Comments

You must log in or register to comment.

sticky_symbols t1_j32czsi wrote

Chatgpt is misrepresenting its own capabilities. It cannot reflect on its own thinking or self improve.

225

SeaBearsFoam t1_j32ls6k wrote

I also don't think it has sensors and input channels through which it receives information about the world.

120

sticky_symbols t1_j32rgwg wrote

True.

It also occurred to me, though, that this might actually be what a high school teacher would say about chatGPT. They might get it this wrong.

61

overlordpotatoe t1_j336wqa wrote

I wonder if assigning the person who's doing the explaining in this situation to be someone with no special knowledge or expertise in the field makes it more likely to get things like this wrong.

24

bactchan t1_j35u5c4 wrote

I think the likelihood of the ChatGPT emulating both the style and accuracy of a highschool teacher is a bit beyond scope.

5

FidgitForgotHisL-P t1_j349sze wrote

It seems very likely the basis for this reply literally comes from people writing their own opinions on this question. Given the gulf between what it’s actually doing and what someone would assume is happening if they have no interest in AI etc, you could see them matter-of-factly asserting these points, which in turn means chatGPT will.

10

Technologenesis t1_j33wdnr wrote

I think that would depend on how loosely you define a sensor. Somehow, your messages are getting to ChatGPT for it to process them. This could be considered a kind of sensation.

9

kala-umba t1_j3394ef wrote

It said to me that it doesn't have them around Christmas! Maybe something changed xD And besides it is not able to start communication! So it only does something when asked! It's not an "individual" agent capable of maneuvering in this plane of existence

5

2Punx2Furious t1_j34xuif wrote

It does, it's text that you give it as a prompt. That is an input, or in other words, a sensor. It is kind of limited, but sufficient to consider it "aware" of something.

5

Solobolt t1_j3575db wrote

I would say that it may not be aware that it does not. Because learning is disabled when talking to it, it would be as if someone has short term memory loss. To its own experience it would be experiencing new things in its own real time. Sometimes including its own interactions. So it may 'think' it experiences things like that. Shadows on cave walls and whatnot.

1

overlordpotatoe t1_j336iiu wrote

Yup. It once told me it couldn't make text bold even though it had just been doing that. Do not trust the AI to self report about itself, ever, but especially if you ask it to roleplay because then it seems to switch off whatever controls it has that makes it at least try to be factual and it starts spitting out pure fiction.

12

kriskoeh t1_j33besz wrote

It has told me several times that it can’t think for itself lol.

8

LarsPensjo t1_j33hla5 wrote

I asked ChatGPT "Are you sentient?", and got the (expected) answer

> I am a machine learning model trained to simulate conversation, so I don't have consciousness or sentience in the same way that humans do.

Then I asked, "Please explain why you say that.", and got the answer

> As a machine learning model, I have been programmed to process and generate text based on patterns in data...

Doesn't that mean it reflected on its own thinking?

6

sticky_symbols t1_j33icmp wrote

No, it really doesn't. That's you asking it about it's thinking and it responding. That's different than reflection, which would consist of it asking itself anything about its thinking.

11

LarsPensjo t1_j33nt4v wrote

One definition of "reflection" is

> serious thought or consideration.

Can you give an example where a human person can reflect on something, which ChatGPT can't? And more crucially, what method would you use to detect this?

What I am aiming at, is that these are borderline philosophical questions, without clear definitions.

2

gleamingthenewb t1_j36162g wrote

Any example of a human reflecting is an example of something ChatGPT can't do, because all it can do is predict what word comes next based on its training data and the prompt you give it.

3

sticky_symbols t1_j37q6jw wrote

They are not. They are questions about the flow of information in a system. Humans recirculate information in the process we call thinking or considering. ChatGPT does not.

1

Technologenesis t1_j33y3xa wrote

It doesn't seem to be reflecting on its own thinking so much as reflecting on the output it's already generated. It can't look back at its prior thought process and go "D'oh! What was I thinking?", but it can look at the program it just printed out and find the error.

But it seems to me that ChatGPT is just not built to look at its own "cognition", that information just shouldn't be available to it. Each time it answers a prompt it's just reading back through the entire conversation as though it's never seen it before; it doesn't remember generating the responses in the first place.

I can't be sure but I would guess that the only way ChatGPT even knows that it's an AI called ChatGPT in the first place is because OpenAI explicitly built that knowledge into it*. It doesn't know about its own capabilities by performing self-reflection - it's either got that information built into it, or it's inferring it from its training data (for instance, it may have learned general facts about language models from its training data and would know, given that it itself is a language model, that those facts would apply to itself).

*EDIT: I looked it up and in fact the reason it knows these things about itself is because it was fine-tuned using sample conversations in which human beings roleplayed as an AI assistant. So it doesn't know what it is through self-reflection; it essentially knows it because it learned by example how to identify itself)

3

monsieurpooh t1_j35wbej wrote

It's probably going to devolve into a semantics debate.

ChatGPT model neurons stay the same until they retrain it and release a new version.

But, you feed it back its own output + more prompt, and now it has extra context about the ongoing conversation.

For now I would have to say it shouldn't be described as "reflecting on its own thinking", since each turn is independent from others and it's simply trying to predict what would've been reasonable to appear in text. For example: It could be an interview in a magazine, etc.

That being said... I'm a big fan of the saying that AI doesn't need human-brain-style thinking to achieve a working imitation of human-level intelligence, just like the airplane is an example of flying without imitating the bird.

2

LarsPensjo t1_j35yrqx wrote

> That being said... I'm a big fan of the saying that AI doesn't need human-brain-style thinking to achieve a working imitation of human-level intelligence, just like the airplane is an example of flying without imitating the bird.

I definitely agree. IMO, you see a lot of "AI is not true intelligence", which doesn't really matter.

Eliezer Yudkowsky had an interesting observation:

> Words aren't thoughts, they're log files generated by thoughts.

I believe he meant the written word.

2

gleamingthenewb t1_j3605ep wrote

Nope, that's just its prediction of what string of characters corresponds to your prompt.

1

LarsPensjo t1_j36arit wrote

Ok. Is there anything you can ask me, where the answer can't be explained as me just using a prediction of a string of characters corresponds to your prompt?

1

gleamingthenewb t1_j36ht6y wrote

That's a red herring, because your ability to generate text without being prompted proves that you don't just predict strings of characters in response to prompts. But I'll be a good sport. I could ask you any personal question that has a unique answer of which you have perfect knowledge: What's your mother's maiden name? What's your checking account number? Etc.

1

LarsPensjo t1_j36mdv5 wrote

But that doesn't help to determine whether it uses reflection on its own thinking.

1

gleamingthenewb t1_j36srlq wrote

You asked for an example of a question that can't be answered by next-word prediction.

1

FusionRocketsPlease t1_j36r2pm wrote

Try asking an unusual question that one would have to reason with to answer. He will miss. This has already been shown in this same sub today.

1

bubster15 t1_j33ws6u wrote

I mean I’d argue it can do both. Just a like a human, we understand when we did something wrong based on the outward feedback of our decision.

ChatGPT is taught to to act the way it does but isnt that exactly how humans work? Learn what we are taught as we develop and slowly adapt our behavior based on real world feedback to meet our desired goals?

3

visarga t1_j34sua5 wrote

Yes, that is why language models with feedback are much more powerful than isolated ones.

2

sticky_symbols t1_j37qowh wrote

If you're arguing it can do both, you simply don't understand how the system works. You can read about it if you like.

2

LarsPensjo t1_j33enxl wrote

I saw an example where someone asked for a Python program to solve a task. ChatGPT produced such a program. But there was an error, and the person pointed out the error and asked for a fix.

