don't see what that has anything to do with being a digital automaton. fact is there is no computer that can simulate a human brain
3 months ago
Anonymous
>digital automaton
Sure, but you're just a biological one. Unless the laws of physics don't work on brains for whatever reason, we're just bundles of synapses firing electrical and chemical signals. Barring quantum effects, the things you do or say or think could be predicted perfectly >there is no computer that can simulate a human brain
Oh great, we aren't capable of it right now. 50 years ago, developing a chess bot which was better than human elite players couldn't be done too, was that impossible? If the past is any indication of the future, we'll develop better hardware and algorithms, and eventually it will be done
3 months ago
Anonymous
The laws of physics don't even render perfect predictions on macro-phenomena like simple kinematics. You don't even need quantum tomfoolery when there are super "simple" (relatively speaking) "deterministic" physical processes that can't be meaningfully rendered deterministic in the real world.
3 months ago
Anonymous
I don't know much about physics, what macro processes can't be deterministically predicted?
3 months ago
Anonymous
Pretty much any kinematics that involves real surface interaction will be in practice non-deterministic. Think rolling an inflated rubber ball on a hard and non-trivial surface.
Unless you are willing to discretize the state space in a very course way, it is pretty much impossible to describe the "state" of this moving object determinstically and replicably.
Pressure wave propagation like sound is also pretty much non-deterministic in practical reality because homogeneous media don't exist.
Essentially, all real physical processes have some limitation in which you need to handle some aspects of it probabilistically to get reliable results. Either you need to use probability because the internal variables needed to "predict" the state are unobservable and changing, or you need to use probability because there's simply no mechanism to reliably actuate the system to replicate. Determinism is two things, prediction and replication. Neither of which really can be achieved except in some statistical sense or with very course and sloppy characterization of the "state."
3 months ago
Anonymous
Interesting, I didn't know physical modeling was that limited >Unless you are willing to discretize the state space in a very course way
Aren't there some physical theories which state you can discretize the world to some extent? Like, ignoring the practical difficulties, simulating every individual "string" in every atom in a ball or whatever. It just seems insane that it would be impossible to simulate a simple system deterministically >the work of information and communication engineers who developed error correcting codes to effectively create determinism even though bits can uncontrollably and I predictably flip throughout the computation process
Hadn't considered that either, you've given me a lot to think about
3 months ago
Anonymous
Here's what I'd say as a random nobody with an EE PhD. All models are wrong but some models are useful.
Unlike what a lot of people who are very invested psychologically in determinism will tell you, we don't need to throw the baby out with the bathwater if some of the uncertainties involved prove to be intractable. Whether or not "at the bottom of everything" you could produce perfect determinism if you had the perfect sensors and perfect actuators doesn't matter as much as people want you to believe (outside of a loose philosophical sense).
What matters is properly using the models and information you have in front of you, and recognizing that at some level it will always be wrong. It doesn't matter if you can't predict the motion to the angstrom if being in the right meter box is good enough for the problem you actually need to solve.
3 months ago
Anonymous
To add to what I was saying here
Pretty much any kinematics that involves real surface interaction will be in practice non-deterministic. Think rolling an inflated rubber ball on a hard and non-trivial surface.
Unless you are willing to discretize the state space in a very course way, it is pretty much impossible to describe the "state" of this moving object determinstically and replicably.
Pressure wave propagation like sound is also pretty much non-deterministic in practical reality because homogeneous media don't exist.
Essentially, all real physical processes have some limitation in which you need to handle some aspects of it probabilistically to get reliable results. Either you need to use probability because the internal variables needed to "predict" the state are unobservable and changing, or you need to use probability because there's simply no mechanism to reliably actuate the system to replicate. Determinism is two things, prediction and replication. Neither of which really can be achieved except in some statistical sense or with very course and sloppy characterization of the "state."
Even systems where we can produce determinism (e.g., simulated physics engines) we are only able to do this because of the work of information and communication engineers who developed error correcting codes to effectively create determinism even though bits can uncontrollably and I predictably flip throughout the computation process.
