What's next after AlphaGo?
.jpg?w=728)
Tech titans work on artificial general intelligence
By Lee Min-hyung
With Google-developed artificial intelligence (AI) AlphaGo achieving landslide victories against human go champions, all eyes are on what will be its next target.
Last week, AlphaGo defeated the world’s No. 1 go player Ke Jie from China. The matches were less than sensational, as it was somewhat predictable, given AlphaGo’s records of 68 wins and only one defeat against professional human players.
After AlphaGo’s total victory, the AlphaGo developer, Google DeepMind team, announced the AI go player would retire from the competition, raising expectations over what an upgraded version of AI systems will turn up down the road.
It was long before the term AI was developed and people have already gotten senses of how it will affect human activities.
But the advent of AlphaGo has crystallized how powerfully the AI can evolve, fueling research and development on the big data-based technologies across the globe.
Some argue that the AI supercomputer — such as AlphaGo — has yet to reach a level to replace major tasks currently done by human labors, and it will take decades for the AI systems to fill the role of humans without any errors.
To overcome such weakness, the concept of the artificial general intelligence (AGI) appeared a few years ago, garnering growing attention from the tech industry.
With the focus on the “general,” the AGI refers to an advanced level of existing AI bots, which can perform most of intellectual activities that humans do, regardless of restricted fields — such as the organized game competition.
The AGI means a certain computer system can generally think, behave and perform tasks with almost the same level of human beings, which is the final destination of the ongoing AI research race across the globe.
Some even predict that the AGI would the final invention of human beings because after the arrival of the AGI, it would invent anything instead of human.
But not everybody buys such a gloomy scenario.
“The existing ‘narrow AI’ systems — such as the AlphaGo and IBM Watson — can replace only a tiny part of human activities,” a senior engineer at the nation’s mobile industry said, asking for anonymity.
“But there are so many fields that a computer system should master when it comes to big data learning processes. If an AI system progresses to master all the separate areas and reaches into a tipping point, this is a genuine AGI era, which we call singularity.”
No one can say for sure when the era will arrive, but it is expected to take a long time, given the previous AI development record, according to him.
This is shown in the historical matches between the AI and human beings.
The first in a series of the AI-human competition began in 1996 when chess grandmaster Garry Kasparov defeated IBM’s Deep Blue supercomputer 4-2.
But the next year, they held a rematch where the AI system became the world’s first computer system to defeat a human.
The U.S.-based technology and chip giant then prevailed in the global AI research for about two decades. In 2011, the company held the second AI-human quiz competition at the TV show, Jeopardy, where its Watson supercomputer defeated former winners Brad Rutter and Ken Jennings.
“It took almost 15 years for the AI to defeat human beings in a different field,” said the engineer. “This shows it is expected to take some decades before the arrival of the AGI era.”
With the advent of AlphaGo last year, the AI boom resurfaced again, with Google taking the machine learning limelight away from IBM.
Demis Hassabis, founding CEO of AlphaGo developer DeepMind, also said last year in a press event that the arrival of the human-level AI system is still decades away.
With AlphaGo making a retirement, the DeepMind team is set to hold the next StarCraft matches between a human and its AI bot. StarCraft is a real-time strategy game where players have to make different decisions every second, unlike the turn-based go matches.
“General AI, or the AGI, has been an ambition of researchers since the advent of the AI, but current systems are still far from achieving this goal,” the National Science and Technology Council of the U.S. released an AI report last year. “While there is no general consensus, most AI researchers believe that general AI is still decades away, requiring a long-term, sustained research effort to achieve it.”