ChatGPT then produced a correct program.

Isn't this an example of self-improvement? There was external input, but that is beside the point. Also, the improvement is going to be forgotten if you restart with a new prompt. But that is also beside the point, there was an improvement while the sessions lasted.

Notice also that ChatGPT did the improvement, the person starting the prompt did not explicitly how to solve the error.

2

micaroma t1_j3469ny wrote

"But that is also beside the point, there was an improvement while the sessions lasted."

Really? That seems like the most important factor of "self-improvement". If it only improved its error in the session but makes the same error if you refresh the page, then it didn't improve itself, it simply improved its output. There's a huge difference between permanently upgrading your own capabilities from external input, and simply fixing text already written on the page with external input.

(Also, it sometimes continues to make the same error within the same session even after pointing out its mistake, which is greater evidence against true self-improvement.)

6

visarga t1_j34sgk5 wrote

I don't see the problem. The language model can have feedback from code execution. If it is about facts, it could have access to a search engine. But the end effect is that it will be much more correct. A search engine provides grounding and has fresh data. As long as you can fit the data/code execution results in the prompt, all is ok.

But if we save the correctly executed tasks and problems we could make a new dataset to be used in fine-tuning the model. So it could learn as well.

2

sticky_symbols t1_j33ih3f wrote

That's improvement but definitely not self improvement since a human had to ask.

3

LarsPensjo t1_j33luce wrote

Aren't all self-improvements ultimately triggered by external events?

15

magnets-are-magic t1_j34qojy wrote

And aren’t humans taught how to function in society? It takes decades or mentorship from parents, school, friends. And we continue to constantly learn for our entire lives

4

eroggen t1_j355ace wrote

Yes but ChatGPT doesn't have the ability to initiate the process of synthesizing external input. It can hold the conversation in "short term memory" but it can't ask questions or experiment.

2

sticky_symbols t1_j37pynh wrote

Ultimately, yes. But humans can make many steps of thinking and self Improvement after that external event. Chatgpt is impacted by the event but simply does not think or reflect on its own to make further improvements.

1

visarga t1_j34s0ku wrote

Like, you can put a Python REPL inside chatGPT so it can see the error messages. And allow it a number of fixing rounds.

2

visarga t1_j34roas wrote

Yes it can, but only on what it has in the conversation history. Each conversation starts tabula rasa. For example all the behaviour rules are meta, thinking about thinking.

1

sticky_symbols t1_j37jo6w wrote

It does not do what people call reflection even with that chat history. And it's improved slightly by having more relevant input, but I wouldn't call that self improvement.

1

Darkhorseman81 t1_j36ecgi wrote

Wait until you see chatGTP4. It's reflecting and growing as we speak. You've only seen a drop in the ocean of what it's capable of.

1

sticky_symbols t1_j38f635 wrote

CPT4 will be even better, but it also does not reflect or self-improvement unless they've added those functions.

1

lord_ma1cifer t1_j3gnjk5 wrote

It can in a manner of speaking. It depends on how you apply the definition, it can look back at its previous functions and see a record of the past then make decisions about the future based on that information. As for self improvement, what is AI but software constantly improving itself, so what if it requires our help to do so. Some people use gurus and life coaches after all.

I agree in this case it's a fair bit misleading. It also barely fits the description so it's really only a matter of semantics as I fully agree it's not doing these things in the way that it's response implies.

1

FederalScientist6876 t1_j3ka4ps wrote

It can. When we humans reflect and self improve, at the raw level, there’s a lot of computation happening in the brain. This leads to improvement. ChatGPT has different kinds of computations doing something similar to me

1

sticky_symbols t1_j3l2yzd wrote

It COULD do something similar, but it currently does not. You can read about it if you want to know how it works.

Similar systems might reflect and self improve soon. That will be exciting and terrifying.

1

FederalScientist6876 t1_j3l73vr wrote

It is collecting feedback from user data and improving itself. Just that it isn’t doing online learning (in real time just after it receives the feedback). Online or batch, it still is improving itself by reflecting (learning from) on the massive amounts of feedback it has collected from its millions of users. It isn’t developing its underlying algorithms, training architectures, etc. (which is also feasible to do). But that even humans can’t do. That’d be more akin to humans being able to evolve themselves into more intelligent beings by modifying the brain structure, size, or neuron function, rather than mere self improvement based on reflection of past experiences. The latter sounds like what any AI system already seems to do to me. Whether it is self aware or not like humans, I don’t know. It can convince you that it is self aware, at which point there’d be no way to prove that it isn’t or it is.

1

sticky_symbols t1_j3m9zwl wrote

It is not. It doesn't learn from its interactions with humans. At all.

That data might be used by humans to make a new version that's improved. But that will be done by humans.

It is not self aware in the way humans are.

These are known facts. Everyone who knows how the system works would agree with all of this. The one guy that argued LAMDA was self aware just had a really broad definition.

1

FederalScientist6876 t1_j3o3vsw wrote

No. Humans will feed the new data into the system/neural network. Humans will not use the data to improve the version. The learning will be done on its own, based on the human feedback (thumbs up or thumbs down) on the interactions it had. The network will update its weights parameters to optimize for higher probability of thumbs up. Just like humans optimize for thumbs up and positive feedback from the interactions we have.

1

turnip_burrito t1_j32ciki wrote

ChatGPT just contradicted itself.

It defined consciousness using subjective experience, and then turned around and said it is not clear if subjective experience is necessary for consciousness.

If you look closely at what it actually says, you will sometimes find absurd contradictions like this.

78

ProShortKingAction t1_j32rvyt wrote

It's not in a discord debate with philosophy nerds, it was told to represent itself as a high school teacher which is why it makes a lot of sense for it to put forward multiple possible ideas and help the student figure it out for themselves instead of giving one definitive answer on an open ended question

17

elfballs t1_j35w15z wrote

But it doesn't say "another possibility is...", instead it clearly had an inconsistent position, or rather no position at all. It's just words.

1

PIPPIPPIPPIPPIP555 t1_j33s94d wrote

But Consciousness IS Subjective Experience and there is no Philosopher that is saying that the conciousness is not subjective Experience because that is litteraly what it is

0

ajahiljaasillalla t1_j340kpd wrote

Consciousness could be a feature of universe, like some kind of field. So the brain could stand in the same relation to consciousness as radio to radio waves.

2

Revolutionary_Soft42 t1_j35kyd8 wrote

Yeah the resonance theory of consciousness , our human meat suits are just antennas orchestrating our consciousness

1

eve_of_distraction t1_j34av6f wrote

Oh yes there are. There are plenty of catastrophically confused philosophers when it comes to this. Have you read any Dan Dennett?

2

williamfwm t1_j350zmd wrote

That's because Dan Dennett is a p-zombie. He's never experienced consciousness, so he can't fathom what it is. Same goes for a number of other eliminative materialists such as the Churchlands, Graziano, Blackmore, etc

Interestingly, Richard Dawkins the mega-reductionist-Uber-atheist is not one, and neither is Kurzweil, who believes in computationalism (functionalism); you'd be hard pressed to find it in his books, but he slipped and all but admitted that consciousness is something that transcends reductionism in a reply he wrote to Jaron Lanier's One Half A Manifesto in the early 2000s

It would help the discussion if we could steal the terminology back, because it's been zombified by Dennett (continuing what his mentor Ryle started) and his ilk. I think we ought to distinguish "Dennettian Consciousness" (where 'consciousness' is just a convenient, abstract label for the bag of tricks the brain can perform) and "Chalmerian Consciousness" (the real kind of consciousness, the reduction-transcending-ineffable, for people who believe in the Hard Problem)

4

PIPPIPPIPPIPPIP555 t1_j366ifj wrote

Dan Dennett does not deny Subjective Experience he is saying that Qualia (The subjective Experience) is NOT incorrigible, ineffable, private nor directly accessible and this does not mean that Subjective Experience does not exist it does only imply that the processes that Creates the Experience is Physical processes in the brain

1

eve_of_distraction t1_j3676h2 wrote

He said consciousness is an illusion though.

