Lee Sedol and AlphaGo
In March 2016, there was a sport that captured the attention of the world. A human and an AI battled it out on a checkerboard, with the AlphaGo AI defeating Lee Sedol in four games to one. For many, the sight of Lee Sedol, the number one player in the world at the time, being defeated was impressive. It was reminiscent of what we often see in science fiction movies, where an AI completely transcends its human counterpart.
The game betweenAlphaGo and Lee Sedol had a profound impact on both industries. Not only did the world of Go embrace the excellence of AI and use it to improve their own players, but it also brought the game to the attention of the general public.AlphaGo was awarded the honorary rank of 9th dan, and was honored to be name done of MIT's "Top 10 Innovations to Watch in 2016." It was a positive synergy for both industries.
AlphaGo after playing against Lee Sedol
After the Go match oft he century, AlphaGo received a lot of attention and went on to change the world. With Alpha Zero, Alpha Tensor, and other algorithms, Google's DeepMind team continued to push the boundaries of computational speed in AI. However, not many people know about the recent history of AlphaGo, which has been labeled a "failure" due to a surprisingly low predictive hit rate.Here's the story behind AlphaGo, the program that was supposed to change the world.
AlphaGo's past glory and dramatic achievements
In 2017, the year after the game of the century, AlphaGo was first put to work on a project to reduce power consumption in Google's data centers. Google's power consumption was very high, at 4.4 terawatts in 2014, which is about the same as the power consumed by 360,000 homes in the U.S. A large portion of that is used to cool the data centers, but that's not something that can be abandoned, and it's not something that can be easily adjusted. In October 2022, Kakao suffered a datacenter fire that brought South Korea to a standstill and was caused by a faulty cooling solution.
AlphaGo in Google's data centers fine-tuned more than 120 variables that affect factory power, including fans, windows, and cooling systems. The end result was a 40 percent reduction in cooling costs, which led to a 15 percent reduction in overall power consumption. The company was able to utilize a safe and efficient cooling system with no human intervention, proving that AI can be efficient in many areas.
AlphaGo began to accelerate its efforts to enhance the capabilities of AI and develop efficient systems in various fields. Now that the warm-up exercise was over, it was time to take on the challenge of controlling a national system.
AlphaGo, do monkeys fall from trees too?
Google DeepMind, the developer of AlphaGo, has begun to learn more about combating climate change through power efficiency. Mirroring its success in reducing power use inGoogle's data centers by more than 40%, it was ready for its next project. It formed a team called DeepMind Energy to work with the UK's National Grid PLC.The goal was to reduce 10% of the UK's national electricity use. Ultimately, the project fell apart in 2020 and the team disbanded.
According to CNBC, deep-minded technology works well in controlled environments and with established rules, such as in games like Go or chess, but not in higher-level situations. The complexity of the real world combined with the unpredictability of the future only served to amplify doubts about its performance. Still, there was certainly potential to continue with the project, but the lack of commercial viability compared to the existing power system stood in the way.Another dilemma was the lack of clarity on who was responsible for the variables and events that resulted from incorrect predictions.
Controlling the variables and unpredictability of vast amounts of data is probably what we want from AI, but ironically, the possibilities shown by AlphaGo made it difficult to glimpse this future. In addition to AlphaGo, Google's other AI devices developed by the DeepMind team were largely unable to reach the market for commercial reasons.
Although it failed....
AlphaGo tasted its first defeat in the Fourth Great Game with Lee Sedol, and its second defeat in the UK's power efficiency scheme, Google has since made meaningful progress in the life sciences with its successor, Alpha Fold. It has also made strides in the field of "predicting life phenomena," which DeepMind CEO Demis Hassabis has identified as a major challenge alongside combating climate change.
When Alpha Fold succeeded in analyzing the structure of proteins, it captured the world's attention by solving a protein structure problem that had been unsolved for 10years in 30 minutes. In the future, it may be possible to find a way to count era new virus from existing treatments in a matter of hours, rather than days of constant experimentation.
In short, AlphaGo is, for the time being, an abstention from big data that has become impossible to learn from. Experts are beginning to recognize that AI can't handle all data, which is why, even as AI becomes more sophisticated, the "human-in-the-loop"approach, where humans and AI collaborate, shines. It's proof that human control over the errors that inevitably occur when processing data can lead to more accurate results.
AlphaGo should make us think.
AI is not a miracle box, and while it boasts immense learning capabilities that transcend humans, it resembles the human process in that it fails and becomes a better model, and even when it fails, it creates success stories in new fields. At Datahunt, we promise to keep this history in mind and use it as a mirror to provide betterAI solutions.