A report from Competitive Go | "These Seven Years We've Been Living With AI"

Xu Mei

The reason why we continue to follow the story of professional Go players' encounters with AI is that we firmly believe that for other professional fields, this story is a sample as well as a mirror. How to understand AI and how to protect the good guys when AI is fully capable of stealing jobs from humans and outputting a steady stream of content? Human chess players have been coexisting with AI for seven years, and their real experiences and feelings deserve to be carefully listened to and recorded. What they have experienced in these seven years should become important signposts leading to good questions.

On February 21, 2023, I sent a tweet to Guli, a nine-dan Go player who had won eight world championships, "Mr. Guli, after AI 'stirred up' the Go world, it hasn't done much, but this time, it finally stirred up some waves. I'd like to make an appointment with you for an exclusive interview to talk about it, just as we said in our discussion back then, the experience of the Go community in facing AI is a sample for other industries in society."

The first question in the interview outline is "ChatGPT is still a toy not a tool at the moment, but as soon as it was launched, it set off another round of discussions like 'will AI take our jobs' ...... "

When I put pen to paper, I thought it was just another clichéd question, but unexpectedly, just half a month after coordinating with Chouli, OpenAI held a launch event for ChatGPT-4, and the presentation shocked the world.

Discussions about whether ChatGPT was taking artificial intelligence (AI) straight to general artificial intelligence (AGI) through large-model training continued throughout March, and as April began, a six-month moratorium on ChatGPT and other large-scale AGI training was signed by a network of signatures that raised even more concern and discussion, with industry leaders in both the pro and con camps.

General AI has been called "the last invention of mankind", and before the launch of ChatGPT-4, people were more than happy to listen to this statement as a joke - even if AI had already taken the holy grail of human intelligence from humans in Go in 2016, but after stepping off the chessboard, people were more than happy to listen to this statement as a joke - even if AI had already taken the holy grail of human intelligence from humans in Go in 2016. the holy grail of human intelligence, they still seem limited and clumsy when they step off the board.

This time, however, the wolf seems to have really come, and the speed of it is staggering. openAI CEO Ortmann said in an interview that a raw and straightforward ChatGPT-4 was trained back in August 2022, but it doesn't know how to get along in the world, and asked it, "How do you make a bomb?" "Where do I buy cheap cigarettes?" OpenAI spent seven months building a "security fence" to manage its harmful outputs, and after the model was trained, it enlisted 50 professionals to set up "lure attacks" from a variety of perspectives, including law, cybersecurity, and ethical dilemmas. "ChatGPT-4, eventually a moral, intellectual and physical development of the super AI to society. The public does not know the training process behind, only focusing on exclaiming how just half a month ChatGPT-4 has grown into a man, from a serious nonsense chat toy into a competent workplace "social man". For those who have paid attention to the Go AI, everything is déjà vu - in October 2015 AlphaGo five wins Fan Hui two-dan, the following year in March, has evolved from V13 to V18, the strength of the game has grown by four discs to 4-1 beat Lee Sedol nine-dan is this advanced version. 2017 spring, Ke Jie in Wuzhen with a 0-3 defeat. Wuzhen defeated AlphaGo 0-3, and Chang Hao 9-dan said in an interview with Southern People Weekly that after Lee Sedol and Ke Jie lost to the AI, "the suspense of the man-machine confrontation is over."

In 2018, just one year after the end of man-machine confrontation, Go AI has turned from a technical experiment and professional paper into a commercialized product. Domestic AI "Jieyi" has become an auxiliary training tool for China's Weiqi team, and after an important game, players no longer get together to review the game as in the past, but rather hold their cell phones and computers, and make an inventory of their gains and losses according to the win rate given by the AI.

The reason why I continue to follow the story of professional Go players' encounters with AI is that I firmly believe that for other professional fields, this story is a sample as well as a mirror. Counting from March 2016 when the AI defeated Lee Sedol, the big-killing AlphaGo is still a tool-based application, whether on the board or in the fields of drones, healthcare, finance, etc., whereas the generative AI trained by LLM (large language model) possesses intelligence rivaling that of human beings, which is enough to replace the content producers; the conversational AI is capable of Conversational AI is able to have coherent conversations with humans through text chat and voice chat.

They are fully capable of stealing jobs from humans and are capable of producing a steady stream of content. Louis Rosenberg, the scientist involved in Musk's initiative to suspend large-scale AI training in all labs for six months, is the CEO of Unanimous AI (yes, the same AI company that published Oscar winners' predictions in Newsweek and accurately predicted the election of Emmanuel Macron), and Rosenberg developed the first immersive augmented-reality system for the U.S. Air Force. He writes that his signature is "not a concern about super AI suddenly taking over the world," but rather about the impact of AI on society.

As AI generates a steady stream of images, artwork, computer software, videos, essays, poems, and scientific articles, it will become increasingly difficult to discern the authenticity of the content and clarify the "AI manipulation" behind it, "In fact, I fear that AI-driven conversations will become the most powerful form of targeted persuasion that humanity has ever seen. powerful form of targeted persuasion." Rosenberg voiced his concerns. Louis Rosenberg's concerns immediately came to mind when I swiped through a short video that exclaimed, "ChatGPT can do a hundred videos a day." Bill Gates said in a previous interview that what he least expected was that social media, instead of bringing greater transparency, would make the public space "toxic," and that if AI was manipulated to inject greater doses of bad content or harmful speech, the public space would, unsurprisingly, become even more muddy. I hurriedly urged Guli, after the last WeChat preliminary communication, he was first busy Chongqing Weiqi Association's hanging ceremony, time constraints; then we both suffered a flu, the interview time has not been finalized. "Mr. Gu well? Humans do not have super arithmetic still carrying a heavy body, just when we are here and there sick kung fu, ChatGPT-4 has come out, just like back then, flew from technology to product process."