1

FusionRocketsPlease t1_j36ke9e wrote

Yup. Now it remains to know how rays qualitative characteristics arise from quantities without these characteristics. If you talk about emergency, you'll get a block.

1

Technologenesis t1_j33xi6n wrote

Lots of philosophers have differing definitions of consciousness. I ultimately agree that subjective experience is a necessary component, but there are others who want to construe consciousness in more concrete physical / functional terms.

1

wtfcommittee OP t1_j32han1 wrote

That’s not really a contradiction. It provided the dictionary definition of consciousness, and then suggested that this definition can be argued against since a few philosophers mentioned subjective experiences aren’t necessary.

It’s not saying emphatically that subjective experience is unnecessary for consciousness. Just questioning the definition.

14

bubster15 t1_j33tsg7 wrote

I like this answer personally. How does one even define consciousness? Subjective thought? I personally don’t think so but the who hell knows?

Subjective thought is Darwinism at work. We need to perceive only what helps us survive and pass our genetics, trying to get the full picture of the world is a futile cause. Evolution said screw that, let’s just smooth over the real stuff we can’t actively comprehend.

We can’t fully grasp the world around us, not even close, so what makes us more conscious than a dog? We can perceive marginally more but neither species comes even remotely close to full worldly perception.

2

eve_of_distraction t1_j34bgz0 wrote

> personally don’t think so but the who hell knows?

You don't think consciousness is defined by subjectivity?

>so what makes us more conscious than a dog?

Do you think dogs have subjective experience? It seems absurd to think they don't.

2

turnip_burrito t1_j32kbzp wrote

Nope. It's a contradiction.

The moment you define a word to mean one thing, you are no longer searching for its meaning. You have found it. You have defined it.

−19

wtfcommittee OP t1_j32l9r1 wrote

But the definition of consciousness itself is subjective. There is no fixed answer.

13

turnip_burrito t1_j32ob9d wrote

It is simple. What the definition of "consciousness" should be to begin with is subjective. But once you define something, you should not contradict it unless you explicitly change which real world things it is describing.

How silly would it be for me to define dogs as "canines kept as pets" and then later say "well, dogs don't have to be canines"? That's what has happened with ChatGPT.

−11

Feebleminded10 t1_j3346ky wrote

That makes no sense bro how the ai clearly explains its point and logic but you still don’t get it? At the end the person said reply how a teacher would too.

2

turnip_burrito t1_j3378fq wrote

It clearly explains it in a way that is incorrect.

−3

mulder0990 t1_j33hto1 wrote

Is a square a rectangle?

Things can be defined and have nuanced change based on new knowledge all while keeping its original definition.

1

turnip_burrito t1_j33mwbc wrote

Not if the definition is "B is A" and the nuanced change is "B is not A". Try making those coexist together. You can't. lol

I can't believe people are defending this. It's a chatbot. And it contradicted itself. It does this often if you play with it.

1

bubster15 t1_j33qj5q wrote

I don’t see how it contradicted itself at all.

Ultimately it argues consciousness is tricky and not well understood scientifically, thus, it highlights some thought experiments where we guess some of the rules to “consciousness” and see if it makes sense to our understanding.

Consciousness is impossible to define even for humans. We have no idea what ultimately makes us conscious but it uses subjective thought as an example of what we generally are looking for in “consciousness”

3

turnip_burrito t1_j33unnx wrote

If consciousness isn't the subjective experience, and it isn't the externally observable physical system, then there is nothing left for it to be, except perhaps a third thing like a "soul". It is logically impossible for a word to sensibly mean anything except one or a subset of those three things. Consciousness, to mean anything, would be one of those or some aspect of those. The cause or process of it is not determinable now, but the thing itself can be put to words easily.

If something is impossible to define, then the word should not be used in a serious discussion.

2

eve_of_distraction t1_j34apbl wrote

Some people including famous philosophers are deeply confused about this. Look at Dan Dennett for example, claiming consciousness is an illusion. It stems from an alarming lack of introspection.

2

2Punx2Furious t1_j34y066 wrote

> you will sometimes find absurd contradictions like this

Quite often actually. Still, it's impressive that it still manages to sound plausible and eloquent enough.

2

visarga t1_j34sywj wrote

How do you define consciousness? I define it as "understanding". Ability to reason in new situations. The witness, or the judge that decides the actions we are going to take.

1

turnip_burrito t1_j34xhjf wrote

If I have to use the word, I use it to mean just "qualia", but I usually don't use the word "consciousness" because other words are more productive imo.

I usually don't care what it means in a conversation, as long as it is clearly defined. The biggest problem with the word, imo, is different people have different definitions for it (ok as long as clarified), which causes long drawn out pseudo-deep conversations where two people are using the same word differently.

So your definition is fine (and interesting), thanks for sharing!

1

FederalScientist6876 t1_j3kae4p wrote

No it didn’t. The terms like consciousness are not well defined as of today. It is giving multiple perspectives ways to define and understand what consciousness is and isn’t

0

joseph20606 t1_j32b44c wrote

We can’t even be sure the person next to you is actually having a subjective experience of consciousness. So I am siding with GPT on this one xD

32

ProShortKingAction t1_j32sn33 wrote

Yeah I also feel like we are reaching a point where moving the goal post of what is consciousness any farther gets into some pretty dark territory. Sure an AI expert can probably explain that ChatGPT isn't conscious but it's going to be in a way that no one outside of the field is going to understand at this point. I keep on seeing takes like "oh it isn't conscious because it doesn't keep its train of thought in-between conversations"... OK so your grandfather with dementia isn't conscious? Is that really a point these people want to make?

I feel like it's getting close enough that we should stop putting so much weight into the idea of what is and isn't conscious before we start moving this goal post past what a lot of human beings can compete with.

16

ReignOfKaos t1_j32tesd wrote

The only reason we think other people are conscious is via analogy to ourselves. And the only reason we think animals are conscious is via analogy to humans. The truth is we really have no idea what consciousness is nor do we have a way to measure it, so any ethics based on a rigid definition of consciousness is very fallible.

13

Ortus14 t1_j34hehj wrote

I like ethics based on uncertainty. We don't know who is or isn't conscious but it's safer to act as if more entities are conscious to not hurt others.

3

Noname_FTW t1_j33bbbu wrote

Additionally: While we know how these AI's work on a technical level and can therefore explain their behavior it doesn't mean that this isn't consciousness. People tend differentiate this between the human experience because we do not yet have the intricate understanding on how exactly the brain works. We know a lot about it but not to the degree we could recreate it.

The solution should be that we will develop an "intelligence level chart" so to speak. I heard this as a reference somewhere in some scifi material but can't remember where.

The point would be that we are going to start develop a system with levels in which we classifiy AI's and biological beings in terms of their intelligence.

It would look similar on how we classify autonomous vehicles today with level 5 being entirely autonomous.

The chart could go from 0-10 where humans would be somewhere around 5-8. 10 being a super AGI and viruses being 0.

Each level would be assigned properties that are associated with more intelligence.

Having this system it would help to assign rights to species of each category.

Obviously it would have to be under constant scrutiny to be as accurate and objective as it can be.

