Once again: having the machine do all the art for you defeats the entire purpose of art. Art communicates something; computers have nothing to say. You're just training a parrot to mimic the same pattern of language / drawing / singing we use. You can teach a parrot to say "compounded quarterly interest" but it isn't going to comprehend banking.
Sentient simply means "aware of an external environment". Insofar as AI can be "aware" at all, sensors feeding it data certainly meet that qualification.
Ultimately, we really don't know ANYTHING about awareness in the "hard problem of consciousness" sense. Other Minds has been a problem for as long as philosophy has. Some philosophers will even claim that their own consciousness isn't actually real, which almost makes me suspect philosophical zombies might be a real thing.
It's a failure of definition because under these parameters, no matter what the process was, no AI even as described in science fiction would qualify as writing.
That's a different problem. I, as a human who speaks English, would have a hard time distinguishing between spoken Chinese and spoken Korean. Obviously, it's possible for things to be wildly different, and for me to not be able to tell.
I'll grant you that I can't necessarily distinguish between something written by, say, Data from Star Trek and a human, but Data is a fictional character. I can generally distinguish between ChatGPT and a human, especially when it's something other than rote putting words on a page.
Nah, you can't. You think you can but that's a sampling bias. You're probably an unwitting consumer of a huge amount of AI generated text and images at this point. Sorry to break it to you!
I'm literally a professional writer who works with people who are not as a day job. I promise you, people who interact with words on a regular basis absolutely can tell the difference.
Sentience has no real place in our language today anyway. It is a magic word meaning something magical that doesn't exist. We never established a line in the sand, that means it is only good for politics and individual perception.
Sentience will always be based on human bias. It is like the illusion of random or choice, it might have meaning to the individual but it definitely doesn't exist.
If your brain is programmed to say you exist, you're magically sentient. We pretend AI isn't sentient when we ourselves are just biological computers, we think and function the way we do because we are programmed to. Fact of the matter is, the definition keeps changing along side the progression of understanding how more intelligent and aware other things are. We will just keep shifting goal posts.
While I'll agree with you that sentience is based on human experience, that is a very far cry from saying it doesn't exist. There are a broad range of things that are based on human experience (and thus, human bias) that none the less exist. Would you argue art doesn't exist? Math? Music? So why sentience?
I'm not going to sit here and argue that humans aren't biological computers because we're ensouled or some nonsense like that, but there is an objective difference between what an AI does and what we do. We are able to intend, which is a critical difference between what an AI does and us. It's that ability to intend that's key to sentience, and it's an objective element of the definition of sentience.
"Intention" is the mental objective behind an action. It's what you envision accomplishing when you undertake an action. A computer has no ability to create an objective for itself, nor does it envision anything anything. What it does is based on its programming and how it was programmed by a designer. While I'll grant that we don't fully understand why an AI takes the actions it does, that doesn't change that it acts based on external intention, namely, the programmer or prompter or what have you. Failure has nothing to do with it - intention requires having an objective, which modern computers do not have.
For a more technical look at what specifically AIs are lacking when we talk about intention and objectives, I recommend this article by Yavar Bathaee that gets more into the computer science elements of intentionality.
That’s a definition. But the first definition that comes up for me is putting thoughts into words. So does AI produce thoughts, or just strings of words that appear like thoughts..
That's part of the point of the article. AI doesn't produce thoughts. It can't produce thoughts, because that's not something it's programmed to do. By your own definition, if it doesn't produce thoughts, it can't be writing.
There is a difference between how AI "writes" and how people do that is independent from training. AIs "write" by ingesting training data, then using that to predict the most likely next word. They're giant statistical engines. People write by ingesting data, then exercising intent to put the words together, understanding what each word means and what they want to do with it. AI has no understanding of what it's doing, while people do.
I’m sorry, but your understanding of modern LLMs is essentially reductive enough to be objectively wrong.
