Dev.to AI 🤖 Ai 👁 0 📖 5 min read

Knowing Is Not Understanding: Why Searle Still Wins Against Today's AI

In today’s society, AI has taken computing to a whole new level. Sure, it has been in development for decades, but it has never advanced this rapidly. With its place being solidified in society, it definitely sparks a lo

In today’s society, AI has taken computing to a whole new level. Sure, it has been in development for decades, but it has never advanced this rapidly. With its place being solidified in society, it definitely sparks a lot of discussion about it and its role here on. Truthfully, discussions concerning AI date back to the 1950s, built on the foundation of the Turing Test, and later the Chinese room. Is AI actually thinking and understanding, or is it just a mimic? Personally, I lean toward Searle’s view. AI has the capability to replicate patterns but does not actually “think”. Instead, AI can recognize and utilize them, but it cannot truly understand the semantics behind them. Passing the Turing test can show a machine has the knowledge about the world, but not the experience to understand it. Searle’s view aligns more with current LLMs, that understanding is not possible for machines as they lack experience.

Turing (1950) proposed replacing the question “Can machines think?” with the imitation game. In this game, an interrogator poses questions to a hidden human and a hidden machine and tries to determine which is which. If the interrogator cannot tell them apart, then Turing supposes that the machine should be considered capable of thinking (Dowe & Oppy, 2021). Searle (1980) challenges this with the Chinese Room argument. He describes a man who does not know Chinese sitting in a room, using a rulebook to match Chinese symbols to the appropriate replies. This makes people outside believe that the man is a native speaker. According to Searle, a computer running a program operates in the same way by manipulating syntax without understanding the semantics. He argues that syntax alone is not sufficient without semantics, and by that, a machine running a program is not the same as a machine understanding.

Combining Turing and Searle’s arguments into one question given our current knowledge on AI, we could derive another question, “Can something that has never cried differentiate tears of joy from tears of sadness, or does it only know how people describe them?”. In my opinion, an LLM can know what tears are; it has the research and resources available in its dataset for training. AI has progressed to the point where AI can perceptively classify feelings of sadness or happiness from an image of a person’s face. Humans also use their sense of sight to perceive if a person was crying, learned behaviour and mannerisms from knowing a person can also determine that. Everyone has gone through a rough patch in their life, including me, where they can no longer hold in the tears; I have personally been reduced to tears purely through exam season at college. A person can upload a photo of their face to ChatGPT and ask whether the expression shows sadness or happiness, and it will tell them. This raises the question however; can AI relate an experience to share in which they were sad? A personal experience, something that has not been pulled from the massive data on the topic of sadness and human emotions? A personal anecdote that can give a sense of shared connection knowing that someone has gone through the same? It does not actually know your experience other than what is the textbook case of it, some examples that seem like it understands it enough, and its ability to regurgitate that information in different ways. As in Searle’s argument, AI can read the syntax it is given in and identifies the response that is most likely to fit at that time.

The Systems Reply concedes that the man in the room does not understand Chinese, but argues the man is a part of a larger system at hand (Cole, 2026). This view holds that understanding belongs to the system as a whole, not solely the man. Searle responded that the man could memorize the rules and rely on his head; in that way he could be considered the entire system, yet he would still not understand Chinese. In my opinion, the man would still not be able to understand Chinese solely because he has not experienced what it means to live the language. If he were to go to China and talk with different generations of Chinese, he would have to navigate through dialects that exist across China. Cantonese sounds different to Standard Mandarin, Mandarin sounds different to Hakka, not to mention the difference in formality speaking with elders versus when speaking amongst your peers. Knowing the symbols of the Chinese cannot equate to keeping up with slang and getting context to the same words being applied to a different meaning. That is what separates knowing and understanding; even as the entire system, he has not experienced what it means to live the language, and neither has the system.

I think Searle is correct in the sense that AI is convincing as a machine that seems to think for itself, but it does not truly think. It lacks the experience humans gain throughout life which is what allows them to truly understand. AI can present as human-like and convince some people, especially those who may crave more of a connection, but at the end of the day it is an imitation, not true understanding, a concern raised against the Turing Test since Turing’s own time (Dowe & Oppy, 2021). Making the system “bigger” does not give it experience, since understanding comes through the experience of using it. To bridge that gap, a machine would need time to live and grow, just as a child grows, learns right from wrong and becomes immersed in a culture. With AI developing at a fast rate, there is always room for that to change, and there will always be more to come from the field in the future. Until then, AI remains an imitation.

References

Cole, D. (2026). The Chinese Room Argument. Stanford Encyclopedia of Philosophy (Summer 2026 Edition). Retrieved from https://plato.stanford.edu/archives/sum2026/entries/chinese-room
Dowe, D., & Oppy, G. (2021). The Turing Test. Stanford Encyclopedia of Philosophy. Retrieved from Stanford Encyclopedia of Philosophy: https://plato.stanford.edu/archives/win2021/entries/turing-test/
Searle, J. R. (1980). Minds, Brains, and Programs. Behavioral and Brain Sciences, 3(3), 417-457.
Turing, A. (1950, October). Computing Machinery and Intelligence. Mind, 59(236), 433-460. Retrieved from https://www.jstor.org/stable/2251299

📰 Read the original article on Dev.to AI

Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.