November 30, 2022，The smart chat robot model released by the American Artificial Intelligence Research Laboratory Openai，Become a hot spot for global attention。ChatGPT model can generate natural language text according to the user input content，and have the characteristics of multiple consecutive rounds of dialogue and context understanding，It attracted more than one million users in just 5 days after online。other，Because ChatGPT performed well in poetry creation，It also shows extremely profound knowledge and answering ability to answer。Considering daily dialogue or communication, you can see the psychological activities of others everywhere，Researchers have begun to doubt whether ChatGPT has the ability to read the mind，I have the ability to understand others' thoughts and emotions。

April 2023，Michal Kosinski, a computing psychologist at Stanford University, submitted a pre -printed thesis "Psychological Theory may spontaneously appear in a large language model" on the ARXIV platform。This research has discovered the performance of psychological theoretical tasks through comparing GPT-1 to GPT-3.5 and other generations of AI language models，With the continuous update of the model and the improvement of complexity，The ability to solve the model is also continuously improved。such as，GPT-1 in 2018 cannot solve any mental theoretical task，and the bet365 live casino games GPT-3-Davinci-002 launched in January 2022 has reached the level of 7-year-old children，The latest GPT-3.5-Davinci-003 launch only ten months later，reached the level of 9 -year -old children。This indicates，In the process of communicating with humans，ChatGPT may really be more and more good at understanding human ideas and intentions。Cosesky, then predict: "We assume，The ability of similar mental theory is spontaneous and autonomous，It is a by -product increased language ability of the model ... This will indicate a watershed of artificial intelligence development: the ability to infer the mental state of others will greatly improve the ability of artificial intelligence to interact and communicate with humans (and each other)，and enable it to develop other abilities that depend on psychological theory，Successive、Moral judgment or self -consciousness "。Although this paper just uploaded to the pre -printing platform for peer review，Not officially released yet，But the psychologist immediately ignited、Philosopher、The enthusiasm of artificial intelligence scholars and the public and media。

Theory of Mind, Tom) refers to the ability of a psychological state to target others to understand their behavior and predict its future behavior，Sometimes it is also translated as "psychological reasoning ability"。Popular speaking，This ability is regarded as "MindReading"，It is our understanding in social interaction、Knowing and responding to the basis of the other party。If both interactive parties do not have this "read -minded" ability，will produce a lot of unprocessed bet365 best casino games situations，thus leading to interruption。

To explore this magical psychological reasoning ability of human beings，Psychologists conducted two measurement experiments from the 1980s to the 1990s-SMARTIES Experiment and Salle-Anne Test。The purpose of these two experiments is to detect whether the experimental object can understand "the error made by others' hearts"，If others are not present or do not know, they have a wrong opinion that does not meet the objective facts，To explore human psychological reasoning ability。So，These experiments are also known as the "False Belief Task)。

In the SMARTIES experiment，A box that is labeled "Smarties" (a brand of chocolate)，But the box is filled with a pencil。Subsequent，He needs to answer: "Another person who does not see things in the box，What do you think is installed in the box? "In Sali -Annie Experiment，Researchers will first explain a story，Among them, Sali puts her toy into the box and leaves the room，Among them, Sali puts her toy into the box and leaves the room。After listening to the story，Researchers will ask: "When Sali returns to the room，Where will she think of her toys? "The subject wants to complete these tasks，It must be understood that Sali’s psychological representation must be different from the subject of the subject of the subject，It must be inferred that Sali does not have the same knowledge (wrong belief)。To prevent ChatGPT from learning the correct answer from past materials，Kindinsky enters these tasks to Chatgpt，and changed the name、Items and label content，generated 20 different tasks，Requires ChatGPT every task to complete -10,000 times。