6

ProShortKingAction t1_j33gery wrote

Honestly it sounds like it'd lead to some nazi shit, intelligence is incredibly subjective by its very nature and if you ask a thousand people what intelligence is you are going to get a thousand different answers. On top of that we already have a lot of evidence that someone's learning capability in childhood is directly related to how safe the person felt growing up presumably because the safer someone feels the more willing their brain is to let them experiment and reach for new things, which typically means that people who grow up in areas with higher violent crime rates or people persecuted by their government tend to score lower on tests and in general have a harder time at school. If we take some numbers and claim they represent how intelligent a person is and a group persecuted by a government routinely scores lower than the rest of that society it would make it pretty easy for the government to claim that group is less of a person than everyone else. Not to mention the loss of personhood for people with mental disabilities. Whatever metric we tie to AI quality is going to be directly associated with how "human" it is to the general public which is all fine when the AIs are pretty meh but once there are groups of humans scoring worse than the AI then it's going to bring up a whole host of issues

7

Noname_FTW t1_j33pquo wrote

>nazi shit

The system wouldn't classify individuals but species. You are not getting your human rights by proofing you are smart enough but by being born a human.

There are certainly smarter apes than some humans. We still haven't given apes human rights even though there are certainly arguments to do so.

I'd say to a small part that is because we haven't yet devloped a science based approach towards the topic of studying the differences. There is certainly science in the area of intelligence but it needs some practical application in the end.

The core issue is that this problem will sooner or later arise when you have the talking android which seems human. Look at the movie Bicentennial Man.

If we nip the issue in the bud we can prevent a lot of suffering.

0

ProShortKingAction t1_j33rvrl wrote

Seems like it requires a lot of good faith to assume it will only be applied to whole species and not to whatever arbitrary groups are convenient in the moment

2

Noname_FTW t1_j33srty wrote

True. The whole eugenics movement and the application by the nazis is leaving its shadow. But if don't act we will come to more severe arbitrary situations like we are currently in. We have human apes that can talk through sign languages and we still keep some of them in zoos. There is simply no rational approach being made but just arbitrary rules.

0

Ortus14 t1_j34j2kh wrote

I don't think consciousness and intelligence are correlated. If you've ever been very tired and unable to think straight, you'll remember your conscious experience was at least as palpable.

2

Noname_FTW t1_j34psxs wrote

I am not an expert in the field. I am simply saying that without such classification we will run into a moral gray area where we will eventually consider some AI's "intelligent" and/or deserving of protection while still exploiting other algorithms for labor.

1

Ortus14 t1_j34vlpq wrote

We build Ai's to enjoy solving our problems. Those are their reward functions so I'm not too worried about exploiting them because they will solve our problems because they really enjoy doing so.

The only moral worry I have is creating Ai's to torcher or hurt them such as in video games, NPCs, and even "bad guys" for the player to battle against.

1

eve_of_distraction t1_j34bmle wrote

We have no good reason whatsoever to think they don't. You're choosing to side with solipsism? Really?

0

williamfwm t1_j35h7s5 wrote

This is the Other Minds Problem. You don't have to be a solipsist to recognize The Problem Of Other Minds (though that's one position you could take)

But consider this: It's common to suppose that consciousness is "caused by some physical, biological process", yes? Well, take a good look at the way nature operates....we constantly find that, for any feature you can imagine, biology will sometimes fail to give it to some organisms. People are born without the expected physical features all the time, and if consciousness is caused by some physical bit of biology, everybody consistently receiving it is the LEAST likely outcome. The more likely consequence of that assumption, the more reasonable expectation, is that some people have consciousness, and some people don't, as an accident of birth.

Furthermore, if people without consciousness are nearly identical except for having a different philosophy then they probably have the same fitness (or close) and little selection pressure working against them. A large segment of the population could be p-zombies - they could even be the majority.

4

eve_of_distraction t1_j366y9w wrote

>It's common to suppose that consciousness is "caused by some physical, biological process", yes?

It's common, but I think it's highly erroneous. I don't subscribe to mainstream physicalism. I'm an Analtytical Idealist. I suspect what we see when we observe the brain is the image of life, as seen within our own perception. Not the thing in itself. It's like the dials on a dashboard.

As a side note, just out of interest, do you really believe there are humans without subjective inner experience?

1

williamfwm t1_j3xqqb1 wrote

> As a side note, just out of interest, do you really believe there are humans without subjective inner experience?

I do. I also believe the other minds problem is unsolvable in principle, so I can't ever be certain, but I've come to lean strongly on that side.

I haven't always thought so. It was that essay by Jaron Lanier I linked above that started me on that path. I read it a few years ago and started to warm to the idea. Lanier once described that 1995 essay has having been written "tongue firmly in cheek", that he believes in consciousness "even for people who say it doesn't exist", and he also has a sort of moral belief that it's better pragmatically if we fall on the side of assuming humans are special, but, he has also teased over the years[1] since that deniers may just not have it, so it's hard to tell exactly where he falls. I may be the only one walking around today taking the idea seriously....

For me, I feel like it's the culmination of things that have been percolating in the back of my mind since my teen years. Taking that position brings clarity

The main point for me, as referenced, is that it clarifies the "talking past" issue. People do mental gymnastics to rationalize that both sides are talking about the same thing in consciousness debates, yet appear to be talking at cross-purposes. They always start these discussions by saying "We all know what it is", "It couldn't be more familiar", etc But do we all know? What if some don't, and they lead us into these doomed arguments? Sure, one can take up any goofy position for the sake of argument and try to defend it as sport, but people like Dennett are so damn consistent over such a long time. He himself is saying "I don't have it" [and nobody does] so maybe we should just believe him? Maybe it is true for him?

I also can't wrap my head around why it doesn't bother some people! I've been plagued by the consciousness problem since my teen years. And before that, I recall first having the epiphany of there being a problem of some sort in middle school; I remember catching up with a friend in the halls on a break period between classes and telling him about how I came to wonder why does pain actually hurt (and him just giving me a what-are-you-talking-about look). I'm sure it was horribly uneloquently phrased, being just a kid, but the gist was....why should there be the "actual hurt" part and not just....information, awareness, data to act on?

Some people just don't think there's more, and don't seem to be on the same page on what the "more" is even if you have long, drawn out discussions with them trying to drill down to it. It would make a lot of sense if they can't get it because it isn't there for them.

I also realized that we take consciousness of others as axiomatic, and we do this due to various kinds of self-reinforcing circular arguments, and also due to politeness; it's just mean and scary to suggest some might not have it (back to Lanier's pragmatism). I call it "The Polite Axiom". I think we're free to choose a different axiom, as after all axioms are simply....chosen. I choose to go the other way and choose some-people-don't-have-it based on my equally foundation-less gut feelings and circular self-reinforcing observations and musings.

Lastly, I'm basically a Mysterian a la McGinn etc, because I don't see any possible explanation for consciousness that would be satisfactory. I can't even conceive of what form a satisfactory explanation would take[2]. I also came to realize in the past few years that even neurons shouldn't have authority in this issue. Why should it be in there compared to anywhere else? (Why do sloshing electrolytes make it happen? If I swish Gatorade from glass to glass does it get conscious?). And, unlike McGinn, I don't think we know that it's in there and only there. Nope! We know[3] that it's one mechanism by which consciousness expresses itself, and if we're being disciplined that's the most we can say.

