“Humanity should benefit from AI”
Demis Hassabis, CEO of Google DeepMind, gives lecture at KAIST on March 11 “AI still has long way to go” “Artificial general intelligence can be seen as a tool. This tool must be used ethically and responsibly to support human creativity and pursuits.”
The father of AlphaGo, an AI-based Go program, visited Daedeok on March 11.
Demis Hassabis, the CEO of Google DeepMind, gave a special lecture at KAIST on the theme of “Artificial Intelligence and the Future.” The event was organized by the Department of Bio and Brain Engineering.
The lecture attracted a crowd of more than 500, consisting of faculty, students, and the media. ◆ “Applications of artificial general intelligence include health, robotics, and smartphones”
DeepMind, an AI-startup first established in 2010, was acquired by Google in 2014. A team of roughly 200 researchers is actively working on AI research and developing learning algorithms to solve other problems.
The ultimate goal of DeepMind is to develop artificial general intelligence, such that machines learn automatically from raw material and operate flexibly across a wide range of tasks.
Artificial general intelligence is different from artificial intelligence. For instance, IBM’s Deep Blue defeated world champion Garry Kasparov in a chess match, but it was not capable of any other tasks. Artificial general intelligence, on the other hand, can be applied to various areas.
AlphaGo was trained by deep reinforcement learning, which is a combination of deep learning and reinforcement learning. Hassabis explained the learning process using video clips of iconic games such as Space Invaders and Blackout. The agent makes mistakes at first but eventually masters the gaming strategies.
To test the performance of AI, the team chose Go, which is known to be the most complex game ever devised. The game often relies on the intuition and experience of professionals, and mastering it would involve combining pattern recognition with planning.
For AlphaGo to master the game, the team trained a value network on game positions obtained while the policy network played against itself.
Hassabis said, “I think of AI as the meta-solution to information overload in genomics and physics, and system complexity in climate change and disease.”
He added, “The dream is to make AI scientists or AI-assisted science possible.”
◆ “AI should serve humanity”
The lecture was followed by a ten-minute Q&A session.
In response to a question on the impact of AlphaGo’s victory on the future of humanity, Hassabis said, “We see AI as a tool to automate some tasks that are quite hard for humans to do. We are a long way away from human-level general intelligence.”
Another student’s question was on the dependency of humans on artificial intelligence. Hassabis replied, “Like the Internet and smartphones, if we use these things correctly, they will help us. We should use AI to help improve us.”
The lecture held great significance for the students.
Junseong Im, a student majoring in computer science, said, “I was amazed by AlphaGo’s learning process. Reinforced learning is a slow method, and I’m curious how it can be improved and applied more effectively to other fields.”
Kwang-Hyun Cho, the head of the Department of Bio and Brain Engineering, said, “Hassabis expressed interest in hiring young and talented people at KAIST, and we invited him a month ago to give a special lecture to the students.”
He added, “AlphaGo’s development is fascinating, but it is not all-powerful. I look forward to a more advanced AI, one that is able to solve mathematical challenges and make effective decisions.”
12 3 4 5 6 7 8