It is almost impossible now to talk about cutting-edge technology without thinking about artificial intelligence AI, a technology that is increasingly becoming a part of everyday life. Nearly all governments have realized the importance of having the most sophisticated AI system possible and do not hesitate to fund research in this area by hundreds of millions.
So far, advances in this area have not yet created AIs capable of holding conversations and this is a limitation that can be seen in the virtual assistants present in many mobile devices and smart devices for the House. But it seems that we are no longer so far from achieving this goal in view of the progress that has been made recently.
The Allen Institute for Artificial Intelligence, a leading laboratory in Seattle, recently unveiled a new system called Aristo, which would have made a huge breakthrough in the field of artificial intelligence. Indeed, this system was able to succeed with a score of over 90%. This represents a major breakthrough especially when we know that four years ago, 700 computer scientists tried unsuccessfully to invent a system capable of passing such an examination. At the time, the highest score was a score of 60% correct answers. The test in question is a standard written exam for New York students.
Aristo was built solely for multiple choice tests and so this test only contained such questions. It was devoid of questions including images and diagrams, because answering such questions would have required additional skills that combine the understanding of language and logic with a so-called computerized vision.
Researchers at the Allen Institute built Aristo on a different technology developed by Google, called Bert. Bert was designed to scan thousands of Wikipedia entries on books and articles written by humans, and like other machine learning algorithms, she is able to improve. Although the system was designed specifically to pass tests, the researchers claim that it could have practical implications in search engines and other systems involving the processing of user queries.
While some find these results very promising for the future of AI, others think they do not represent much, because they believe that having removed from the test all visual issues, thus decreases the quality of the test and the implications that these results could bring.
Source : NYTimes