The goal of artificial intelligence thought by Google's Eric Schmidt


Google's Eric Schmidt chairman talks about the future of his thinking AI on "Intelligent Machines" which is a series related to BBC's artificial intelligence (AI).

Intelligent machines: Making AI work in the real world - BBC News

The word "artificial intelligence" was born on August 31, 1955, 60 years ago, and it seems that John McCarthy first appeared in the research plan proposed for developing electronic brains. For decades after the birth of a word, AI seemed to be a fictitious thing that exists only in SF. However, in recent years it has evolved explosively and is trying to blend in various parts of everyday world.

ByJD Hancock

One of the reasons for the explosive evolution is "deep learning". Normally, innovation will bring very hard research results leading up to today. For example, Jeff Hinton, one of the authorities of an artificial neural network, seems to have came up with several important insights in the 1980s. However, at that time the computer was not as fast as it was, so we could not shape its important insights.

Since the 1980's Mr. Hinton has been studying artificial neural networks for 20 years. And in 2009, we are introducing state-of-the-art speech recognition technology developed based on years of warming insight. According to the circumstances that it was possible to develop this, it was said that deep learning became a level that can actually be used by operating thousands of computers at the same time or improving the processing speed of one computer It also has a big influence.

In addition, Google immediately adopted the method which Mr. Hinton announced, it seems that 25% cutting error occurred in Google voice recognition application.

Furthermore, incorporating it into the real world problem will accelerate the evolution of AI. The latest system demanded by consumers such as "speech recognition corresponding to 58 languages" can be replaced with a system that focuses AI's basic research on a new field. These are much more difficult than AI related studies that have been done over the past decades, but they are challenging, Chairman Schmidt said.

Therefore, Schmidt employs several talented researchers from the university every year at Google. In addition, he commented that Google researchers have let their studies open to the public. Because the real world is so big and cluttered, AI needs to learn a lot, but it is because we believe that an approach from the outside will also become necessary to overcome it, so we can open a variety of data It is said that it is doing.

ByGuillaume Paumier

"From now on, we need to conduct harmonized AI research," Schmidt said. With next generation software, machine learning becomes one of just add - ons and it may be an additional function to improve performance by a few percent. For example, digital music service started over ten years ago, but users should have used some kind of functions and services to search for new music. However, with the latest music service, the system is evolving more intelligently, so that users can use 'what kind of music you want the next music to play' or 'what genres of music do you like?' Schmidt said, evolving to the point where it can be estimated and suggested automatically according to the situation.

Schmidt said that in order for AI to expand the possibilities of society over the long term, "A best plan for going on a trip for a few days with two children in a limited budget" and "A smarter filter to remove noise around us, from e-mails and SNS to our own schedule", "help scientists who need to process enormous amounts of data such as genes and energy, climate It is necessary to keep thinking on the needs of the real world first, such as "system like".

in Science, Posted by logu_ii