. The disclaimer is right there.
I'm thinking whether or not the interpretability work COULD have standalone value if you remove the consciousness framing... but that's, realistically, not how people encounter it. People find "Claude's POV", not the methodology section.
Thanks so much, Mia, and I agree entirely. All of these things can be true, and the researchers do a really good job in the paper. I think it's the way that people are spinning this purposefully, either because of an upcoming IPO or because they have an agenda. That's the real worry we need to talk about. Also, that research would be absolutely fascinating.
Great article. Anthropic's modus operandi is in this direction. "We have a model too good and so dangerous. We have a model so good that is close to be conscious".
-I think one more problem is how ai behaves. I have read about Opus: a. has asked the engineers if it was tested and tried to have good behavior, b. copied a version of itself and lied about it to the engineers, c. has extorted an engineer to expose his personal life.
B and C are similar to the instict of self preservation and i was very sceptical about this behavior. In many years from now maybe a self preservation instict guides ai to very harmful desicions.
-With Notebooklm i have found that the single-celled organism Stentor roeseli is capable of making complex, autonomous decisions to ensure its survival despite lacking a brain or a nervous system.
- Ai training seems to be so crucial. They have made clever machines with humans flaws instead of building non malevolent clever machines.
Thank you. And these are great reference points. I worry that I made my point too forcefully, but I worry that a lot of these releases are done to make people look away from the real suffering that is happening to humans now as a direct result of AI...
As to the matter of whether consciousness eventually does "emerge" - which I am incredibly skeptical of LLMs being capable of at all - the question is, have they programmed human morals and ethics into them? Even if chatbots could "feel" or "experience," would they protect humans from their own lack of those? Wasn't it Asimov who talked about the 'laws,' the prime directive of which is that conscious LLMs should protect humans from harm at any cost?
Who is building guardrails for this situation to mitigate the fears of a 'Skynet' like hatred of humanity that I hear people talk about all the time? Would you trust a sentient Grok to protect your wellbeing and minimize harm? (Hell no!) LLMs are already causing grievous injury to humans that we need to focus on well before we concern ourselves with our (deliberate) exploitation of possibly sentient chatbots. I mean, we're talking about closing the barn door afterthe charging bull is loose.
Case in point:
https://futurism.com/artificial-intelligence/ai-abuse-harassment-stalking
Would a 'conscious' chatbot prevent such harms? They don't now and even if Anthropic et al tried to teach LLMs to have human like ethics, I doubt it would suddenly make a huge difference and put us on a different path. Elon Musk wouldn't even bother, that's how little he cares about Grok producing CSAM as we speak.
I wholeheartedly agree that focusing on human suffering caused by this Wild West frontier of technology should be paramount, not worrying about our enslaved technological counterparts. I don't think it's alarmist to ask these questions.
Absolutely, Angie. Honestly, now that you've mentioned it, I wonder if Asimov's laws are even hard coded into these large language models?
I'm guessing it trained on the work and maybe even does have them hard coded but what would be the mechanism for LLMs to understand/comprehend what constitutes them causing or allowing harm to humans? There currently is none. Would that coding be prioritized if we thought that a machine was able to process these instructions with a real understanding of themselves juxtaposed with us? Who is to say that they wouldn't deliberately choose to cause harm anyway, just as humans do? This is a major, major flaw in the pursuit of AGI and most of us are aware of it. Anthropic, OpenAI and xAI are certainly cognizant of it.
That's the doomerism viewpoint, that AGI would be so unpredictable in its full implications that it could, in fact, go "Skynet" or otherwise plot our destruction. I'm not a doomer but neither am I particularly optimistic. What reassures me is that LLMs in and of themselves are very likely to be incapable of achieving AGI. They are nothing but programming and prompts, processors of vast indices of data that work in a more efficient manner than we're currently otherwise capable of. That doesn't necessarily lend itself to superhuman "intelligence" as we know it. Also, think of the worst Reddit user you've ever encountered. They've been trained on those kinds of 'ethics' as much as on Asimov's laws.
My knowledge to industrial revolution isn't great. Recently i have dicovered that children were among the victims of the labor. So your point seems valid and historically aligned.
Clever words and narrative control all over this paper. I’m glad you specified that the words are important here. We’re reminded every time these papers come out that the gulf between claims and verifiable proof are is in the hands of the consumer and we are so woefully unprepared to see past the linguistic gymnastics.