How DeepMind, OpenAI, and Anthropic were founded until the people who were most afraid of the risks of AI decided they should be at the top and built them.

The New York Times summarizes how three AI organizations were established: DeepMind, which developed AlphaGo and AlphaZero, OpenAI, which developed GPT-4 and ChatGPT, and Anthropic, which developed the chatbot AI Claude. .

How Elon Musk and Larry Page's AI Debate Led to OpenAI and an Industry Boom - The New York Times

·table of contents
◆Debate between Elon Musk and Larry Page
◆Birth of DeepMind
◆Until DeepMind was acquired by Google
◆Establishment of OpenAI and departure of Mr. Musk
◆Establishment of Anthropic by former members of OpenAI
◆How OpenAI technology was adopted by Microsoft

◆Debate between Elon Musk and Larry Page
In July 2015, Elon Musk held a three-day party at a wine country resort in California. After dinner on the first day of this party, then-Google CEO Larry Page and Mr. Musk talked about the theme of ``AI'' near the poolside bonfire.

Mr. Page said, ``Humans will eventually become one with machines with AI, and one day many types of intelligence will compete for resources, and the best will win.'' In response, Mr. Musk, who believes that humans should be prioritized, said, ``If that happens, we will be destroyed. Machines will destroy humanity,'' and there was a conflict of opinion.

Mr. Page and Mr. Musk have known each other for over 10 years and are said to be good friends, but this talk completely heated up due to conflicting opinions. 'Musk is a 'specieist' who prefers humans to future digital life forms,' Page said in frustration. Mr. Musk recalled that Mr. Page's remark was 'the final blow (to confront Mr. Page).'


UK Government

The New York Times said, ``This is one of the more difficult arguments common in Silicon Valley. But eight years later, their argument seems prescient.'' I am.

◆Birth of DeepMind
Five years before the Musk-Page debate, then 32-year-old neuroscientist Demis Hassabis attended a party hosted by investor Peter Thiel. At the time, Mr. Hassabis was looking for funding to conduct research on ``artificial general intelligence (AGI)'', an artificial intelligence that can do anything the brain can do.


George Gillams

Mr. Thiel apparently liked science fiction, especially works that deal with the singularity , in which AI and intelligent technology will no longer be controlled by humans, and Eliezer Yudkowsky, a self-taught AI researcher, said that he liked Thiel's works. With financial support, he expanded the research institute he founded, MIRI (Machine Intelligence Research Institute), and created an annual conference on the singularity. Mr. Hassabis' colleague got to know Mr. Yudkovsky, and Mr. Hassabis was able to receive an invitation to Mr. Thiel's party.

However, Mr. Thiel, who was also successful as a co-founder of PayPal and earned a huge amount of wealth, was very cautious as he thought that most people were attending the party for the purpose of asking for money. So, taking advantage of the fact that Mr. Thiel is a chess enthusiast and that Mr. Hassabis was once a skilled chess player who was ranked second in the world in the 14-and-under category, he asked Mr. Thiel, ``(Chess 'There is a deep tension between the bishop and the knight (pieces). Although these two pieces have the same value, top players can see that their strengths are very different.'

This strategy was successful, and Mr. Hassabis was able to arrange a meeting with Mr. Thiel at a later date, and obtained an investment of approximately $2.25 million (approximately 225 million yen, hereinafter converted at the exchange rate at the time) from Mr. Thiel. I was able to get it. Mr. Hassabis named his startup ``DeepMind'' after ``deep learning,'' a machine learning technology, and ``Deep Thought,'' a supercomputer that appears in the novel ` `The Hitchhiker's Guide to the Galaxy.'' .

Mustafa Suleiman, co-founder of DeepMind, said, ``AI technology has great benefits. The goal is not to eliminate the technology or pause its development, but to reduce its downsides.'' doing.

Furthermore, Mr. Hassabis was able to meet with SpaceX's CEO Musk, as Mr. Thiel's investment fund invested in the space company SpaceX. When Musk said, ``We are planning to colonize Mars to escape overpopulation and other dangers,'' Hassabis said, ``Unless superintelligent machines come to Mars and wipe out humanity. The plan will work.' Mr. Musk was surprised to receive an unexpected comment from Mr. Hassabis, and it seems that he immediately invested in DeepMind.

With plenty of funding, DeepMind hired researchers specializing in neural networks, algorithms modeled after the human brain. Neural networks themselves have been researched since the 1950s, but DeepMind has taken this neural network research further and succeeded in building AI that plays games such as Space Invaders and PONG. DeepMind's success drew attention from Google, especially Mr. Page. Mr. Page saw a demo of AI actually playing the game and became strongly interested in DeepMind.