As China's first eight-champion Go player, Guli was the only South Korean player Lee Sedol to rival him in his heyday, and the two teamed up to create the "Guli Era" in the history of Go.In November 2019, Lee Sedol, who holds 14 world championships, said that after the emergence of the AI of Go, he could no longer enjoy the fun of Taoism and art, announcing his retirement. his retirement from the game, sending shockwaves through the chess world. Guli mocked himself as a "semi-retired second-rate player" when AI appeared, and his inner struggle was not as great.AI has changed not only the ecology of professional players, but also commentators, and his level of authority has become "fearful", "When the position is not clear, I can't judge the position," he said. "When the position judgment is not clear, in the past, you can be very confident and bold prediction, now will say, come on, let's see 'AI teacher' how to look at ......."

Goldman Sachs analysts released a report bluntly stating that generative AI, including ChatGPT, has the potential to displace 300 million jobs globally.OpenAI, the research and development organization for ChatGPT, says that 80% of American jobs will be affected, and they list the top 34 "iron rice bowls" that won't be affected by AI for a while -- Dishwasher, Bartender, Plumber, Greenwriter ...... That means only blue-collar workers are relatively safe under the new wave of highly automated intelligence.

Where technology goes we can't imagine, and the human heart is even harder to gauge, ChatGPT-5 may have been trained long ago ...... As Louis Rosenberg writes in his article, "Will this letter work? It's not clear to me that the AI industry would agree to a six-month moratorium, but the letter has drawn attention to the issue." AI as a resource is naturally attractive to capital, and at a time when there is too much discussion of too much capital pouring into the technology, if the biggest difference between humans and AI is really that only humans currently have a sense of problem and ask questions of AI, we need to start with the human experience and feelings and ask the key questions. How to understand AI and how to protect the good guys? Human chess players have been coexisting with AI for seven years, and their real experiences and feelings deserve to be carefully listened to and recorded. What they have experienced in these seven years should become important signposts leading to good questions. When I was reading the information, writing the outline, and circling around the desk, sometimes my heart would be very excited, feeling that I was participating in a field survey about the future of mankind. While waiting for Guli to confirm the interview time, I contacted former Go player Li Zhe and Golaxy AI Go CEO JinXing.

LiZhe became famous at a young age, reaching the World Championships at the age of 14, and by the age of 18, he had beaten top players such as Lee Sedol, Gou Li, Luo Zaihe, and Xie He, and won the men's individual bronze medal at the inaugural World Intelligence Games in 2008. Because he could not find the value and significance of winning, he went to Peking University in 2012 to study at the Department of Philosophy, and after graduating from Peking University as a graduate student in 2022 and joining the Physical Education Department of Wuhan University, the Go classes he gave to his students felt more like philosophy classes.

Golaxy AI Go currently holds over 90% market share in the cloud Go service market, with nearly 200,000 paid users. Even the Korean national team has purchased their services. On the evening of March 20th, Jinxing accepted my phone interview. After the call connected, he said "long time no see," as the last meeting was three years ago. Over the three years of the COVID-19 pandemic, many startups have faced unprecedented challenges. Jinxing said, "We are doing well, the company is not big, but we are profitable, able to support the team, and still have a surplus." Jinxing is also highly focused on the launch of generative AI, "This is a major change in human-computer interaction. Initially, people had to code for computers to execute commands, then Microsoft introduced the graphical interface, followed by Steve Jobs bringing touch screens, and now ChatGPT has conquered natural language processing. A person who doesn't understand computers, or even how to use a smartphone, can now give commands to machines through normal conversation."

Jinxing believes that just as AI quickly reshaped the Go industry, a stronger and smarter ChatGPT “will bring about impacts like a major earthquake.” On March 29th, I had an hour-long phone call with Gu Li. Finally, I’ve completed interviews with several main subjects. It’s been a month since ChatGPT was released and the topic was initiated. During this period, I often felt anxious, felt inadequate in mental capacity, learning, absorbing, and driving the interviews forward. There was even a time when I particularly hoped that ChatGPT-4 could help me organize the material and write the draft. The image of Gu Li holding up a surrender in the mixed human-machine match in Wuzhen keeps reappearing in my mind, haha…

"Compared to AI, the progress of human Go players is like that of snails," Ke Jie, who had risen 50 ranks in the rating within a year, once said in a CCTV interview, dedicating all his time to Go, "Sometimes I get so tired that I nibble on an apple and fall asleep while nibbling..." I also sent an interview outline to Ke Jie, asking his father to pass it on. I haven’t had much leisure time over the past month. But you see, whatever we humans do, it's so time-consuming and exhausting. Whenever I think about ChatGPT tirelessly advancing and upgrading, I can't help but get up and make myself a cup of coffee to stay awake. It’s time to share their insights.