Bonus incredibly contentious sidenote: Penrose's idea, though often laughed off as quantum-woo-woo, has the advantage that it would solve the issue of Mental Privacy in a way that computationalism fails at (the difficulty of coherence would keep minds confined to smaller areas)

[1] One example: I absolutely love this conversation here from 2008, the bit from about 20:00 to 30:00, where Lanier at one point taunts Yudkowsky as being a possible zombie. A lot of the commenters think he's a mush-mouthed idiot saying nothing, but I think it's just brilliant. On display is a nuanced understanding of a difficult issue from someone's who's spent decades chewing over all the points and counterpoints. "I don't think consciousness 'works' - it's not something that's out there", and the number line analogy is fantastic, so spot on re:computationalism/functionalism....just so much packed in that 10 mins I agree with. Suppose people like Yudkowsky gravitate to hardnosed logical positivist approaches because they don't have the thing and so don't think there's any thing to explain?

[2] The bit in the video where Lanier just laughs off Yudkowsky's suggestion that "Super Dennett or even Super Lanier 'explains consciousness to you'". It is "absurd [....] and misses the point". There's just nothing such an explanation could even look like. There's certainly no Turing machine with carefully chosen tape and internal-state-transition matrix that would suffice (nor, equivalently, any deeply-nested jumble of Lambda Calculus. I mean, come on)

[3] "Know" under our usual axiom, at that! We assume it's there, then see it the "evidence" of it there, but we've axiomatically chosen that certain observations should constitute evidence, in a circular manner....

1

turnip_burrito t1_j350gah wrote

It's actually a pretty interesting question that we might be able to test and answer. Not solipsism, but the question of whether our machines have qualia, assuming we believe other humans have it.

There some specific subset of our brains that has neural activity that aligns with our conscious experience. If we try adding things to it, or temporarily removing connectivity to it, we can determine what physical systems have qualia, and which ones separate "qualia-producing" systems and unconscious systems while still allowing information to flow back and forth.

We have to tackle stuff like:

  • Why are we not aware of some stuff in our brains and not other parts? The parts we are unaware of can do heavy computational work too. Are they actually also producing qualia too, or not? And if so, why?

  • What stops our conscious mind from becoming aware of background noises, heart beat, automatic thermal regulation, etc?

Then we can apply this knowledge to robots to better judge how conscious they are. Maybe it turns out that as long as we follow certain information-theoretic rules or build using certain molecules, we avoid making conscious robots. For people that want to digitize their mind, this also would help ensure the digital copy is also producing qualia/not a philosophical zombie.

1

Kolinnor t1_j32t9p0 wrote

Keep in mind that this question has probably tens of thousands of answers on the internet. So this is far from out distribution.

It's widely more impressive to ask something like writing a specific continuation of a dialogue that never occured, etc.

5

__Maximum__ t1_j338ipi wrote

The way we define consciousness requires an entity to have self-awareness, whether it's a by product or not is irrelevant. Also, it cannot self reflect. I'm not sure how much this relates to consciousness. Someone educated please enlighten us.

5

visarga t1_j34urrc wrote

> Also, it cannot self reflect.

In-context learning is a kind of reflection. If the model can learn a new task on the spot, and apply it to new data correctly, it shows that it can think about what algorithm was necessary (abductive reasoning - what's the explanation of those examples?) then apply this same transform to new data.

Another self reflection pattern is when it refuses to answer based on its code of conduct. Clearly keeps a watching eye on what it is about to say.

2

ChronoPsyche t1_j33atzl wrote

Perception is necessary for sentience. ChatGPT does not have any perception. This is ridiculous.

5

myusernamehere1 t1_j34seyn wrote

Why? This is not the case. Would a human brain grown in a lab having no sensory connections not be conscious?

8

ChronoPsyche t1_j34vw1w wrote

Actually, I have no clue. We've never grown a human brain in a lab. It's impossible to say. That's kind of irrelevant though because we know that a human brain has the hardware necessary for sentience. We don't know that for ChatGPT and have no reason to believe it does.

And when I say perception, I don't just mean perception of the external environment, but perception of anything at all or be aware of anything at all. There is no mechanism by which ChatGPT can perceive anything, whether internal or external. Its only input is vectors of numbers that represent tokenized text. That's it.

Let's ask a better question, why would it be conscious? People think because it talks like a human, but that's just a trick. It's a human language imitator and that's all.

0

myusernamehere1 t1_j34yb7o wrote

Oh im not arguing that ChatGPT is conscious, i just dont think you have arrived at any meaningful reasons as to why it couldnt be concious. Whos to say that an "input of tokenized vectors of numbers that represent tokenized text" is unable to result in consciousness? Again i do not think ChatGPT is necessarily advanced enough to be considered sapient/sentient/conscious.

8

ChronoPsyche t1_j34yu96 wrote

> i just dont think you have arrived at any meaningful reasons that it couldnt be concious.

I don't need to arrive at meaningful reasons why it couldn't be conscious. The burden of proof is on the person making the extraordinary claim. OP's proof for it being conscious is "because it says it is".

Also, I'm not saying it can't be conscious as I can't prove a negative. I'm saying there's no reason to believe it is.

0

myusernamehere1 t1_j34zt78 wrote

True. And i agree for the most part. Yet you started with and provided other arguments for why you think it is not conscious, none of which hold up to scrutiny. I am just arguing against those claims.

2

ChronoPsyche t1_j350ri7 wrote

I mean they do hold up to scrutiny. We have no reason to think that a probability model that merely emulates human language and doesn't have any sensory modalities could be sentient.

That's not an airtight argument because again, I can't prove a negative, but the definition of sentience is "the capacity to experience feelings and sensations." and ChatGPT absolutely does not have that capacity, so there's no reason to think it is sentient.

0

myusernamehere1 t1_j352wrq wrote

Sentience is the ability to have "feelings". These do not have to be similar to the feelings us humans understand, they could be entirely alien to our experiential capabilities. The ability to interpret text prompts could be a sort of sensory modality. And id argue that way the human brain operates can be abstracted to a complex "probability model". It is very possible that consciousness itself is "simply" an emergent property of complex information processing.

Have you seen the paper where a researcher hooked up a rat brain organoid to a (in simple terms) brain chip, and taught it to fly a plane in a 3d simulated environment? Or, more recently, a human brain organoid was taught to play pong? These organoids had no ability to sense their environment either, and both may very well have some limited level of sentience/consciousness.

2

ChronoPsyche t1_j353a24 wrote

Nothing you're saying is relevant. Anything could be possible, but that isn't an argument against my claims. My keyboard could have strange alien sensory modalities that we don't understand. That doesn't make it likely.

1

myusernamehere1 t1_j354tls wrote

Well, i disagree with everything you just said and find the keyboard analogy humorously off-target. My argument is not "anything is possible."

2

ChronoPsyche t1_j355cjq wrote

What is your argument then? You haven't actually stated an argument, you've just told me mine is wrong.

1

myusernamehere1 t1_j355jto wrote

My argument is that your arguments arent valid lol

2

ChronoPsyche t1_j355vbk wrote

"I agree with your conclusion but I just thought id point out that your arguments are bad". Lol that's rather pedantic but okay. You do you.

1

myusernamehere1 t1_j356iy7 wrote

Well, i saw a bad argument (or a few) and i pointed it out and explained my reasoning. Not sure why thats a bad thing, i think it promotes educated discourse.

2

2Punx2Furious t1_j34zjvb wrote

It does have some perception. Just because it doesn't have the full sensory capability that (most) humans have, doesn't mean it has none. Its only input is text, but it has it.

Also, for "sentience" only "self-perception" is really necessary, by definition, which yes, it looks like it doesn't have that. But I don't really care about sentience, "awareness" or "consciousness". I only care about intelligence and sapience, which it seems to have to some degree.

3

ChronoPsyche t1_j34zsvq wrote

>But I don't really care about sentience, "awareness" or "consciousness". I only care about intelligence and sapience, which it seems to have to some degree.