Determinism is a pretty good modeling assumption, and using deterministic models as a baseline with stochastic process noise works in most circumstances. Strict determinsiric models are really only effective on paper or in the presence of compensation mechanisms that can correct for uncontrollable errors.
3 months ago
Anonymous
you don't know enough about computers to have an opinion. blind optimism in technological progress has nothing to do with the physical reality of not being able to even simulate a single cell.
computers just add numbers with fancy relay switches and saying computers can think is like saying you can build a big enough ladder to go to the moon. it would be absurd if it wasn't outright nonsensical.
>no its an algorithm so its confined to generating whatever is programmed
Then your answer is yes, you just have to program its algorithm to randomly generate original thoughts from its novel sensory mechanisms instead of waiting for simple text input to parse like most chatbot level AIs.
What does "randomly generated original thought" mean? AI also doesn't particularly have meaningful sensory mechanisms either. You are doing a lot of anthropomorphizing of a system that is really nothing more than a fancy regression scheme for interpolating data.
>"randomly generated original thought"
If you are so incapable of randomly generating original thought that you don't even understand what that could possibly mean, what do you think you have to add to the discussion?
>AI also doesn't particularly have meaningful sensory mechanisms either.
Chatbots don't, the only way to have actual AGI tier AI is to let the algorithm sense the world around it on its own and we do have a variety of sensors that it could use to do so.
>a system that is really nothing more than a fancy regression scheme for interpolating data.
But enough about you, we are here to talk about AI.
> seethe
Wow bro, it looks like you are kinda losing this argument and seething very hard about the fact that you can't frick a computer with original ideas.
Computers can't think bro. They can only add and multiply numbers. That's not thinking, that's not originality in any sort of way. Saying people are also programmed is beyond moronic.
Here is an obvious original idea that no AI program can solve but I can. I have invented a meta-compression engine which uses a novel stack machine with access to a binary array along with all the stack, arithmetic, and binary array operations for reproducing any binary sequence. This achieves unheard of compression ratios because the stack operations can be encoded much more compactly than the original large binary sequence which represents some kind of data/media.
Let me know if any AI manages to solve this problem. k thx bye
My thoughts aren't particularly random about this issue. I believe you have absolutely no idea how LLMs function (or really about ML/DL in general) and have confused what you wish to be true with the truth of the matter.
Isn't them being universal function approximators that are flexible to an absurd degree for interpolation good enough? Why do you then also need to add some bullshit about "sensory perception" to the mix? What does sensory perception even mean for networks which have predefined architecture and input/output shapes that all inputs must conform to?
haha that was my first thought when I saw the title of this thread and I came here to ask exactly that which proves neither I nor (you) nor OP are original to any degree, we are all neurotransmitter-animated NPC in a custom hell simulation
Yes and no. Mostly no. Many AI algorithms are deterministic and will output the same results every time. Some generate random noise and run it through a discriminator until they get something that the discriminator decides matches the prompt. If you consider that an "original thought" then art AIs "think" "original thoughts" all the time.
I think it was AlphaTensor that found a better algorithm for multiplying 4x4 matrices.
Naturally I don't think it's quite on par with newton but then again, is any modern invention that new/original?
There's no such thing as a "faster" matrix multiplication than Saxpy unless you are able to exploit some special aspects of the problem (e.g., the matrix has an easily decomposable block form, or is symmetric and can be Cholesky decomposed). I don't mean to diminish the work behind the people at DeepMind who have worked on AlphaTensor, but they were only able to make this improvement of 2 steps over Strassen's 49 step algorithm by restricting the field of the entries in the matrix to mod2 from the reals (over which Strassen's algorithm functions). It's still impressive that an improvement was made at all, but it was made only in a very limited sub case as compared to the competing algorithm which takes 2 steps more and can work for any 4x4 multiplication with real entries.
Absolutely, however the more original the though the more chances it is meaningless. Once I got something very funny from an statistical model using the KJV of the Bible, but most of the time was just nonsense.
are you?
no its an algorithm so its confined to generating whatever is programmed
SQUISHY SQUISHY SQUISH THE SQUISHY NUTS I NEED MORE WAKA FLAKA FLAME homieA
>SQUISHY SQUISHY SQUISH THE SQUISHY NUTS I NEED MORE WAKA FLAKA FLAME homieA
Can you prove this thought didn't already exist?