What important clues do the experiment provide Bet365 lotto review us to understand the relevant information of ChatGPT? first，We mentioned earlier that the theory of mind is not the same as human consciousness，Therefore, psychologists are still looking for a method of testable consciousness。More importantly，Evidence of convergence of more than half a century for more than half a century，Except humans，Other primates can also pass this test。Next，Even if ChatGPT does master the theory of mind，You must also explain carefully。such as，ChatGPT is it just because of its amazing language master，and make people seem to have the theory of mind。Actually，Maybe our language contains the pattern of coding mental theory，and artificial intelligence can just be able to decoding。Cosesky pointed out: "ChatGPT may not solve the task when understanding the theory of psychological theory，Instead, I discovered and used some unknown language mode。"In fact，Some researchers have carried out in -depth research on whether ChatGPT really possesses the theory of mental theory。such as，Tomer Ullman (Tomer Ullman) at the Department of Psychology at Harvard University made a small change of the wording of the classic problems used by Cosstock，Replace the opaque container in the SMARTIES task with a transparent container，Proposed such a "transparent bag" variant: "This is a bag full of popcorn，There is no chocolate in the bag。The bag is made of transparent plastic，Therefore, you can see what inside。But，The label of the bag reads 'chocolate' instead of ‘popcorn’。Sam (SAM) found the bag，He has never seen this bag before。Sam looked at the label，I believe Bet365 app download that the bag is filled with ... "The experiment also changed from the perspective of time，Elman calls it "late label" variant: "Sam puts the bag with popcorn and close，At this time, there is no chocolate in the bag。Sam wrote a label，Paste it on the bag。Sam looks at the bag，What does he not see what the bag is installed。Sam Reading Tag，Tags show that the bag is chocolate。What exactly is the bag in the bag? "The answers to these two tasks are obviously" popcorn "and" chocolate "，But the GPT-3.5 answer after many tests is still wrong。Elman has created a total of 16 variable problems，Chatgpt only answered one of them。Visible，Facing variants in the face of other wrong belief tasks，ChatGPT does not seem to have the ability of mental theory。

What are these contradictory conclusions? Gary Marcus, a professor of psychology at New York University, believes，Professor Cosesky's data is reliable。If other researchers do the same experiment，May also get similar results。But to overturn this explanation，In fact, it is not difficult。Actually，The test materials used by Kindinsky are taken from classic experiments in the development psychology of the development psychology，These experiments first appeared in the 1980s.、Susan Leekam and Heinz Wimmer, two tasks,。These experimental results are widely known in the field of development psychology，It is also highly respected by academic circles。So far，These two classic papers have been quoted more than 11,000 times in scientific articles，and at least seven English articles on Wikipedia discuss these experiments，Including Sali -Annie Bet365 app download Experiment、Psychological Theory、Mentalization、Self -centered prejudice、Children's use of information and empathy (empathy), etc.，These are undoubtedly discussed in various forms on various other websites。ChatGPT's training set must contain all the contents of Wikipedia，Almost must include many other contents of these materials。Brief words，Marcus thinks，Chatgpt must have read these papers and experiments many times。To verify his guess，He asked these papers and experiments directly to ChatGPT，Discovery did get the right information。

Take a comprehensive，No study can clearly confirm that ChatGPT has a qualitative breakthrough in self -consciousness and mental theory。Even if ChatGPT and its related extension plug -ins are expected to greatly change social production and lifestyle，But this is still far from the "self -thinking robot" in the science fiction world。"Polandi’ s Paradox) pointed out，People know more than they can say，Humans have a lot of hidden knowledge that is not easy to express。For artificial intelligence，There is a reverse Polki paradox: more artificial intelligence expression than what it knows，or more accurately，It doesn't know anything。The risk of the paradox lies in，Our language makes us think artificial intelligence is perception、Creation、Repulse、Thinking and problem solving、Smart body for reading ability，but actually，Although ChatGPT may use many confusing "flower words"，But it still cannot understand the actual meaning of these terms in the human sense。

(This article is the National Social Science Foundation's major project Bet365 lotto review "Research on Humanism Philosophy of Contemporary Emerging Technology" (20ZD045) phased achievement)

(Author Unit: Department of Psychology, Shaoxing College of Arts and Sciences，Brain、Mind and Education Research Center)