◆Until DeepMind was acquired by Google
In the fall of 2012, Professor Jeffrey Hinton of the University of Toronto was conducting research on neural networks. The neural network built by Professor Hinton can accurately recognize objects such as flowers, cars, and people, astounding scientists around the world. In particular, the Chinese company Baidu paid attention to Professor Hinton's research results, and offered Professor Hinton and his students $12 million (about 1.3 billion yen) and asked them to join Baidu. However, Professor Hinton declined the offer.

by Collision Conf

Professor Hinton did not lead a wealthy life, but he believed that ``we do not understand our own value properly,'' and decided to organize an AI academic conference that brings together major tech companies such as Google, Microsoft, and Baidu. He planned to participate in the event and put himself up for auction.

Naturally, various companies were interested in Professor Hinton, and his contract fee continued to rise. The main competitors were Microsoft and Google, and the contract amount eventually rose to $44 million (about 4.7 billion yen). Although there were still signs that the bidding price for the contract would rise, Professor Hinton and his colleagues ultimately decided to work at Google, so the auction was stopped here.

DeepMind also participated in this auction of Professor Hinton's team, but withdrew because Professor Hinton's contract amount became too high. Hassabis didn't think DeepMind's technology would be dangerous because it was an independent company, but the involvement of big tech like Google and Microsoft has made it difficult to secure future talent. I thought it would be difficult.

Therefore, DeepMind decided to come to the table for acquisition negotiations, which they had been approached with since the end of 2012. However, DeepMind issued the following conditions: ``DeepMind's technology will not be used for military purposes'' and ``AGI technology will be supervised by an independent committee consisting of engineers and ethicists.''

At that time, Google and Facebook were interested in acquiring DeepMind. Google offered an acquisition price of up to $650 million (approximately 70 billion yen), while Facebook reportedly offered a higher acquisition price, but did not agree to the terms proposed by DeepMind. thing. As a result, DeepMind was acquired by Google for $500 million (about 50 billion yen).

Google acquires mysterious artificial intelligence development startup company 'DeepMind' for 50 billion yen - GIGAZINE

On August 14, 2015, about a month after Musk and Page had a heated discussion about the future of AI, the first meeting of the ethics committee proposed by DeepMind was held in a conference room at SpaceX. . Participants in the meeting included Page and Musk, Google co-founder Sergey Brin, former Google CEO Eric Schmidt, PayPal co-founder Reid Hoffman, and Australian It was Toby Ord, a philosopher.

At the Ethics Committee's first meeting, DeepMind co-founder Suleiman gave a presentation called 'The Pitchforkers Are Coming,' in which he said, 'AI could lead to an explosion of disinformation,' and 'DeepMind As Google's technology begins to do more jobs in the coming years, the public will accuse Google of stealing their lives. 'We will need to provide a universal basic income that will share wealth with the people of the world.'

Mr. Musk agreed with this claim of Mr. Suleiman. However, Schmidt said, ``That worry would be completely overdone,'' and Page agreed, ``AI will create more jobs than it will take away.'' This was the only meeting of the Ethics Committee, and a second one was never held.

◆Establishment of OpenAI and departure of Mr. Musk
Musk left DeepMind shortly after this meeting. It is extremely rare for Mr. Musk to not be involved in the management of a company he has invested in, and The New York Times believes that he left after seeing that DeepMind was completely controlled by Mr. Page and Google. However, one reason why Mr. Musk so easily withdrew from DeepMind was because he had invested in another AI-related organization.

A few weeks before DeepMind's ethics committee met, Musk dined with Sam Altman, then CEO of tech incubator Y Combinator, and several researchers at a hotel in Menlo Park, California. I did. This dinner led to the birth of OpenAI, which was founded in December 2015. OpenAI's philosophy of ``general purpose artificial intelligence that benefits humanity'' is due to the fact that Mr. Page argued by the poolside bonfire, ``One day, machines and humans will merge.'' This is to protect humanity from the idea that 'the best will survive.'


Village Global

OpenAI poached Ilya Satskyvar from Google to start the project. Mr. Satsukivar is one of Professor Hinton's students and is part of the group that Google bid on in the auction.

Mr. Mask seems to have wanted to free AI research from economic incentives like Google and Facebook by operating OpenAI as a non-profit organization. However, when DeepMind announced results such as AlphaGo and became a hot topic, the opinion changed to ``OpenAI should produce results sooner.'' According to sources, Musk was planning to take control of OpenAI from Altman and other co-founders in late 2017 and turn it into a commercial operation using a supercomputer being developed at Tesla.