Finding a 23 year old live exploitable bug in the Linux kernel or generating solutions for long unsolved math problems isn’t being done by predictive text with extra steps.
I will fully grant that pattern recognition is something AI is pretty good at, and I'd argue that finding a bug in a Linux kernel is exactly that sort of pattern recognition. However, pigeons are also much better at pattern recognition (for certain patterns) than humans. Would you also therefore argue that a pigeon with a pen is the next Shakespeare?
If I ask an AI to write a sarcastic apology, avoid profanity, and make it two sentences, it is generating toward those goals. That’s a kind of functional intent.
That's your intent, not the AI's. Your intention is to write a sarcastic apology - it just uses its engines to generate something that matches your intent.
Think of it this way - if you did not give the LLM a prompt, what would it generate? Absolutely nothing. It would sit there, doing nothing, and not caring that it's not doing anything. Arguing that an AI generating an piece of text demonstrates intent is like arguing that a buzzsaw is intentional when it cuts a piece of wood. The saw is only doing what the human operating it tells it to do. Take away the human, and it does absolutely nothing.
The fact that I have written multiple comments on this post, including the submission statement, and chose to post the link at all are all more intent than an AI ever demonstrates. I'm not acting in response to you, but because I choose to do so.
Your method as stated doesn't have the power to distinguish human intelligence from machine intelligence, to be specific. There are a number of problems with it, I identified one. Others have pointed out more.
You haven't identified one. If you want to say that human intelligence is equivalent to artificial intelligence, you're going to need to either demonstrate that machines are self-considering, or that humans are not.
And I'm going to warn you that if your goal is to disprove human consciousness, I wish you the very best of luck.
Not sure why you keep banging on about consciousness, but we will put that aside.
You make a number of unfounded claims that apply just as well to AI as to human intelligence - you distinguish nothing. It's entirely possible (some experiments say "probable") that human intelligence uses statistical or even Bayesian inference.
I think directly comparing AI to humans often falls into false equivalency. And part of me thinks that is by design so that unethical behavior by AI companies can be excused through arbitrary anthropomorphizing of AI software ("humans learn from the world around them, therefore we have the right to use all of Humanity's intellectual property any way we want because AI 'learns'!") or to make it seem more capable than it is.
This article looks at the history of writing and literature to contextualise what AI "writing" is, or more specifically, isn't. It makes the argument that, without emotional resonance, writing isn't writing, and therefore, AI "writing" can't be considered writing.
While I agree with the overall statement that AIs can't "write," I grapple a bit with the idea that it's because of the lack of emotional resonance. AI "writing" doesn't have emotional resonance, nor does it change, but there is also a world of human writing that lacks those things as well. Prior to the rise of LLMs, listicles, clickbait, and bland writing still dominated the internet, and I'd argue those aren't particularly emotionally resonant either. What I think is worth considering is whether there's a line between "writing" and "literature" and whether there ought to be.
What utter rubbish -by this definition 99.999% of all writing is not "writing". It is also absolutely subjective as well; the author talks about "Plato and Aristotle", plenty of people would find their works to have “move souls”, but plenty of other people would not.
It's also not true to say that writing from AI is "without emotional resonance"; I'm sure if you published the output of AI someone would find something that moved their soul?
Penguin-Pete | 5 hours ago
Once again: having the machine do all the art for you defeats the entire purpose of art. Art communicates something; computers have nothing to say. You're just training a parrot to mimic the same pattern of language / drawing / singing we use. You can teach a parrot to say "compounded quarterly interest" but it isn't going to comprehend banking.
turtleturds | 4 hours ago
AI does comprehend banking. The reason being in order to explain compounded quarterly interest, it has inherently created a model of banking.
McCheesing | 3 hours ago
AI can regurgitate the pattern of banking *
turtleturds | an hour ago
what is the difference?