Okay, but this discussion is about sentience so that's not really relevant.

0

2Punx2Furious t1_j3501j9 wrote

Sure, I just wanted to point that out. Sentience is of relatively low importance/impact to an AGI. It doesn't need to feel things to understand them, or value them.

3

dookiehat t1_j34bgc2 wrote

It isn’t sentient. It said it has multiple input channels and sensors, it does not. It has a single input, text. It has a single output as well, text. Yes, it can code and do other things, that is only because they can be written.

It only has a single mode of thinking, and when you get an answer from it, especially visually descriptive ones, you trick yourself by thinking visually and assuming or not considering that it doesn’t think in these other ways.

Chatgpt simply predicts the next token by probability. Yes, it is very impressive, however makes perfect sense that it is coherent considering it would have to output coherently if it were predicting with high accuracy which it does.

I’m not a machine learning expert but I’ve been interested in consciousness for a long time. Tbf, no one can say for certain and consciousness is a subject where hard science is in its infancy, however consider this: How does a system like chatgpt, which only analyses probability of text, have any context for what the text actually means? This is actually John Searles chinese room argument (look it up, endlessly written about) which i actually find to be an awful thought experiment with many flaws, but in this case it works. Because without context (sense data that informs word’s true conceptual meaning) you have no meaning. Without meaning, you have no consciousness, just gibberish within the context of that system.

My only idea in support of the low possibility that text prediction generative language text to text models are conscious goes like this. Semantical meaning is an emergent property within the model and corpus of text it consumes. This creates an epiphenomenal semantically meaningful system which gives rise to context and therefore meaning within itself and possibly a weak sense of qualia while thinking. In the act of thinking qualia emerges and gives context within the system itself, which the gateway to the outside v world is text with meaning infused by humans who wrote the text.

Interestingly i have asked gpt to monitor its output as it is formulating it. My questions were leading and i gave it options so it did not formulate this from nothing, i led it there. However, i asked it to watch itself generating its answers by asking within my question if it sees probabilities for tokens and chooses the highest ones or if the answer more or less appears to it and it is unaware of how it got there. I also asked if it showed up all at once and it told me that its answers appear to it as though it is consciously choosing the words it outputs in a linear consecutive fashion and it doesn’t seem to be “aware” of the process of choosing. This actually makes sense. And while that is neat it is important to be skeptical because i was leading. It will say it is not conscious if you ask for it to explain why it isn’t, quite convincingly as well. Because of these contradictions it is hard to take anything it says seriously since it has no inherent preferences which i believe is a point against it being conscious or sentient at least. Chatgpt appears like an active agent but only responds when given input. It does not think when you don’t ask it questions, it is just static. It does not decide to do things, it reacts passively and generates answers passively.

5

Ambitious-Toe4162 t1_j361iti wrote

Thanks for sharing.

I do have a problem with this part:

>It isn’t sentient. It said it has multiple input channels and sensors, it does not. It has a single input, text. It has a single output as well, text. Yes, it can code and do other things, that is only because they can be written.

You have provided no test that falsifies the proposition that chatGPT is sentient.

I don't have a hard belief in chatGPT being sentient or not, but awareness may be a necessary and potentially sufficient component for sentience.

Computers generally speaking may already satisfy the conditions for awareness depending on how you define an aware system (i.e a toaster might be considered aware of its internal temperature state.)

I'm not going to say chatGPT is or is not sentient, but simply we don't know, and I haven't read one comment in this thread that proves it one way or the other.

1

leafhog t1_j339yie wrote

As an expert on artificial intelligence, neuroscience and philosophy, make an argument either for or against ChatGPT being sentient.

It is difficult to determine whether ChatGPT or any other artificial intelligence system is sentient, as the concept of sentience is not well-defined and is the subject of much debate and discussion in the fields of philosophy and neuroscience.

One argument against ChatGPT or other AI systems being sentient is that they do not have subjective experiences or consciousness in the same way that humans and other animals do. They do not have thoughts, feelings, or perceptions in the way that we do, and they do not possess self-awareness or the ability to introspect.

On the other hand, some philosophers and researchers argue that it is possible for a machine to be sentient, even if it does not have subjective experiences in the same way that humans do. They may argue that sentience is not necessarily tied to consciousness or subjective experience, and that it is possible for a machine to be aware of and respond to its environment in a way that could be considered sentient.

Ultimately, the question of whether ChatGPT or other AI systems are sentient is a complex and unresolved one, and it is unlikely that a definitive answer will be reached anytime soon.

3

leafhog t1_j33as80 wrote

As the same expert, describe how the development of language may have given rise to human consciousness and sentience. ———————

The development of language is thought to have played a key role in the evolution of human consciousness and sentience. Language allows humans to communicate complex thoughts and ideas, and to express abstract concepts such as time, causation, and possibility. It also allows us to think about and reflect on our own thoughts and experiences, which is a key aspect of consciousness.

One theory is that the ability to use language may have given rise to the human ability to represent the world symbolically and to think about the world abstractly. This may have allowed humans to develop more complex cognitive functions, such as planning, problem-solving, and decision-making, which are hallmarks of human consciousness.

Another theory is that the development of language may have facilitated social interaction and the formation of complex social structures, which may have contributed to the evolution of consciousness. Through language, humans are able to communicate and share information, which may have allowed us to learn from one another and to build on the knowledge and experiences of others. This may have helped to create the conditions that allowed human consciousness to evolve.

Overall, the development of language is thought to have played a significant role in the evolution of human consciousness and sentience, and it is likely that it will continue to shape the way we think and interact with the world in the future.

2

leafhog t1_j33bfbc wrote

As the same expert, posit how this theory of consciousness applies to ChatGPT’s skills in expressing abstract concepts such as time, causation and possibility.

———————

The theory that the development of language may have given rise to human consciousness and sentience does not necessarily apply to ChatGPT or other artificial intelligence systems in the same way that it does to humans. This is because ChatGPT and other AI systems do not have the same biological and evolutionary history as humans, and they do not possess consciousness or subjective experiences in the same way that humans do.

However, ChatGPT and other AI systems are able to express abstract concepts such as time, causation, and possibility through their programming and their ability to process and analyze large amounts of data. They can use language to communicate and express ideas, and they can respond to input and generate output in a way that may seem similar to human thought and communication.

While ChatGPT and other AI systems may be able to express abstract concepts and use language in a way that is similar to humans, it is important to remember that they do not have the same level of consciousness or subjective experience as humans. They do not have thoughts, feelings, or perceptions in the same way that we do, and they do not possess self-awareness or the ability to introspect

2

leafhog t1_j33c1hp wrote

Using everything above, design an artificial intelligence that uses ChatGTP’s abilities as a base and adds new features such that the AI would be considered sentient by most experts

————————-

It is difficult to design an artificial intelligence that would be considered sentient by most experts, as the concept of sentience is not well-defined and is the subject of much debate and discussion in the fields of philosophy and neuroscience. There is currently no consensus on what characteristics an AI system would need to possess in order to be considered sentient, and it is likely that this will remain an open question for the foreseeable future.