However, after receiving backlash from CEO Altman and others, Musk parted ways with OpenAI in 2018, resigning from its board of directors. Mr. Musk, who was impatient with DeepMind's success, complained at the OpenAI office that ``we need to move faster,'' but an OpenAI researcher retorted that it was ``reckless,'' and yelled ``You idiot!'' at the office. It is reported that he left.

Musk's departure leaves OpenAI in need of new funding. So, CEO Altman met with CEO Satya Nadella through Kevin Scott, Microsoft's chief technology officer, whom he knew. We then established a for-profit corporation under OpenAI's non-profit organization and succeeded in obtaining an investment of $1 billion (approximately 108 billion yen) from Microsoft.

Microsoft invests more than 100 billion yen in OpenAI to develop artificial general intelligence (AGI) - GIGAZINE

◆Establishment of Anthropic by former members of OpenAI
One person who was present when Mr. Musk had dinner with Mr. Altman and decided to establish OpenAI was OpenAI co-founder Dario Amodei. Mr. Amodei, who was active as an altruist , was involved in research to build neural networks that learn from vast amounts of digital text, such as Wikpiedia articles, digital books, and bulletin boards. This neural network was the prototype for the large-scale language model that was later released in the form of GPT.

by TechCrunch

However, Mr. Amodei was dissatisfied with Microsoft, which was trying to take OpenAI in a commercial direction. Therefore, in exchange for a huge investment from Microsoft in 2021, Mr. Amodei planned to oust Mr. Altman, who was the trigger for Microsoft to gain more leverage. However, when this scheme failed, Mr. Amodei left OpenAI and established a new research institute called Anthropic with 15 engineers and scientists.

However, Anthropic spokesperson Sally Aldous said, ``Anthropic's co-founders have never attempted to remove CEO Altman from OpenAI. 'I came to the conclusion that I wanted to launch it, told the representatives at OpenAI about it, spent several weeks negotiating terms that were mutually agreeable, and then left the company.'

Anthropic has since received an investment of 4 billion dollars (about 420 billion yen) from Amazon and hundreds of millions of dollars from Google.

Amazon invests up to $4 billion in OpenAI rival ``Anthropic'' to advance AWS AI accelerator and chip development - GIGAZINE

CEO Altman, who experienced the departure of Musk and Amodei, said, ``There are differences of opinion, mistrust, and ego. This can also be seen in social gatherings.Because they are the closest people, intense conflicts arise between them.''

Two years after Mr. Amodei left OpenAI, CEO Altman was suddenly dismissed from OpenAI's board of directors and will leave OpenAI, confirming his statement. It was reported that one of the board members who announced the dismissal of CEO Altman was Mr. Satsukivar.

Sam Altman, CEO of AI development company OpenAI, steps down; CTO Mira Murati becomes interim CEO - GIGAZINE

Although CEO Altman was appointed chairman of Y Combinator, he was ``often prioritizing his personal circumstances over the organization, and his absences tended to make colleagues and the startups he was responsible for uneasy.'' I also had the experience of having the chairman of Y Combinator fired.

Sam Altman has been fired before OpenAI - GIGAZINE

◆How OpenAI technology was adopted by Microsoft
Of the three AI organizations DeepMind, OpenAI, and Anthropic, the one that was generally greeted with the most surprise was OpenAI's ChatGPT. ChatGPT can be used for free by anyone, even those who have little knowledge of AI, so it can be said to be more well-known than DeepMind and Anthropic's AI.

In March 2022, after OpenAI received investment from Microsoft, CEO Altman and co-founder Greg Brockman visited Bill Gates at his home in the Seattle suburbs. Bill Gates, co-founder of Microsoft, said he was in regular contact with OpenAI executives.

Gates, who initially doubted how well GPT would work, said, ``I will remain skeptical of large-scale language models until I do tasks that require critical thinking.'' About. Therefore, CEO Altman and President Brockman visited Gates again in August 2022, accompanied by OpenAI researcher Chelsea Voss.

Mr. Vos had GPT-4 take a multiple-choice biology test, then graded and explained the answers. There were 60 questions, but Gates' eyes widened as he sat in his chair as he only got one wrong. The presentation showing the performance of GPT-4 was effective enough to break Gates' skeptical attitude.

Shortly after this, Microsoft decided to adopt OpenAI technology across its online services, including its Bing search engine. Furthermore, two months after this presentation, OpenAI released ChatGPT.

The New York Times writes, 'This time, OpenAI defeated the altruists at Anthropic. An optimistic Google rushed to release ChatGPT and released Bard, a chatbot AI, but Google's stock price fell 11% as the perception that it had lost the competition spread, and Mr. Musk was nowhere to be seen in this competition.''

in Note,   Software, Posted by log1i_yk