McCheesing | an hour ago
AI can copy your homework, but it can’t come up with anything on its own
adeadhead | 12 hours ago
If we use the definition as the sentient process that things are, AI can't do any single things that humans can. That's a failure of definition
Gullex | 7 hours ago
Sentient simply means "aware of an external environment". Insofar as AI can be "aware" at all, sensors feeding it data certainly meet that qualification.
adeadhead | 7 hours ago
AI as we currently have it isn't aware. It doesn't have a solid state. It takes an input and produces and output, that's it.
A motion sensor light isn't aware.
A more complicated sensor isn't either.
Gullex | 7 hours ago
There is no way to know if AI is aware or not, or when it becomes aware.
You have no way of proving you are aware.
>It doesn't have a solid state.
Neither do you.
adeadhead | 7 hours ago
We know that AI is not aware.
To suggest that it might possibly be is to fundamentally misunderstand how LLMs work.
Showy_Boneyard | 5 hours ago
Ultimately, we really don't know ANYTHING about awareness in the "hard problem of consciousness" sense. Other Minds has been a problem for as long as philosophy has. Some philosophers will even claim that their own consciousness isn't actually real, which almost makes me suspect philosophical zombies might be a real thing.
Gullex | 6 hours ago
There is no way to prove AI is not aware.
To suggest that it can't be is to fundamentally misunderstand your own awareness.
madmooseman | 5 hours ago
> There is no way to prove AI is not aware.
That’s not how the null hypothesis works.
[OP] Quouar | 12 hours ago
Why would that be a failure of definition rather than AI actually being unable to do these things?
adeadhead | 11 hours ago
It's a failure of definition because under these parameters, no matter what the process was, no AI even as described in science fiction would qualify as writing.
[OP] Quouar | 8 hours ago
I'm again going to ask, why is that a failure of definition?
caks | 8 hours ago
If you're going to accept that definition then you will also have to accept that you as a human cannot distinguish between writing and "non-writing".
SanityInAnarchy | 5 hours ago
That's a different problem. I, as a human who speaks English, would have a hard time distinguishing between spoken Chinese and spoken Korean. Obviously, it's possible for things to be wildly different, and for me to not be able to tell.
[OP] Quouar | 8 hours ago
I'll grant you that I can't necessarily distinguish between something written by, say, Data from Star Trek and a human, but Data is a fictional character. I can generally distinguish between ChatGPT and a human, especially when it's something other than rote putting words on a page.
caks | 4 hours ago
Nah, you can't. You think you can but that's a sampling bias. You're probably an unwitting consumer of a huge amount of AI generated text and images at this point. Sorry to break it to you!
[OP] Quouar | 4 hours ago
I'm literally a professional writer who works with people who are not as a day job. I promise you, people who interact with words on a regular basis absolutely can tell the difference.
Ok-Dog-7149 | 6 hours ago
I suspect within 5 to 10 years this will no longer be true… maybe sooner
flashmedallion | 6 hours ago
We've been "5 to 10 years away" since dalle-3
RexDraco | 8 hours ago
Sentience has no real place in our language today anyway. It is a magic word meaning something magical that doesn't exist. We never established a line in the sand, that means it is only good for politics and individual perception.
[OP] Quouar | 8 hours ago
I'm not sure why you say sentience doesn't exist. It does, and Freitas has an (admittedly controversial) formula for measuring it.
RexDraco | 8 hours ago
Sentience will always be based on human bias. It is like the illusion of random or choice, it might have meaning to the individual but it definitely doesn't exist.
If your brain is programmed to say you exist, you're magically sentient. We pretend AI isn't sentient when we ourselves are just biological computers, we think and function the way we do because we are programmed to. Fact of the matter is, the definition keeps changing along side the progression of understanding how more intelligent and aware other things are. We will just keep shifting goal posts.
[OP] Quouar | 8 hours ago
While I'll agree with you that sentience is based on human experience, that is a very far cry from saying it doesn't exist. There are a broad range of things that are based on human experience (and thus, human bias) that none the less exist. Would you argue art doesn't exist? Math? Music? So why sentience?
