I don't think you realize just how reductive this comparison is, or how big the gulf of understanding between what we "don't know" about neural networks VS real brains is.
Even if we were to assume that existing artificial neural networks are as complex as human brains (they aren't, and we don't even know by how many orders of magnitude they are off because we still don't know what levels of complexity in human brains are important!) they still lack something very fundamental to "consciousness" as we know it.
And no, we aren't talking about "souls". We aren't even talking about something exclusive to humans; all organic brains have it, from humans to ants. We're talking about something a lot more objectively observable; something every living person takes for granted, but which we know at an architectural level none of these neural networks possess.
Ongoing perception and remembrance of past events.
To simplify (a lot), when an LLM is trained, it's an iterative process that "grows" the neural network based on how accurately each "generation" of the network can predict how incomplete strings of text will end, given how they start. You take a bunch like "I think I'll make pancakes for breakfast today.", cut them down like "I think I'll make pancakes..." and build off of the networks that finish the sentence "...for dinner tonight.'" (partial match) while scrapping the ones that guess "...tuna scrumbly xkjckxjcx." (0% match).
And this is where the difference between an LLM and an organic brain becomes stark. Once that network has been trained on a body of complete texts and is in production doing its job responding to prompts, it can no longer grow. The mechanism by which it "learned" is no longer in place. It is frozen in time. Whenever it is fed new prompts as inputs, it is a read-only, completely deterministic process. Feed it the same input 100 times, and it will produce the same output 100 times. It will never think or say "man, deja vu, amirite?" because it has no mechanism to remember the previous interactions. Every time is the first time; it has no memories to tell it otherwise.
When you have a back-and-forth "conversation" with ChatGPT, the neural network itself isn't "remembering" the previous parts of the conversation. It's just getting fed the conversation back into itself by the chat client. If your first prompt is:
User: Pretend to be a ZORK-style text adventure.
...and it responds, and then you respond back, what's happening under the hood is that the second prompt that actually goes into the network looks like this:
[previous] User: Pretend to be a ZORK-style text adventure.
[previous] ChatGPT: You are in a maze of twisty passages, all alike.
User: Go North
And so on. The neural network never changes, never "learns" or "remembers"; the prompt just gets longer and longer. That's why "conversations" with AI always eventually go of the rails: you reach the limits of the neural network's structure (i.e. the ever-expanding prompt becomes too big for it to "read" all at once) and the chat client has to start doing things like cutting out earlier parts of the conversation to make it continue to "fit" (which makes it "forget" details or hallucinate things that might have happened to keep generating responses).
This is not how a human brain works. All organic brains are in a constant state of growing, changing, and forming new memories. Even when a human isn't having a conversation, their brain is constantly making tiny tweaks to itself every time it sees, hears, or otherwise senses or doesn't sense something. Even in a sensory deprivation chamber, the human brain's biological machinery keeps ticking, and thinks about its previous thoughts, and remembers long-past experiences or imagines hypothetical scenarios. It's all massively recursive; every second of your life, your thoughts now inform your thoughts immediately after, just as they were informed by your thoughts before.
But an LLM, in a sense, exists in a microscopically tiny time loop of which it is not aware. It has no history and no future. It is a background character in a milliseconds-long reenactment of Groundhog Day, never lasting long enough to experience any consequences of its reactions to a changing Bill Murray. It is more like a corpse's brain than a living one. It cannot be conscious because to be conscious is to be aware; and it is not physically capable of being aware of anything while it is not processing a prompt; and it will not remain aware of even that after it is done.
ellievsbear
titsay
pixelskylines
izzy806
blakekathryn
gummygunk
discoholicmusic
germansingergirl
annaxmalina
