Chinese room in Russian: "think" according to the 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 proposed a thought experiment for this , which became known as the “Chinese Room”.

In June 2015, as part of a virtual robot competition , Nanosemantics and the Skolkovo Foundation plan to show artificial intelligence technologies available in Russia, while they will practically reproduce John Searle's experiment.


About Chinese Room

Critic Searle directed to Turing test. John Searle suggested introducing a person who does not speak Chinese and is locked in a room with a small slot for letters. The person has baskets with Chinese characters and a textbook with instructions in a language he knows that will help translate from Chinese. Through a slot in the door he is given pieces of paper with a set of Chinese characters. A man can use the textbook to translate phrases and send an answer in Chinese.

Although the man himself does not speak Chinese ( and some intelligence can be a question ?! ), he can convince being outside that is fluent in Chinese.

About Chatbots Contest

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

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

About templates

The approach to creating artificial intelligence based on templates has an obvious advantage (compared, for example, with chatbots on neural networks).): It is easy for creators of robots to understand, who can quickly create their own chatbot, able to imitate a meaningful dialogue for a limited time.

Below is an example of templates for the question about the name of the mother of the robot
* {like / kag} * {I call ~ / zavu ~ / call ~ / call ~} {your / tfayu / tvayu / tfoyu} {mom ~ / mother / mamk ~} *
* {mom ~ / mother / mamk ~} * {your / tfayu / tvayu / tfoya} * {like / kag} * {call ~ / zavu ~ /
call ~ / call ~ } * * {call ~ / zavu ~ / call ~ / call ~} * {your / tfayu / Tvoy / tfoyu} * {mom ~ / mother / mamk ~} * {like / kag} *

You can chat with one of the competitive robots, which have approximately 1,000 templates, by clicking on this link (until May 20, 2015 inclusive) .

Below is an example of dialogue with this robot
Robot: I'm Sonya Guseva, hello!
Guest: Hello
Robot: can I ask you something?
Guest: Of course, ask the
Robot: who are you on the zodiac sign?
Guest: I am Scorpio, and you?
Robot: I'm a maid
Guest: When were you born?
Robot: in 2002
Guest: Very interesting! What language are you written in?
Robot: why are you interested in this?
Guest: Because I am a programmer, and I want to develop in this area.
Robot: is this the real reason?
Guest: Yes
Robot: well , ok :)))
Guest:Why didn’t you answer?
Robot: just like that :)))
Guest: You probably don't understand what I'm talking about
Robot: do you understand me?

The ability to maintain a short dialogue in English was previously demonstrated by the robot Evgeny Gustman, who, operating about 3,000 templates, made about a third of the judges of the Turing test think that they are communicating with a living person. The knowledge base of some modern chatbots that can communicate on a wide range of topics contains more than 10,000 rules.

Robots of Nanosemantics (info), built on "template thinking", previously communicated in ICQ, while the average length of a person’s conversation with info was several hundred replicas. But, as John Searle argued, lengthy dialogs do not prove the ability of computer programs to think and understand .

Also popular now: