Chinese room in Russian: “think” according to a pattern

In the 1980s, the American philosopher John Searle decided to refute the assumption that computers or other types of artificial intelligence can think and understand, and suggested a thought experiment for this, which became known as the “Chinese Room”.

In June 2015, within the contest of virtual robots , Nanosemantika and the Skolkovo Foundation plan to show the artificial intelligence technologies available in Russia, and they will practically reproduce the John Searle experiment.


About the Chinese room
')
Searle's critique was sent to the Turing test . John Searle proposed to introduce a person who does not speak Chinese and is locked in a room with a small letter slot. A person has a basket with Chinese characters and a textbook with instructions in a language he knows that will help translate from Chinese. Through a slit in the door he is handed a piece of paper with a set of Chinese characters. A man can use the tutorial to translate phrases and send the answer in Chinese.

Although the person himself does not speak a word of Chinese ( and what kind of intelligence can we talk about ?! ), he can convince those who are outside that he is fluent in Chinese.

About chatbots contest

In the competition, robots will have the task of convincing the judges that they can speak Russian. In the Chinese room in Russian, the judges will communicate with the robot in a text chat for 3 minutes. Instead of the Chinese characters, the robot will have the words of the Russian language, and instead of instructions in English - special rules described by the creators of chatbots. Judges will need to evaluate how the responses of the robot are similar to human ones.

Only robots based on the same technology will participate in the competition. This technology involves the description of a set of templates with which the replicas of the interlocutors are compared, and the corresponding reaction options that the robot returns as answers. This approach does not imply independent learning of the robot during the conversation. In this case, artificial intelligence does not analyze the semantics of texts, but acts as a person from the Chinese room.

About templates

The approach to creating artificial intelligence on the basis of templates has an obvious advantage (compared to, for example, chatbots on neural networks ): it is simple to understand by the creators of robots who can quickly create their own chatbot, capable of simulating meaningful dialogue for a limited time.


You can communicate with one of the competitive robots, which has about 1,000 templates, by following this link (until May 20, 2015) .


The ability to maintain a short dialogue in English was demonstrated earlier by the robot Eugene Gustman, who, using approximately 3,000 templates, made about a third of the Turing test judges think that they were communicating with a living person. The knowledge base of some modern chatbot, able to communicate on a wide range of topics, contains more than 10,000 rules.

The Nanosemantics robots (info), built on “patterned thinking”, previously communicated in ICQ, while the average conversation length of a person with info was several hundred replicas. But, as John Searle argued, lengthy dialogues do not prove the ability of computer programs to think and understand .