Five points to pay attention to in "artificial intelligence" of great success in various fields
ByDr J Clarke
I won a professional shogi player "AlphaGo"Automatic composition of jazz"Deepjazz",Pop song created with artificial intelligenceAnd 2016 was the year of development of artificial intelligence (AI) and machine learning. I think that artificial intelligence will continue to make a great breakthrough in 2017, so that five points that are likely to be particularly keyMIT Technology ReviewIs cited.
5 Big Predictions for Artificial Intelligence in 2017
◆ 1: reinforcement learning
AlphaGo developed by Google in 2016 will become the top player of govictoryWe accomplished a historical achievement to do. This seems to have been one major goal for the AI industry, especially reinforcement learning.
Its reinforcement learning was invented inspired by the fact that animals learn things from positive or negative consequences of certain actions. With this approach, the computer learns "How to act" with trial and error while taking positive results as a successful example.
MIT Technology Review predicts that in the year 2017, more cases will be used for reinforcement learning in the automatic driving car and robot industry. actually,Google's efforts to more efficiently manage data centers using deep-level reinforcement learningWe are doing. However, the reinforcement learning approach is experimental, and it seems that simulation still takes time.
◆ 2: Hostile Network
When Neural Information Processing Systems (NIPS), an international academic society of AI in Barcelona was opened, it seems that the hostile generation network (GAN), which is a new machine learning technology, was proud of its great popularity.OpenAII am doing AI research by Dr. Ian Goodfellow invented by GAN is a single network generated using new data after learning from the training set, a network that attempts to distinguish between correct data and fake data . Using this GAN seems to be able to create very realistic synthetic data.
Joshua Benjo, the foremost expert on AI research, said in the NIPS that "GAN is a particularly exciting approach in terms of" learning computers using unclassified data ".
◆ 3: AI Boom in China
In 2017 it may be the year that China will evolve into a major player in the AI industry. China's technology industry is changing from the stage of copying technologies of European countries and the United States to the stage of innovation using AI and machine learning. China's search engine Baidu has opened a laboratory to focus on AI and, as a result, benefits from various fields such as speech recognition and natural language processing.
There are also rumors that chasing Baidu, Didi has purchased Uber's operation rights in China, and building a laboratory for the development of automatic driving cars. Many other Chinese investors are investing in AI related startups, "There are many people who desire the boom of the AI industry in China," MIT Technology Review. There is also speculation that China's AI industry will reach $ 15 billion (about 1.7 trillion yen) by 2018.
◆ 4: Language learning
"If you ask AI researchers" What is the next target? ", MIT Technology Review says," Many people will answer "language learning." Language learning is a technology that can be used for speech recognition and image recognition, etc. It also leads to an attractive theme called "conversation with AI".
If the computer can understand the language more correctly, the machine will definitely become more convenient. However, that challenge is also very difficult. Therefore, it seems that it will be a while before it becomes possible to have deeper meaningful conversation as AI on smartphone.
◆ 5: The reaction to the hype
A pure progress of AI and a lot of exciting applications making use of it have appeared in 2016, but in the same way a lot of hype about AI has appeared. In these circumstances AI researchers are obviously irritated. As a matter of reality, it is also pointed out that there are people who adopt AI due to hype advertisement but fail to achieve the expected results as expected, and "disappoint" for AI as a whole.
in Note, Posted by logu_ii