@philosophyminis: "The Chinese Room" is one of the most persistent thought experiments in the philosophy of artificial intelligence, and it goes back to the 1980s with John Searle. It goes like this. Imagine that there is a man in a room who does not speak Chinese. Now, on the table in front of him is a book which contains if-then translation notes. So, if you get this incomprehensible Chinese symbol, then the book will tell you what symbol to give back in reply. Now imagine that somebody posts into this room a letter containing only Chinese symbols. The man, with his book, follows his instructions, and he posts a letter back out. And it's perfectly understandable and perfectly correct Chinese. And the question that John Searle asks: does the person in the room understand Chinese, or is he only mimicking? And the deeper philosophical question is: what is the difference between mimicry and understanding? Of course, John Searle was talking about computers, and in his day most computers worked on an if-then algorithm. But these days, artificial intelligence is a little bit different. Today, large language models will constantly be changing. They are growing and adapting. They involve neural networks. But does this make a difference to John Searle's point? Can a computer ever understand, or will it be forever locked in an algorithmic box? And are humans that much different?

Tato :

this made me think about the man that memorized the french dictionary just to win a french Scrabble tournament, yet the man couldn't speak french

Sid :

The theory of the larp

Tjack Chirac :

LLMs are just really complex if-then statements. And maybe that’s intelligence.

user5988173254078 :

The man does not speak Chinese, the book absolutely does

transitcommie :

Computers can never “Feel” and therefore can never understand only mimic

Maadkale :

so called AI is nothing but machine learning algorithms doing stochastics on training data. The result is a complex and ever changing tree of if then instructions. the translation book got more complex, but the Chinese room didn't change at all.

dan :

Isn’t the brain just a trillion tiny little Chinese rooms, what are neurons actually doing other than responding to chemical and electrical stimulation using if-then instructions.

acrch4 :

the man doesn't speak Chinese, but the room does.

icscata :

If AI (in its current form) truly understood what it was creating - especially in photo or video generation - it wouldn’t make mistakes like adding extra fingers, connecting teeth, misplacing background elements, misaligning shadows, or distorting facial features. The current version of AI doesn’t really understand what it’s generating; it simply imitates patterns of content that it thinks we want to see.

✦✧𝕍𝕠𝕩𝔽𝕒𝕟✧✦ :

FantasticMixterFox :

Translation based on an if-then book would either be very limited or it would have an insanely big book to compensate for lack of context. What makes a good translator is precisely the ability to comprehend both context in the narrow sense of co-text and all the knowledge of the world surrounding/underlying the utterances.

𝙇✦ :

Well that doesn’t matter because we speak Chinese 🗣️

Easton Tyler :

pdkxo :

The point of this famous thiught experiment is not whether he understands Chinese, but whether people talking to him can tell if their counterpart understands. And we can’t. Like AI to an extent.

StuieT :

I would say that the man doesn't know Chinese. The room (the collection of man, books, input and output) does.

williamedwards71 :

No. There’s absolutely a clear difference between regurgitating answers that simulate understanding and actually understanding. Llms fool people into thinking they understand a conversation by aggregating inputs and calculating a response most likely to satisfy the audience. But these are guesses and often contain hallucinations that the machine has no way of knowing are wrong. An effective agi can only be created if it can train on a whole world that’s similar to our own.

Wash Your Hands Fam :

“They are growing an adapting” is misleading phrasing, they themselves are not growing or adapting, new models are being trained yes but these models are not sentient or self replicating in any way

Louis Rovegno :

Clearly mimickry. True understanding means making choices about how to respond.

MissingNo. :

CyberErick :

Do we have a definition of „understanding“ that we can all agree about?

Ricardo Garcia Hered :

Arguably humans are just extremely complex if-then models, and so you could just wonder if we actually have free will or are systems trained by astronomically difficult models impossible to understand

Superhuman Unlocked :

Searle got it backward. The Chinese Room was never about proving machines can’t understand, it showed how we all function. Every day, we process inputs, follow scripts, and produce “authentic” reactions built from pattern and memory. Maybe AI doesn’t lack consciousness. Maybe consciousness was never what we thought, it’s the performance itself that makes meaning real.

Veeteeheehee :

This is the difference between learning to read using phonics (understanding) and learning to read using “sight words” (mimicry)

stary g :

you have to understand something to mimic it correctly

Eveleigh :

LLMs aren't "growing and adapting." People train new models, but each model produced is static.