In December last year, a few weeks after the AI chatbot ChatGPT3.
This is the beginning of a nightmare.
Over the past four months, Pichai may have been the most flustered CEO in American tech. He officially took charge of Google in 2015, and a year later he proposed "AI first", making Google the most determined technology giant in the field of AI. However, ChatGPT's broken circle made the outside world feel that he seemed to have done nothing in recent years.
Google, which had laid out AI 10 years ago, was left behind by the dark horse OpenAI. Conservatism and slowness are the two major "culprits" for its slow pace. The ChatGPT-like product Bard, which was launched in a hurry, was also complained by the outside world.
After the "lost decade", will Google lose the AI battle?
flustered google
Faced with the threat of ChatGPT, Pichai quickly made his first decision: let Bard launch as soon as possible, which is an AI dialogue product that is benchmarked against ChatGPT. On March 21, Google held a Bard launch conference, but Pichai did not attend. In the tech world, this means that the CEO is either not serious about the product or has little confidence in it.
At the press conference, Google played a demonstration video-asking: "What new discoveries did the James Webb Space Telescope have that I should tell my 9-year-old child?" Bard gave three answers, one of which was: "Telescope The first photo of an extrasolar planet was taken."
It was soon discovered that the first photo of a planet outside the solar system was taken in 2004 and can be found on NASA's official website. Then, more Bard trial users began to complain. There is a highly praised user comment on Twitter: "After seeing Google's products, I don't worry about the threat of AI anymore."
From this moment on, Google's two major AI teams—GoogleBrain and DeepMind—began to realize that OpenAI had surpassed itself.
Bard's behind-the-scenes R&D team, GoogleBrain, used to belong to Google Research Institute. It was established in 2010 and was once the world's most cutting-edge AI research team. It has a large number of talents in the fields of neural network and visual recognition. Driving cars have used their technology.
DeepMind was also established in 2010, specializing in machine learning technology, and was acquired by Google for $500 million in 2014. It focuses more on boutique projects, such as Alpha Dog and Alpha Fold. The former once beat the world Go champion Lee Sedol and made headlines all over the world; the latter successfully predicted the structure of protein DNA, which caused a huge shock in the international biological community.
Bard's first defeat made Pichai calm down.
On April 20, he announced the merger of the two major AI teams. The new team was named "Google DeepMind", with DeepMind founder Demis Hassabis as CEO. Jeff Dean, the head of Google Brain, was named chief scientist, essentially out of the game.
Hassabis, one of the co-founders of DeepMind, is 46 years old. He has set up the company's headquarters in his hometown - King's Cross Street, London, England. Hassabis’s father is from Greece and his mother is a Singaporean Chinese. He was once known as a chess prodigy and graduated from Cambridge University at the age of 15. The first thing he did with machine learning was to teach AI to play games.
Inside Google, no one questioned his abilities.
Hassabis source: DeepMind official website
Why is DeepMind winning? Natasha Jaques, a former Google employee who worked on both teams, said: "Brain's way of working is bottom-up, and research is more driven by personal interests. DeepMind is top-down. , the boss sets the direction, and the employees have a standard OKR assessment."
Hassabis quickly announced the structural adjustment, and the DeepMind team took over. For the employees of the Brain team, the free-range scientific research style disappeared.
lost decade
Counting from the acquisition of DeepMind in 2014, Google has not produced a truly disruptive AI product in the past 10 years. Why?
Answer: Conservative and obtuse.
Gaurav Nemad, a former product manager at Google, revealed: "Google has a lot of concerns, and they are very afraid of damage to the company's reputation... They tend to be conservative."
Google's conservatism is dressed in the cloak of an ethical defender, and it continues to convey the argument that "AI is dangerous" to the outside world. In 2018, Google released the "Artificial Intelligence Principles", emphasizing that AI research and development must benefit society and oppose discrimination. Google is quite sensitive to AI ethics internally. Many employees broke the news to the media that the company's AI products already have human consciousness, calling for a suspension of research.
Google's emphasis on AI ethics is not only influenced by the value of "don't be evil" and the politically correct public opinion environment in American society, but also out of consideration for the company's interests.
As early as 2017, Google developed a conversational AIGC product - LaMDA. At that time, Google internally assessed the risks of LaMDA, believing that it made mistakes from time to time, would mislead users, and make users feel distrustful of Google. While search, one of Google's main businesses, is dedicated to helping users find the right answer, AIGC is considered to have the potential to shake the company's foundation.
