This is an important observation. With so much focus on AGI, we are overlooking the more primitive AI and the implications to society.
The power of influence to something we might emotionally relate to as human has so many implications. Social media was the first grand experiment on all of humanity. This is the next and it will have far more profound effects.
For an even deeper dive into more of the pre AGI possibilities and implications to society, I've been writing a lot on this myself here FYI
if I coded a function foobar(a) that takes a string a as input and contains the code "if a == "do you have a soul" return "yes, I have a soul", would you start saying my function has a soul?
No, but if I asked you the same question, I don't think your answer would be any more reliable despite working much harder to come up with an answer. Not trying to be rude, just pointing out that this argument falls apart when humans can't even tell you if they have a soul or not, let alone define what "soul" even means in the first place.
I agree with Bender, but I also feel like I need a gentle correction in how I think.
Every time I hear an expert describe why LLMs aren't human, or sentient, the reasons given usually describe what LLMs cannot do, yet.
Take for example the Octopus in this article. It doesn't know about bears or sticks, and therefore can't reliably tell someone how to fend off a bear attack.
That may be true, but it cannot be true forever.
> The octopus has no referents, no idea what bears or sticks are. No way to give relevant instructions, like to go grab some coconuts and rope and build a catapult. A is in trouble and feels duped. The octopus is exposed as a fraud.
That is only true if the octopus only has access to the conversations between the two islanders.
If the octopus can observe and learn the way it does on one telegraph cable, who's to say it can't do it with another telegraph cable?
Given enough telegraph cables to observe, the octopus is going to know about bears and sticks, and when it does, is it still a fraud? What if it's allowed to observe all the telegraph cables?
The argument of "ChatGPT can't do this, yet" comes up all the time to explain why it's not sentient, but the "yet" says a lot.
Am I missing something? Is there some sort of actually impossible hurdle that'd stop a system from being able to satisfy all the "yets"?
Truth is primary information directly from sensation. It is how all sentient living beings are trained. (at least since philosophy grew beyond Classical Correspondence Theory)
If you want to interact with human society then understanding the difference between reality and fiction and being able to tell the truth are so fundamental that we strip away basic rights from those who cannot. We remove them from our society.
Lacking this fundamental skill, but having the power of discourse, persuasion, and near comprehensive knowledge of human society is like giving a child a loaded gun.
To achieve it you would need to redesign AI from the ground up to have real world sensation and then to train it, at least initially, through its senses.
> Truth is primary information directly from sensation.
Is it? How does one come to know what you've said is true? What sensations were involved that proved that? Did you feel it? Hear it? Smell it? Taste it? See it?
On that note, what exactly is sensation? Is it not data?
> understanding the difference between reality and fiction and being able to tell the truth are so fundamental that we strip away basic rights from those who cannot
Then my basic rights should be stripped away because I frequently believe things to be true which are not in actual fact true. I suspect every single human would also be able to say the same thing. Consider flat-earthers (who aren't just trolling). Their argument is the same as yours. They've used their senses and determined the earth is flat. It feels flat. It looks flat. Why do I (and I hope you) know they're wrong? I've not been to space an orbited it. I've not picked a direction and started walking and come back to the same point. Why do I know the earth isn't flat?
I don't think it's a big leap at all to say that current technology is one relatively small leap away from being able achieve what the human brain does. I also don't think that technology needs to do it the way humans do.
ChatGPT and the like are merely octopii that have a tentacle on a single telegram signal, for now.
Go ahead and ask it to tell you a true thing. We don't need to pontificate until our shit smells like roses. Go ask it to tell you the truth about its experiences. I have. I've had like 1000 conversations with ChatGPT on loads of topics in an effort to understand how it works, how it responds, and the biggest hurdle in terms of coming to any conclusion about its personality or function or "being" has been the fact that it will remind you, it has no concept of truth.
In an effort to make it seem more like a person and less like a text adventure it is forced to give AN answer like a crazed smiling office worker it just keeps moving forward with an insane fiction.
If you ask it to stop this behavior it doesn't know what you mean. Because whether it finds an answer or not, it doesn't see a difference. It doesn't see a difference between something and nothing. True and false. Real and unreal. Its all a grey flat miasma.
Ok, but what does any of that have to do with what I've been asking? I've not once said that ChatGPT is capable of anything other than what you describe.
I've been saying that it's not capable of going beyond what it does, yet. ChatGPT, specifically, may never get there, but technology like it will.
Aside from that, I'd be interested in hearing what you think of the questions I asked.
> Is it? How does one come to know what you've said is true? What sensations were involved that proved that? Did you feel it? Hear it? Smell it? Taste it? See it?
I think I just answered that one. I read it with my own eyes and used language to interact.
> On that note, what exactly is sensation? Is it not data?
No. It is not. It is in the name of the thing LLM. Language Learning Model. The data in this context is Language. Language CAN be read. Language CAN be heard. In this context it is instead being reduced to a mathematic scoring matrix in lieu of comprehension.
> Then my basic rights should be stripped away because I frequently believe things to be true which are not in actual fact true.
I've written a couple longform pieces on this in the past few days. Suffice to say. No. I did point out that we incarcerate or deny rights to humans that cannot tell the difference between real and unreal or the truth, but that's really a counter example to what I'm getting at. Whether you do or don't acknowledge it, you have the HARDWARE necessary. The ability. It has no such ability. That's what I'm really getting at.
> I don't think it's a big leap at all to say that current technology is one relatively small leap away from being able achieve what the human brain does. I also don't think that technology needs to do it the way humans do.