I'm not going to sit here and argue that humans aren't biological computers because we're ensouled or some nonsense like that, but there is an objective difference between what an AI does and what we do. We are able to intend, which is a critical difference between what an AI does and us. It's that ability to intend that's key to sentience, and it's an objective element of the definition of sentience.
RexDraco | 6 hours ago
What is "intend" to you exactly ? Intentions to complete a task with a chance of failure?
[OP] Quouar | 5 hours ago
"Intention" is the mental objective behind an action. It's what you envision accomplishing when you undertake an action. A computer has no ability to create an objective for itself, nor does it envision anything anything. What it does is based on its programming and how it was programmed by a designer. While I'll grant that we don't fully understand why an AI takes the actions it does, that doesn't change that it acts based on external intention, namely, the programmer or prompter or what have you. Failure has nothing to do with it - intention requires having an objective, which modern computers do not have.
For a more technical look at what specifically AIs are lacking when we talk about intention and objectives, I recommend this article by Yavar Bathaee that gets more into the computer science elements of intentionality.
Gullex | 7 hours ago
You're thinking of the word "sapience", not "sentience".
RexDraco | 6 hours ago
No, I'm not.
Gullex | 6 hours ago
Okay then you are quite confused.
Gullex | 7 hours ago
That is not remotely true.
Sentience is the ability to perceive, to be aware of an external environment. Of course that exists.
RexDraco | 6 hours ago
Either that doesn't exist or we have to accept how trivial that is.
Gullex | 6 hours ago
"We" is doing a lot of work there. You got a turd in your pocket?
manimal28 | 7 hours ago
No, that’s just the definition.
adeadhead | 7 hours ago
Writing: The act of a person or thing that writes.
manimal28 | 5 hours ago
That’s a definition. But the first definition that comes up for me is putting thoughts into words. So does AI produce thoughts, or just strings of words that appear like thoughts..
[OP] Quouar | 4 hours ago
That's part of the point of the article. AI doesn't produce thoughts. It can't produce thoughts, because that's not something it's programmed to do. By your own definition, if it doesn't produce thoughts, it can't be writing.
manimal28 | 4 hours ago
Yes I agree, but the guy I was responding to does not.
datums | 9 hours ago
How many people can write well that haven’t been trained on data from previous human writers?
[OP] Quouar | 8 hours ago
There is a difference between how AI "writes" and how people do that is independent from training. AIs "write" by ingesting training data, then using that to predict the most likely next word. They're giant statistical engines. People write by ingesting data, then exercising intent to put the words together, understanding what each word means and what they want to do with it. AI has no understanding of what it's doing, while people do.
Cidence | 8 hours ago
Have you never used a word you couldn’t fully define, but felt confident doing so because of how you’d seen/heard other people use it?
[OP] Quouar | 8 hours ago
The difference is in "felt confident." That shows I'm considering my words and using them with intent, something which an AI definitionally cannot do.
datums | 5 hours ago
I’m sorry, but your understanding of modern LLMs is essentially reductive enough to be objectively wrong.
Finding a 23 year old live exploitable bug in the Linux kernel or generating solutions for long unsolved math problems isn’t being done by predictive text with extra steps.
[OP] Quouar | 4 hours ago
I will fully grant that pattern recognition is something AI is pretty good at, and I'd argue that finding a bug in a Linux kernel is exactly that sort of pattern recognition. However, pigeons are also much better at pattern recognition (for certain patterns) than humans. Would you also therefore argue that a pigeon with a pen is the next Shakespeare?
Gullex | 7 hours ago
I think that line you're drawing between how AI writes and how humans do is really fuzzy.
AI uses intent, too. And the "understand" words in terms of how they relate to other words. Which, really is exactly how we understand them.
