The historic Lee Sedol vs. AlphaGo match took the entire nation by storm, thankfully giving much-needed impetus to the government and companies’ efforts to develop AI technology.
S. Korean President Park Geun-hye said on March 17 that she would create and preside over a strategic council on science and technology that will oversee the nation’s research and development (R&D) sector. In addition, the government will join hands with six large companies to set up an AI research center in order to narrow the gap between the nation’s AI technology and developed countries.’ For now, the highly-publicized human vs. AI Go faceoff is serving as a wake-up call for the government and companies.
As a matter of fact, AI had been applied to various fields before it took on the game of Go. When it comes to software powered by AI, global IT companies have been making progress in image and voice recognition and knowledge processing by exploiting deep learning.
As for facial recognition, Google is working on a captioning system called Neural Image Caption (NIC) while Facebook in 2014 developed facial-processing software “DeepFace” that can match faces with 97.25 percent accuracy. As a comparison, humans generally have an average of 97.53 percent accuracy. This means that DeepFace has nearly the same accuracy as humans.
Furthermore, China’s Baidu in 2014 developed a supercomputer called “Minwa,” which has an error rate of 5.98 percent, lower than Google’s 6.66 percent error rate.
In voice recognition, Microsoft's Cortana personal assistant, Google’s Now and Apple’s Siri reign supreme. In the knowledge processing area, IBM’s deep learning-based Watson, takes the lead. In 2011, IBM’s Watson beat Ken Jennings and Brad Rutter on the TV quiz show Jeopardy!.
Breakthroughs in deep learning have something to do with rapid advances in the performance of computers and the advent of big data technology, which supports the processing of large and diverse data sets.
However, only high-performance supercomputers are capable of deep learning, so deep learning-enabled computers are heavy and bulky and thus have limited mobility. As for bringing deep learning to smartphones, smartphones will receive the output after they are connected to a remote server operating the software and the input goes through pattern analysis and classification. As a result, making smartphones capable of real-time deep learning could be a challenge.
Above all, unlike humans, computers understand our language using statistics, so it is hard for them to answer questions including the word “not.” To develop AI that is exactly the same as the human brain, petaflop supercomputers should be upgraded into exaflop (about 1,000 petaflops) ones. Such an upgrade necessitates several hundred megawatts of power. And due to a technical limit to the development of semiconductors, upgrading computers to such levels could be impossible in the next decade.
On the hardware front, Nvidia took the wraps off its Drive PX 2, the world’s first in-car AI supercomputer for self-driving cars, at CES 2016. It's the size of a school lunchbox. The new Drive PX 2 can accurately understand the full 360-degree environment around the car to produce a robust representation, including static and dynamic objects.
The computational capability of the Drive PX 2 is roughly the same as 150 MacBook Pros, allowing for up to 24 trillion deep learning operations per second.
Meanwhile, neuromorphic chips, i.e. microprocessors configured more like the human brain, are emerging and increasingly being applied to many fields including manufacturing.
S. Korea is playing catch-up with developed countries in AI.
In AI technology, the gulf between S. Korea and developed countries including the US, the UK, Japan and Germany is huge. In 2008, the US embarked on government-led research on AI. In 2014, the US developed an AI chip that is nearly as intelligent as a honey bee. Its image recognition accuracy topped out at 80 percent. Under the Seventh Framework Programme (FP7) of the European Commission, the EU also developed an AI chip, comprised of 262,000 neurons cultured on silicon wafers, which was known to be approximately 10,000 times faster than the human brain.
On the corporate front, IBM has been participating in the DARPA-funded SyNAPSE program since 2008. The company has said that a new kind of cognitive computer with similar form, function, and architecture to the human brain will come out in five years. In June 2013, Intel's investment arm allocated 100 million dollars to the Intel Capital Experiences and Perceptual Computing Fund; Intel unveiled its patter-recognition technology based on an AI chip. Since 2009 Qualcomm has been working on a project called "Zeroth," a neuromorphic chip program for vision sensors, robot controllers and cognitive computing (CC). HP is another SyNAPSE contractor. Nearly 90 HP researchers have been committed to the DARPA-funded SyNAPSE program.
Korean companies are also aware of the importance of AI chips. Since 2011 the Samsung Advanced Institute of Technology has been working on a small-scale AI chip project to develop image-recognition technology. SK Hynix is also working on an industry-academia project. However, due to the nation’s endemic corporate culture of seeking profits that are near at hand (attainable within 5 years) and playing the role of a fast follower, a government-led, long-term R&D plan is badly needed in order to develop high-risk, high-return and high-patience technologies like AI.
It is true of the developed world. Companies’ projects to develop basic and fundamental technologies tend to fizzle, so the governments of developed countries lead and back those projects in the long term. In October 2015, the White House has announced a Nanotechnology-Inspired Grand Challenge for Future Computing; each department of the US government is set to draw up an action plan for the initiative.
In Feb. 2013, the EU kicked off the Human Brain Project (HBP), a large ten-year scientific research project into which 1 billion euros will be ploughed between 2013 and 2023. The goal of the HBP is to simulate the human brain in a supercomputer and develop an AI chip that has the same numbers of neurons and synapses as the brain of a cat.
Releasing the Fifth Basic Science and Technology Plan for 2016-2020, Japan is poised to concentrate on the concept of Society 5.0, aimed at building a ‘super smart society.’
Israel has drawn up a national strategy to position Israel not only as the “Startup Nation,” but also as the “Brain Nation” and is planning to build nearly 20 AI research centers.
The S. Korea government is gearing up to implement the 4th nanotechnology development plan for 2016-2025, designed to channel national resources for developing strategic technologies into one field.
Since 2013 the S. Korean government has invested in AI projects, such as the Exobrain project and the development of visual intelligence software based on real-time image recognition. Its investment in hardware has been made mainly in developing nanomaterials and future-oriented pioneering projects on a small scale.
To close the gap between our AI technology and developed countries’, we have to drastically get rid of regulatory hurdles and craft policies that spur on the development of AI technologies.