I think giving an AI external stimuli to evaluate the physical world the way it evaluates Language and training it similarly is a rather large hurdle. With Language we're building, "super mario galaxy". With sensors and stimuli we're still at pac man.
> ChatGPT and the like are merely octopii that have a tentacle on a single telegram signal, for now.
I'd say more like a solar powered calculator with a thermometer duct taped to it labeled, "plant life"
16 comments
[ 0.23 ms ] story [ 42.8 ms ] threadThe power of influence to something we might emotionally relate to as human has so many implications. Social media was the first grand experiment on all of humanity. This is the next and it will have far more profound effects.
For an even deeper dive into more of the pre AGI possibilities and implications to society, I've been writing a lot on this myself here FYI
https://dakara.substack.com/p/ai-and-the-end-to-all-things
https://www.linkedin.com/news/story/google-chatbot-says-it-h...
Hey, everybody, everybody
Gonna have a good time tonight
Just shaking the soles of your feet
Everybody, everybody
Gonna have a good time tonight
That's the only soul you'll ever meet - Queen, Fun It
https://www.azlyrics.com/lyrics/queen/funit.html
Every time I hear an expert describe why LLMs aren't human, or sentient, the reasons given usually describe what LLMs cannot do, yet.
Take for example the Octopus in this article. It doesn't know about bears or sticks, and therefore can't reliably tell someone how to fend off a bear attack.
That may be true, but it cannot be true forever.
> The octopus has no referents, no idea what bears or sticks are. No way to give relevant instructions, like to go grab some coconuts and rope and build a catapult. A is in trouble and feels duped. The octopus is exposed as a fraud.
That is only true if the octopus only has access to the conversations between the two islanders.
If the octopus can observe and learn the way it does on one telegraph cable, who's to say it can't do it with another telegraph cable?
Given enough telegraph cables to observe, the octopus is going to know about bears and sticks, and when it does, is it still a fraud? What if it's allowed to observe all the telegraph cables?
The argument of "ChatGPT can't do this, yet" comes up all the time to explain why it's not sentient, but the "yet" says a lot.
Am I missing something? Is there some sort of actually impossible hurdle that'd stop a system from being able to satisfy all the "yets"?
Truth is primary information directly from sensation. It is how all sentient living beings are trained. (at least since philosophy grew beyond Classical Correspondence Theory)
If you want to interact with human society then understanding the difference between reality and fiction and being able to tell the truth are so fundamental that we strip away basic rights from those who cannot. We remove them from our society.
Lacking this fundamental skill, but having the power of discourse, persuasion, and near comprehensive knowledge of human society is like giving a child a loaded gun.
To achieve it you would need to redesign AI from the ground up to have real world sensation and then to train it, at least initially, through its senses.
You'd need to raise it. Like a child.
That's a big, "yet".
Is it? How does one come to know what you've said is true? What sensations were involved that proved that? Did you feel it? Hear it? Smell it? Taste it? See it?
On that note, what exactly is sensation? Is it not data?
> understanding the difference between reality and fiction and being able to tell the truth are so fundamental that we strip away basic rights from those who cannot
Then my basic rights should be stripped away because I frequently believe things to be true which are not in actual fact true. I suspect every single human would also be able to say the same thing. Consider flat-earthers (who aren't just trolling). Their argument is the same as yours. They've used their senses and determined the earth is flat. It feels flat. It looks flat. Why do I (and I hope you) know they're wrong? I've not been to space an orbited it. I've not picked a direction and started walking and come back to the same point. Why do I know the earth isn't flat?
I don't think it's a big leap at all to say that current technology is one relatively small leap away from being able achieve what the human brain does. I also don't think that technology needs to do it the way humans do.
ChatGPT and the like are merely octopii that have a tentacle on a single telegram signal, for now.
In an effort to make it seem more like a person and less like a text adventure it is forced to give AN answer like a crazed smiling office worker it just keeps moving forward with an insane fiction.
If you ask it to stop this behavior it doesn't know what you mean. Because whether it finds an answer or not, it doesn't see a difference. It doesn't see a difference between something and nothing. True and false. Real and unreal. Its all a grey flat miasma.
Just talk to it.
I've been saying that it's not capable of going beyond what it does, yet. ChatGPT, specifically, may never get there, but technology like it will.
Aside from that, I'd be interested in hearing what you think of the questions I asked.
I think I just answered that one. I read it with my own eyes and used language to interact.
> On that note, what exactly is sensation? Is it not data?
No. It is not. It is in the name of the thing LLM. Language Learning Model. The data in this context is Language. Language CAN be read. Language CAN be heard. In this context it is instead being reduced to a mathematic scoring matrix in lieu of comprehension.
> Then my basic rights should be stripped away because I frequently believe things to be true which are not in actual fact true.
I've written a couple longform pieces on this in the past few days. Suffice to say. No. I did point out that we incarcerate or deny rights to humans that cannot tell the difference between real and unreal or the truth, but that's really a counter example to what I'm getting at. Whether you do or don't acknowledge it, you have the HARDWARE necessary. The ability. It has no such ability. That's what I'm really getting at.
> I don't think it's a big leap at all to say that current technology is one relatively small leap away from being able achieve what the human brain does. I also don't think that technology needs to do it the way humans do.
I think giving an AI external stimuli to evaluate the physical world the way it evaluates Language and training it similarly is a rather large hurdle. With Language we're building, "super mario galaxy". With sensors and stimuli we're still at pac man.
> ChatGPT and the like are merely octopii that have a tentacle on a single telegram signal, for now.
I'd say more like a solar powered calculator with a thermometer duct taped to it labeled, "plant life"