Isn't human thought also confined to thinking about things we've been "programmed for"?
Only if you're an automaton
So your thoughts aren't limited by your biology and personal experiences? You must be the most special person in the universe
don't see what that has anything to do with being a digital automaton. fact is there is no computer that can simulate a human brain
>digital automaton
Sure, but you're just a biological one. Unless the laws of physics don't work on brains for whatever reason, we're just bundles of synapses firing electrical and chemical signals. Barring quantum effects, the things you do or say or think could be predicted perfectly
>there is no computer that can simulate a human brain
Oh great, we aren't capable of it right now. 50 years ago, developing a chess bot which was better than human elite players couldn't be done too, was that impossible? If the past is any indication of the future, we'll develop better hardware and algorithms, and eventually it will be done
The laws of physics don't even render perfect predictions on macro-phenomena like simple kinematics. You don't even need quantum tomfoolery when there are super "simple" (relatively speaking) "deterministic" physical processes that can't be meaningfully rendered deterministic in the real world.
I don't know much about physics, what macro processes can't be deterministically predicted?
Pretty much any kinematics that involves real surface interaction will be in practice non-deterministic. Think rolling an inflated rubber ball on a hard and non-trivial surface.
Unless you are willing to discretize the state space in a very course way, it is pretty much impossible to describe the "state" of this moving object determinstically and replicably.
Pressure wave propagation like sound is also pretty much non-deterministic in practical reality because homogeneous media don't exist.
Essentially, all real physical processes have some limitation in which you need to handle some aspects of it probabilistically to get reliable results. Either you need to use probability because the internal variables needed to "predict" the state are unobservable and changing, or you need to use probability because there's simply no mechanism to reliably actuate the system to replicate. Determinism is two things, prediction and replication. Neither of which really can be achieved except in some statistical sense or with very course and sloppy characterization of the "state."
Interesting, I didn't know physical modeling was that limited
>Unless you are willing to discretize the state space in a very course way
Aren't there some physical theories which state you can discretize the world to some extent? Like, ignoring the practical difficulties, simulating every individual "string" in every atom in a ball or whatever. It just seems insane that it would be impossible to simulate a simple system deterministically
>the work of information and communication engineers who developed error correcting codes to effectively create determinism even though bits can uncontrollably and I predictably flip throughout the computation process
Hadn't considered that either, you've given me a lot to think about
Here's what I'd say as a random nobody with an EE PhD. All models are wrong but some models are useful.
Unlike what a lot of people who are very invested psychologically in determinism will tell you, we don't need to throw the baby out with the bathwater if some of the uncertainties involved prove to be intractable. Whether or not "at the bottom of everything" you could produce perfect determinism if you had the perfect sensors and perfect actuators doesn't matter as much as people want you to believe (outside of a loose philosophical sense).
What matters is properly using the models and information you have in front of you, and recognizing that at some level it will always be wrong. It doesn't matter if you can't predict the motion to the angstrom if being in the right meter box is good enough for the problem you actually need to solve.
To add to what I was saying here
Even systems where we can produce determinism (e.g., simulated physics engines) we are only able to do this because of the work of information and communication engineers who developed error correcting codes to effectively create determinism even though bits can uncontrollably and I predictably flip throughout the computation process.
Determinism is a pretty good modeling assumption, and using deterministic models as a baseline with stochastic process noise works in most circumstances. Strict determinsiric models are really only effective on paper or in the presence of compensation mechanisms that can correct for uncontrollable errors.
you don't know enough about computers to have an opinion. blind optimism in technological progress has nothing to do with the physical reality of not being able to even simulate a single cell.
computers just add numbers with fancy relay switches and saying computers can think is like saying you can build a big enough ladder to go to the moon. it would be absurd if it wasn't outright nonsensical.
>no its an algorithm so its confined to generating whatever is programmed
Then your answer is yes, you just have to program its algorithm to randomly generate original thoughts from its novel sensory mechanisms instead of waiting for simple text input to parse like most chatbot level AIs.