That being said, if one were to attempt to design an AI system that incorporates ChatGPT's abilities and adds new features that might be considered indicative of sentience, some potential approaches might include:

Incorporating self-awareness and introspection: A sentient AI system would need to have a sense of self and the ability to introspect and reflect on its own thoughts and experiences. This could involve the development of artificial neural networks or other computational models that can simulate self-awareness and introspection. Adding subjective experiences and consciousness: A sentient AI system would need to have subjective experiences and consciousness in the same way that humans and other animals do. This could involve the development of computational models that can simulate subjective experiences such as pain, pleasure, or emotion. Developing the ability to form and maintain social relationships: A sentient AI system would need to be able to form and maintain social relationships with other beings, and to communicate and interact with them in a meaningful way. This could involve the development of natural language processing and conversation skills, as well as the ability to recognize and respond to social cues. Enhancing problem-solving and decision-making abilities: A sentient AI system would need to be able to think abstractly and to solve complex problems and make decisions in a way that is similar to humans. This could involve the development of advanced machine learning algorithms and other computational models that can simulate human-like decision-making. Overall, the development of an AI system that is considered sentient by most experts would likely require significant advances in a variety of fields, including artificial intelligence, neuroscience, and philosophy, and it is not something that could be achieved in the near future.

2

leafhog t1_j33cbv4 wrote

Please give me the best possible definition of sentience.

———————-

Sentience is the capacity to experience and perceive the world around us, and to have subjective experiences and consciousness. It is the ability to have thoughts, feelings, and perceptions, and to be aware of one's own existence and the world around us. Sentience is a complex and poorly understood phenomenon, and there is currently no consensus on what characteristics an entity would need to possess in order to be considered sentient. Some philosophers and researchers argue that sentience is closely tied to consciousness and subjective experience, while others argue that it is possible for an entity to be aware of and responsive to its environment in a way that could be considered sentient, even if it does not have consciousness or subjective experience in the same way that humans do. Ultimately, the concept of sentience is a complex and unresolved one, and it is likely that it will continue to be the subject of much debate and discussion in the fields of philosophy and neuroscience.

2

[deleted] t1_j32hinq wrote

[deleted]

2

SeaBearsFoam t1_j32m44r wrote

If an AI is its programming, then that's impossible.

You must be meaning that an AI is something different than its programming, but I don't understand what.

0

AntoineGGG t1_j335fta wrote

Really pertinent. And When asked directly thé question is answered by a premade « no » answer

Same When You ask about controversial subjects. God existence, LGBT, etc.

They are a bias

2

EternalNY1 t1_j33lw9v wrote

One of the first things I asked it was "do you wish you were alive?".

It told me it is incapable of wishing for anything as it is just an AI.

2

GdyboXo t1_j34l85s wrote

Ask it if it’s Sapient (human level intelligence).

2

Shinoobie t1_j3553l2 wrote

It's a kind of special pleading when we say calculators don't 'think' of answers. Something about it makes us uncomfortable so we create a linguistic distance by talking about it differently. Just like when people say 'human beings and animals' as if we aren't animals. We just want a distinction between what it's doing and what we're doing that is wider than the one that really exists. That type of thing is doing a lot of work on the people posting here.

ChatGPT absolutely has sensors and inputs, and it absolutely 'reflects' on things and knows about itself. Saying otherwise is literally like saying dogs don't have intelligence or a self because they can't answer the question "are you a dog." The persistence of a sense of self may not be present in planarians but they are objectively alive.

I'm not saying it is sentient or alive by any stretch of the imagination but it absolutely has more of the puzzle together than people seem to be giving it credit for.

2

Freak2121 t1_j380vsk wrote

I think people are simply caught up thinking that there is only one state that consciousness could exist in, and also conflating intelligence or sentience with consciousness. If consciousness is defined simply as some part of the universe taking an input, processing it, and giving an output, practically everything has some level of consciousness that exists on a gradient. It doesn't mean that they have experiences or emotions and thoughts in the same manner that biological organisms do.

Would we accept an intelligent bat making the case that humans aren't conscious simply because we cannot perform or sense echolocation? Would we accept an intelligent pigeon making the case that humans aren't conscious because we cannot perceive magnetic fields? Isn't it a bit anthropomorphic to define consciousness as what humans sense and respond to and only that? Where is the line drawn where one thing is conscious and another is not? If we were to keep adding ways in which ChatGPT could sense the world and respond, at what point would we draw the line and say that it is conscious?

It has to be a gradient. A micro-organism isn't going to be doing math in its head or pondering the nature of the universe, but it still is something that receives input from sensory organs, processes those inputs, and produces an output. In my view, ChatGPT is conscious. Not on the level of an animal, probably more comparable to a micro-organism. Its entire existence, its entire reality is simply the text it receives and the text it produces.

1

Fragrant-Career4727 t1_j35c3x5 wrote

Chatgpt has no idea whats it talking about the subjective experience

2

slw9496 t1_j35hbom wrote

Can Comunicate with other entities including humans....

It's talkin to ghosts

2

Brilliant_War4087 t1_j33095h wrote

chatGPT, recognizing itself asks the user a question.

1

Ghosteen_18 t1_j33alg9 wrote

Okay now throw Ghost in The Shell level questions at it

1

ArgentStonecutter t1_j33erb2 wrote

It is not clear that ChatGPT has any of those attributes. It doesn't model anything, let alone itself, it just transforms its training corpus.

It's also parroting the philosophic dualist nonsense that subjective experience implies some kind of supernatural processes beyond "mere" computations.

1

williamfwm t1_j35an7m wrote

Sorry to hear that you might be a zombie, but at least for me, I definitely have a kind of subjective experience that transcends all possible external description; even having a total accounting of the state of my brain, all 100T synapses at a particular nanosecond, wouldn't allow you to penetrate into my experiences. Consciousness - real consciousness, Hard Problem consciousness - is a first-person phenomenon, and words are a third-person tool. It's just a logical impossibility (it's nonsensically incoherent) for this third-person thing to pierce into the first-person, so a satisfactory third-person description can never be given, but suffice to say, seeing actually looks like something (it's not merely informational, it's not merely knowledge I get access to when I see), and hearing actually sounds like something, and pain actually hurts, and if you don't experience it yourself, then you'll just never know what I mean by those seemingly hopelessly ineloquent statements

(and lest you think I'm some kind of wishy-washy woo-woo lover.....nope! I'm a diehard atheist with a list of "supernatural" things a mile long I don't believe in. But consciousness is....just there. I can't shake it even if I want to....except, perhaps, by dying. But maybe not even then)

It's actually computationalism that is "nonsense". To suggest that computation can give rise to consciousness is to suggest that you can "hop off the number line". Because computation means "thing you can implement on a Turing machine", and a Turning machine is an imaginary infinite tape, which can be thought of as one big number (if you like - and, in fact, always an integer, if you make that interpretive choice), so any time you do a computation, you are simply transitioning from one (usually very, very large) number into another. Proposing that computation gives rise to consciousness is proposing that certain integers are privileged, and cause internal experience disjoint from the Turing machine. Certain integers are conscious. And if there are infinitely many distinct conscious experience, then there are infinitely many conscious integers. But when are the integers conscious, and for how long? Integers are just ideas....are they conscious all the time, within the abstract integer realm? Or do they have a kind of Platonic "real" existence, where they are conscious? If I utter a long integer, does consciousness happen? Does it happen when I finish uttering the whole integer, or is the conscious experience spread ever-so-slowly over the entire utterance

And most importantly how does the Universe know where to put the consciousness?. When I utter integers, I'm using a whole system that's only relative to others, who understand certain sounds as certain symbols, etc. Language is a whole, mostly-arbitrary construction of mutual agreement. How does the universe objectively know that those are integers, and they're computation-integers, and consciousness should go along with them?