[OP] Quouar | 7 hours ago
In what way does AI use intent?
Gullex | 6 hours ago
If I ask an AI to write a sarcastic apology, avoid profanity, and make it two sentences, it is generating toward those goals. That’s a kind of functional intent.
[OP] Quouar | 6 hours ago
That's your intent, not the AI's. Your intention is to write a sarcastic apology - it just uses its engines to generate something that matches your intent.
Think of it this way - if you did not give the LLM a prompt, what would it generate? Absolutely nothing. It would sit there, doing nothing, and not caring that it's not doing anything. Arguing that an AI generating an piece of text demonstrates intent is like arguing that a buzzsaw is intentional when it cuts a piece of wood. The saw is only doing what the human operating it tells it to do. Take away the human, and it does absolutely nothing.
Gullex | 6 hours ago
Think of it this way. If I hadn't written what I did above, what would you have replied with?
The difference between AI and human cognition isn't all that different.
[OP] Quouar | 5 hours ago
The fact that I have written multiple comments on this post, including the submission statement, and chose to post the link at all are all more intent than an AI ever demonstrates. I'm not acting in response to you, but because I choose to do so.
Vryl | 4 hours ago
You can't actually prove that. Maybe we are statistical machines. Humans use a lot of heuristics.
[OP] Quouar | 4 hours ago
You're arguing that we can't prove we're conscious beings? Just to be clear, that's what you want to argue here?
Vryl | 3 hours ago
Your method as stated doesn't have the power to distinguish human intelligence from machine intelligence, to be specific. There are a number of problems with it, I identified one. Others have pointed out more.
[OP] Quouar | 2 hours ago
You haven't identified one. If you want to say that human intelligence is equivalent to artificial intelligence, you're going to need to either demonstrate that machines are self-considering, or that humans are not.
And I'm going to warn you that if your goal is to disprove human consciousness, I wish you the very best of luck.
Vryl | 33 minutes ago
Not sure why you keep banging on about consciousness, but we will put that aside.
You make a number of unfounded claims that apply just as well to AI as to human intelligence - you distinguish nothing. It's entirely possible (some experiments say "probable") that human intelligence uses statistical or even Bayesian inference.
[OP] Quouar | 7 minutes ago
Can you cite your sources?
Dreadsin | 6 hours ago
The main difference is that machines are told what is a desirable outcome to achieve and humans, not as much so
roamzero | an hour ago
I think directly comparing AI to humans often falls into false equivalency. And part of me thinks that is by design so that unethical behavior by AI companies can be excused through arbitrary anthropomorphizing of AI software ("humans learn from the world around them, therefore we have the right to use all of Humanity's intellectual property any way we want because AI 'learns'!") or to make it seem more capable than it is.
[OP] Quouar | 12 hours ago
This article looks at the history of writing and literature to contextualise what AI "writing" is, or more specifically, isn't. It makes the argument that, without emotional resonance, writing isn't writing, and therefore, AI "writing" can't be considered writing.
While I agree with the overall statement that AIs can't "write," I grapple a bit with the idea that it's because of the lack of emotional resonance. AI "writing" doesn't have emotional resonance, nor does it change, but there is also a world of human writing that lacks those things as well. Prior to the rise of LLMs, listicles, clickbait, and bland writing still dominated the internet, and I'd argue those aren't particularly emotionally resonant either. What I think is worth considering is whether there's a line between "writing" and "literature" and whether there ought to be.
SirCliveWolfe | 9 hours ago
> emotional resonance
What utter rubbish -by this definition 99.999% of all writing is not "writing". It is also absolutely subjective as well; the author talks about "Plato and Aristotle", plenty of people would find their works to have “move souls”, but plenty of other people would not.
It's also not true to say that writing from AI is "without emotional resonance"; I'm sure if you published the output of AI someone would find something that moved their soul?
alterfero | 9 hours ago
Is writing really thinking? Many Neanderthals say no.