IBM Watson supercomputer learned street lingo: it was necessary to clean its memory

The developers of the IBM Watson supercomputer made a big mistake when they downloaded the street jargon dictionary from the Urban Dictionary site into its memory. IBM Watson is trained to recognize the meaning of suggestions / questions and respond to them using assimilated arrays of unstructured data (data mining). Thanks to this, the computer won the champions of the game show “His Game” (Jeopardy) in 2011, after which he began to study medical reference books in order to diagnose diseases by symptoms, like Dr. House.

The developers thought that information about additional meanings of words would improve artificial intelligence, help it better understand people. They were wrong. Street jargon proved to be extremely harmful to the supercomputer. Once he even answered the question with the indecent word "Bullshit".

As a result, the developers decided to clear the memory of IBM Watson from slang and put a filter on the output words, in order to prevent accidental profanity on the air.
')
Responsibility was taken by one of the project staff, Eric Brown (Eric Brown). In the photo, he is depicted inside the IBM Watson cluster. By the way, the IBM Watson development team consists of 35 people.



In an interview with CNNMoney, Eric Brown says that the problem arose from the fact that the computer could not determine which meaning of the word is decent and which is not. Because of this, he sometimes used words in an indecent sense.

Eric Brown says that IBM Watson also has some bad habits after reading Wikipedia. That is why the removal of the Urban Dictionary was not enough, and I had to put an additional dictionary filter. In the future work of IBM Watson, as a diagnostician, a vocabulary like OMG or LOL is completely out of place.