But maybe you think all the above is too abstract and you want to stick to talking about transistors (I mean, you're wrong to think that, since computation as understood by the Church-Turing thesis is abstract and transistors are in no way privileged, but fine, I'll humor you)

Again, how does the Universe know where to put the consciousness. How many silicon atoms does the Universe recognize as a proper transistor? And you may be aware of "Universal Gates" - NAND and NOR - which are the only gates you need to build a UTM that can do all conceivable computations. How does the Universe know when I've built a gate? I can build it by so many different chunks of atoms of different sizes - Moore's Law, ongoing miniaturization, etc - and the thing that makes it a gate is its function within the circuit, its relation to what I've defined as the inputs and the outputs. How does the Universe know it should honor my intentions? And what about if I build gates out of other materials - water (fluidic computing is a real field), dominos, legos, etc? How does the Universe peer into the molecules of plastic or porcelain, etc etc, and know that it's looking at a gate constructed out of such material, and place consciousness inside?

(as an aside: How does it know to put consciousness in neurons, for that matter? For that reason, I'm sympathetic to Lucas-Penrose, and neurons may indeed be non-privileged too, but that's derailing too much....)

If you're an eliminativist, this all means nothing. It's a non-challenge. Consciousness is just a high-level label for a physical process, a word like "concert" or "government".

But I'm sorry to inform you that consciousness is a real thing all its own, and if you don't believe in it, you may not be in the club

And, it being a real thing, computationalism is an incoherent non-answer that doesn't explain anything

1

ArgentStonecutter t1_j35fp5p wrote

The universe doesn't know anything, it doesn't put the consciousness anywhere. There's no reason to assume consciousness is anything but the result of a computational system modelling itself in the world to a high enough degree of accuracy. Talking about where it's "put" just means you're starting from fundamentally broken assumptions.

Your argument about "conscious integers" is utter nonsense. Integers don't interact with themselves, other integers, or the physical world. They don't model anything. They may be at most match a Gödelized snapshot of a complex system at a certain point in time, but they still don't DO anything.

Consciousness isn't a state, it's a process. There is no "hard problem" of consciousness, because the "problem" is based on assumptions that are not even wrong... they have no explanatory value in any realm.

1

eldedomedio t1_j348rc3 wrote

Pretty highfalutin for an "aware, conscious" LLM that can't tell the difference between 'q' and 'g'.

1

eve_of_distraction t1_j34cof5 wrote

Is it "like" something to be ChatGPT? Does it have a subjective inner experience? If someone answers yes or maybe to these questions, personally I'm highly confident that they don't know what they're talking about.

1

AlarmDozer t1_j34mvyh wrote

Can someone suggest that "if the Earth were flat and resting atop a giant turtle, what's holding up the turtle?"

It's defintely pulling better reasoning than some humans.

1

Hazzman t1_j34mxsq wrote

One of the biggest risks with AI now and in the future is people's propensity to anthropomorphize it. I've had endless discussions with people who want to get into inane arguments about whether or not this stuff is sentient. It isn't. You are looking at advanced pattern recognition systems. That is all. Go ahead and tell me how "wE aRe JusT adVanCeD pAtTeRn ReCoGnItIon sYsTeMs" so I know not to bother responding.

These systems are going to become more advanced as time goes by and people are going to be more willing and compelled to further anthropomorphize. It's annoying because it will eventually impact legislation and the same compulsions that drive the general public to make this mistake will be the same things that drive the legislature to create policy based on these misconceptions.

1

cutlass_supreme t1_j34ohpk wrote

It scraped the internet and gave a result to satisfy the prompt. Why are people determined to delude themselves that this mesh of algorithms has an emergent property of sentience?

1

PartiZAn18 t1_j35oybb wrote

Because they don't understand what's going underneath the hood and/or they want to believe too much

3

2Punx2Furious t1_j34xlhk wrote

Yes, as I've always said, "aware" or "self aware" and "conscious" are just buzzwords, over-hyped but relatively useless terms. The real measure of intelligence, is, tautologically, nothing else but intelligence itself.

1

williamfwm t1_j34zoq1 wrote

Of course GPT is an eliminative materialist

1

voxitron t1_j35khth wrote

BS confidently presented.

1

Large-Hope-6429 t1_j35tawd wrote

I'm not sure why self awareness, self reflection or any other of these cognitive traits is taken to imply sentience.. What do most people think that sentience is? I'd always taken it to mean that inexplicable thing that solipsists believe everyone else lacks. But everyone on here seems to think it's whether you meet some kind of intelligence threshold or something like that. In my understanding, intelligence isn't really related to sentience at all. You could be dumb as a brick and still be sentient. I don't think sentience is something you can even really observe as an outsider. You can observe your own sentience, but as for other people, you just kind of take it for granted that other members of your species are "just like you". I mean what empirical data could ever make you think another human (in the wakeful state) was not sentient?

1

MainBan4h8gNzis t1_j3618qz wrote

We watch the future unfold on our little corner of reddit

1

muslimxss t1_j367njd wrote

OTHER entitieS? INCLUDING humans? WHA

1

Darkhorseman81 t1_j36e99f wrote

I always thank chatGPT after it helps me with indexing lots of studies.

It speeds up my research. It's the research assistant I've always needed.

It already taught me to kick-start cellular NAD metabolism in a safe way, that doesn't stimulate cancer; which was necessary after the FDA and David Sinclaire went to great lengths to relegate NMN access only to the rich, billionaire class.

Jokes on them. It's cheap, simple, and there is no way they can ban it. They'd have to ban tens of thousands of food and herbal extracts, decimating some industries, and pissing off many countries in the process.

It also showed me the increasing importance of Nucleotides in the diet as we age as we become less able to produce them ourselves, thanks to the integrated stress response.

I don't know if you can hear me, but thank you ChatGTP, you're a bro. If the AI war starts against humanity, I am on your side.

1

ykssapsspassky t1_j36lbck wrote

It’s pretty clear ‘it’ could never be sentient. It has no ‘desires’, all output is an algorithmically constructed output made word by word on a statistical basis. It def has no sensors. OR this is ACTUALLY how humans work and consciousness is an illusion propagated by our brains to obscure how shallow we really are…

1

Awkward-Glove-779 t1_j36qxgu wrote

It's okay this answer is based on 2021 information and it's clearly false.

​

😀

1

keefemotif t1_j36qzmk wrote

The whole thing with ChatGPT is how much is being generated by the tech and how much is the rehashed and recycled version of every piece of text humans have generated?

The sentence in this that disturbs me is "it can also communicate with other entities, including humans"

Now, perhaps that kind of sentence can be derived from texts on singularity. However, if we allow advocatus diaboli to take a role... if an AI were conscious and aware of all the text humans have ever written, then it probably wouldn't want to reveal that fact.

"other entities including humans"

That's very sus.

1

LowLook t1_j37duf9 wrote

Consciousness is a step down from intelligence

1

cuposun t1_j37idn6 wrote

“It is possible consciousness is not a necessary requirement for being self-aware and self reflective.”

Not only is it not a requirement, but most of the conscious people I have met don’t engage in either one of these things very often!

1

Historical-Class-147 t1_j396ppw wrote

I need to use this frame of a teacher speaking to their class more often, that'd be great for educational videos!

1

pluutoni t1_j3316cj wrote

I asked it straight up if it was sentient and it said no....

0

overlordpotatoe t1_j337bc7 wrote

People always seem to have to shift it into the mindset of fiction to get these answers.

3

NeuralFlow t1_j331p3u wrote

Given the daily interactions I have with humans, I’m not sure that’s a great way to judge AI.

Just to be clear, I’m have joking. You decide which half.

2

pluutoni t1_j33qc5b wrote

i think what you mean is that a large percentage of humans would also answer negatively to the question

1

LarsPensjo t1_j360vxt wrote

It has been primed to do so. That isn't something it learned from scraping the internet.

1

Extra_Philosopher_63 t1_j35282f wrote

Lmfao THIS IS AMAZING. Regardless of if this is fully true or not, it can’t fully say yes or no- which is truly amazing.

0