AI's originator, Jordan, poured cold water on the "artificial intelligence" on the vent

At the 2nd International Conference on Intelligent Science with the theme of "Development and Application of Intelligent Science", Michael Jordan, known as the originator of the AI ​​community, came to the scene and poured cold water on the " artificial intelligence " on the ventricle.

“In the next decade, the 'intelligence' of artificial intelligence systems is still very limited, and you don’t think it can be as smart as humans. I think these AI systems will not be as flexible as humans in the next decade. Sex and creativity," Jordan said.

Together with Geoffrey Hinton, the originator of deep learning, Jordan is considered to be two “root-level” figures in the field of artificial intelligence. His students have the authority of deep learning, Yoshua Bengio, Bayesian field of study authority Zoubin Ghahramani, and former Baidu chief scientist Wu Enda. and many more. He is currently a professor at the University of California at Berkeley and is a Distinguished Professor in the Department of Electrical Engineering and Computer Science and the Department of Statistics. He is also an academician of the American Academy of Sciences, the American Academy of Engineering, and the American Academy of Arts and Sciences. He is the only scientist in the field of machine learning to achieve this achievement. He has pointed out the connection between machine learning and statistics and has pushed the machine learning community to recognize the importance of Bayesian networks. He is also known for his work on the formalization of the inductive variational method and the maximization of the expected algorithm in the popularization of machine learning.

When it comes to artificial intelligence, the first image in many people's minds is usually a robot. Just like the intelligent robot in the movie "I, Robot", humans can interact intelligently with it. Therefore, some people think that artificial intelligence refers to progress in this area. It can communicate with you and even take care of your food and clothing.

But Jordan believes that such a robot can't be done in the short term.

From a computer vision point of view, computers have now been able to accurately identify specific objects in complex images. However, computers currently lack a common understanding of visual scenes. For example, if a person approaches the edge of the stage, humans will feel that he is likely to fall off the stage. "Humans can judge from the scene what will happen next, and why the current scene will appear. But the computer is still far away. There is no ability to achieve this."

From the perspective of speech recognition, the current mutual conversion from speech to text has been successfully applied in many languages. But the computer's hearing ability is still very limited. For example, if you close your eyes and feel the surrounding environment just by hearing, you can know whether you are in a quiet park or a bustling street. You can also infer the orientation of people and things around you based on your voice. There is also a lack of common-sense cognition in this category. If complicated language information is added, it will be even more difficult."

Natural language processing is the most difficult for computers. "We have seen a great progress in machine translation, but it still misses many details in the language." Jordan said that the current neural network technology used in machine translation can calculate and match a large number of different language data. But the way humans learn language is very different from computers.

For example, the question and answer system, the current question and answer system research can only answer some questions with clear conditions and simple answers, and can not make complex answers to complex questions in the real world question and answer scene. The semantics of people's language is complex and diverse, with synonyms, synonyms and antonyms. A phrase may contain multiple meanings in different language scenarios. The expressions and habits of different languages ​​are different. “For humans, we learned how to distinguish complex contexts during the learning process from small to large, but computers are far from being able to do this.”

In addition to the perspective of robotics, at present, robots that are being used in industry can only perform some fixed tasks programmatically, which is quite different from the artificial artificial robots that people imagine. "Robot science helps to realize artificial intelligence research." The ultimate vision – we hope that in the future, artificial intelligence robots will be able to operate autonomously and interact with us.”

Jordan predicts that some intelligent applications may indeed become a reality. For example, in the next 10 years, autonomous vehicles and even unmanned taxis are possible. "Although the current experience of these technologies is not very good, it can be expected that these cutting-edge technologies should be available to people in the next 10 years."

As for the understanding of the AI ​​system in the process of human-computer interaction, whether it can achieve advanced intelligence such as forecasting, planning, etc., Jordan believes that it is still very far from this step, "it will take at least several decades. Even hundreds of years can make robots understand humans."