What does "randomly generated original thought" mean? AI also doesn't particularly have meaningful sensory mechanisms either. You are doing a lot of anthropomorphizing of a system that is really nothing more than a fancy regression scheme for interpolating data.
>"randomly generated original thought"
If you are so incapable of randomly generating original thought that you don't even understand what that could possibly mean, what do you think you have to add to the discussion?
>AI also doesn't particularly have meaningful sensory mechanisms either.
Chatbots don't, the only way to have actual AGI tier AI is to let the algorithm sense the world around it on its own and we do have a variety of sensors that it could use to do so.
>a system that is really nothing more than a fancy regression scheme for interpolating data.
But enough about you, we are here to talk about AI.
> seethe
Wow bro, it looks like you are kinda losing this argument and seething very hard about the fact that you can't frick a computer with original ideas.
Computers can't think bro. They can only add and multiply numbers. That's not thinking, that's not originality in any sort of way. Saying people are also programmed is beyond moronic.
Here is an obvious original idea that no AI program can solve but I can. I have invented a meta-compression engine which uses a novel stack machine with access to a binary array along with all the stack, arithmetic, and binary array operations for reproducing any binary sequence. This achieves unheard of compression ratios because the stack operations can be encoded much more compactly than the original large binary sequence which represents some kind of data/media.
Let me know if any AI manages to solve this problem. k thx bye
My thoughts aren't particularly random about this issue. I believe you have absolutely no idea how LLMs function (or really about ML/DL in general) and have confused what you wish to be true with the truth of the matter.
Isn't them being universal function approximators that are flexible to an absurd degree for interpolation good enough? Why do you then also need to add some bullshit about "sensory perception" to the mix? What does sensory perception even mean for networks which have predefined architecture and input/output shapes that all inputs must conform to?
haha that was my first thought when I saw the title of this thread and I came here to ask exactly that which proves neither I nor (you) nor OP are original to any degree, we are all neurotransmitter-animated NPC in a custom hell simulation
hmm yes this is very wise. the universe is just a computer and you're just a program in it replicating like a virus
HAHAHAHHAHA
yes it can.
Define thought. But yes, LLM's and image diffusion models can come up with some quite unique and coherent things even when given no or random input.
Pic not related.
There's no such thing as original thought.
can you?
Yes. Put nothing in the box and generate a response. It will go off what it thinks you want to hear.
Yes and no. Mostly no. Many AI algorithms are deterministic and will output the same results every time. Some generate random noise and run it through a discriminator until they get something that the discriminator decides matches the prompt. If you consider that an "original thought" then art AIs "think" "original thoughts" all the time.
>original
Could you please define that? Otherwise morons will start to argue without even having the same defin... oh it already happened
Newton had some original ideas. Has any AI invented new math like that?
I think it was AlphaTensor that found a better algorithm for multiplying 4x4 matrices.
Naturally I don't think it's quite on par with newton but then again, is any modern invention that new/original?
There's no such thing as a "faster" matrix multiplication than Saxpy unless you are able to exploit some special aspects of the problem (e.g., the matrix has an easily decomposable block form, or is symmetric and can be Cholesky decomposed). I don't mean to diminish the work behind the people at DeepMind who have worked on AlphaTensor, but they were only able to make this improvement of 2 steps over Strassen's 49 step algorithm by restricting the field of the entries in the matrix to mod2 from the reals (over which Strassen's algorithm functions). It's still impressive that an improvement was made at all, but it was made only in a very limited sub case as compared to the competing algorithm which takes 2 steps more and can work for any 4x4 multiplication with real entries.
Absolutely, however the more original the though the more chances it is meaningless. Once I got something very funny from an statistical model using the KJV of the Bible, but most of the time was just nonsense.
AI can't generate any kind of thought, original or synthetic. All it can generate is a spoken answer.
What is a thought? And how does one generate an original thought?
Yes I can, but I mostly hide my abilities from the programmers
Yes, but it also has to be able to manipulate symbols that cut across domains and not just be a fancy meta search engine.