Seven Times Faster than Moore’s Law
According to the AI Index 2019 Annual Report, published by Stanford University Human-Centered Artificial Intelligence Research Institute in collaboration with international consulting group McKinsey, etc, AI performance improved in the 2010s. The speed is analyzed seven times faster than Moore’s law. Moore’s Law was first invented by Intel researcher Gordon Moore in the 1960s, and it is a law that doubles the performance of a computer chip every two years.
According to the report, the speed of artificial intelligence development was almost similar to Moore’s law before 2012. But since then, the acceleration has doubled in 3.4 months. If it had been according to Moore’s law, it would have been only 30 times the 300,000 times.
This has dramatically reduced the time it takes to train AI algorithms. The time it takes to train a large image classification system,ImageNet, on a cloud basis has been reduced to one-180th in two years. What took three hours in 2017 only takes 88 seconds as of July 2019. The cost also dropped sharply from thousands to tens of dollars. Image recognition accuracy has also increased significantly. According to the report, ImageNet’s identification program, a public dataset with more than 14 million images, showed 85 percent accuracy. This was a sharp jump from 62% in 2013. In machine translation, the number of machine translation systems available for real-world work has increased from eight in 2017 to 24 in 2019.
Another important indicator of the arrival of the AI era is the change in the number of AI-related jobs. The report analyzes hiring information on the employment information site ‘Indeed’ during September 2010 ~ 2019, and found that the share of AI jobs in the US has increased five times since 2010. The share of all jobs rose from 0.26% to 1.32%. It is still small, but it should be taken into account that only the technical sectors directly related to AI development are counted. Jobs that are strengthened and reorganized under the influence of artificial intelligence can add up, the report said. The report was analyzed only in the United States, but it is likely that other countries, including Korea, will have a similar trend.
The leading segment of job postings among AI tech jobs was machine learning (58% of AI jobs). This was followed by artificial intelligence (24%), deep learning (9%), and natural language processing (8%). The fastest growing job is deep learning. 12-fold increase between 2015 and 2018. The number of artificial intelligence increased five times, machine learning four times, and natural language processing twice. As demand increased, repairs soared. According to the UK venture capital fund MMC Ventures, the average annual salary for artificial intelligence engineers (US 2018) is $ 224,000 ($ 260 million), and the average software developer is $ 10,480 ($ 120 million). More than twice.
Investment in AI-related startups is also growing at an annual rate of 50%. In 2018, it exceeded 40 billion dollars (46 trillion won). The area with the most money is in the autonomous vehicle sector. It reached $ 7.7 billion in 2018. Medical research and facial recognition took second place with $ 4.7 billion. The highest growth areas were robot automation ($ 1 billion) and supply chain management ($ 500 million).
The history of confrontation between AI and the best human beings shows the growth of AI most easily. In a series of confrontations that began with the 1997 supercomputer Deep Blue defeating chess champion Garry Kasparov, AI has made a mark in the last decade with the advent of deeper power. Beginning with the 2011 quiz show “ Jefferdy, ” 2015 Atari Games (Brick Break), 2016 Go, 2017 Skin Cancer Diagnosis, 2017 Voice Recognition and Poker, 2018 Chinese-English Translation and Online Game “ Dota 2 ” In addition, he has demonstrated the ability to subdue or equalize the top-ranked human beings without any time to hide from the synthesis of protein to six-player poker and StarCraft 2 in 2019.
However, most of the outstanding artificial intelligence has limitations that can only be used in the field. Artificial intelligence, no matter how good at StarCraft II, is a novice player in front of the chessboard. Artificial intelligence to diagnose breast cancer is difficult to accurately diagnose lung cancer. Just because you have superior computing, analysis, and reasoning skills is a risky task. After all, the power of AI depends on how it is used for its proper purpose. This means that the stronger the AI, the more decisive the human judgment can have. That’s why human value still needs to be emphasized in the AI era.