"Google has been developing this thing for the longest time, why doesn't it dare to launch a product?" Zhang Yaqin, former president of Baidu, said in an interview with "Sanlian Life Weekly", "Because big companies are afraid of imperfect products and making mistakes, and ChatGPT It is equivalent to providing a user's standard."
In his view, Microsoft can boldly add ChatGPT to Bing search because it has a small market share. But if Google adds a chatbot to search, "it will use a lot of computing power, and it will affect profits."
Although ChatGPT occasionally gives unreliable answers, OpenAI is a non-listed company, so they boldly pushed their products to the public. Compared with being conservative, Google's slowness to technology is incomprehensible.
"Facebook and Google didn't realize that there could be such a big breakthrough in this field." A domestic investor said.
Both Google and OpenAI use machine learning techniques that mimic human neural networks, but take slightly different routes. Google uses a supervised learning model, while OpenAI uses a pre-trained model; Google puts more emphasis on intervention, while OpenAI first lets ChatGPT learn independently, and then introduces artificially labeled corpus for training and guidance.
It's more like an arms race in the dark. Until 2018, Musk thought Google was ahead of OpenAI.
Source of Google’s AI dialogue product Bard: Google’s official website
However, when Google gave up its conservatism under the pressure of competition and publicly released AI dialogue products, Bard received overwhelmingly negative reviews. A senior American technology reporter commented: "(Compared to GPT) this is like a standard chatbot."
Adding to the humiliation, when a Barron's reporter asked Bard to describe what he looked like, it replied: "I'm supposed to be six feet tall (1.83 meters), with short brown hair and blue eyes. My skin should be Light-colored, casual attire, like a T-shirt and jeans.” (Editor’s Note: A typical image of a white American mainstream group.)
For a company that emphasizes "AI political correctness", this is a bit ironic.
Now, Google is starting to lose its most important asset in this arms race: talent. In March of this year, according to media statistics, at least 13 star talents in the field of artificial intelligence have left Google, including Jacob Devlin, the head of the pre-trained language model, who joined OpenAI.
Will Google fall behind?
What if Google loses the AI battle?
The secondary market has already given the answer. One wrong answer on the day of Bard’s release cost Google’s parent company Alphabet $100 billion in market value. Microsoft is still "stabbing", and it is trying its best to portray itself as a disruptor of the search industry. Microsoft CEO Satya Nadella (Satya Nadella) said in a conference call that after integrating ChatGPT, Bing’s share in the search market is growing: “The largest software category, the search business, is undergoing earth-shaking changes. We will continue on this journey."
However, according to the data of Statcounter, an American traffic monitoring agency, in March this year, Google searched for 93.17% of the global market share, which has increased compared with October last year (before the launch of ChatGPT); Microsoft Bing currently has a share of 2.88%, and No significant growth.
In the AI big language model track, there are still people who are optimistic about Google.
He Xiangxin, the investment director of Hongxin Fund who has been following the US AI industry for a long time, does not think that Google has achieved nothing in the AI field in the past 10 years. In his view, the transformer represented by T in ChatGPT was actually invented by Google.
It's just that Google has used many of its achievements to improve its product experience, such as search results and YouTube's content recommendations, "these are not revolutionary." The amazing effect of ChatGPT is that it integrates unsupervised learning, supervised learning and reinforcement learning.
Google is constantly improving Bard and rolling out new features intensively. Alex Kantlowitz, a well-known American technology blogger, said: "Google's chatbot technology may have a latecomer advantage." Some people in the industry believe that OpenAI's "pre-training model" has established a technical route for the industry, and Google's The advantages will be played out soon: computing power, data and talents.
Google's move to merge its AI teams was also seen as timely and necessary. A user commented on the social news platform Reddit: "In the field of large language models, are you going to bet on two teams with 500,000 GPUs (graphics processing units) each, or a team with 1 million GPUs?" April On the 26th, Alphabet, the parent company of Google, released its Q1 financial report for 2023. Both revenue and profit exceeded market expectations, and the cloud business achieved profitability for the first time. Pichai was the first to mention AI on the call, though the business didn't show up in its financials alone.
While "Amway" Bard, he tried his best to downplay the threat of generative AI to the search advertising business: "Users want to choose what they want to see, even if we provide them with answers and content summaries, they are the same."
According to Hassabis, Google's AI No. 1, the battle towards general artificial intelligence has just begun, and Google is not far behind. "AI will eventually have human consciousness, but today it's far from it," he said.