“Google 25th Anniversary | 3.0 – From AI Trendsetter to Chaser”.
In addition to the field of social media, Google has been unique in Silicon Valley with its search technology in the first two decades of the 21st century. The rich advertising revenue brought by search allows the two experimenters to work on innovative projects in the Google X laboratory and have fun.
It wasn’t until another AI wave triggered by OpenAI’s ChatGPT last year that swept the world and was thought to engulf search technology that Google really became anxious.
Ironically, the first wave of AI around 2014 was started by Google, and that became the catalyst for the rise of OpenAI.
“I woke up early and rushed to the late market.” This saying is like the epitome of Google in the AI era .
Unlike the services and products that have been brought to the forefront before, Google’s research on artificial intelligence seems to be more about technology and less about applications. After we review Google’s investment and results in AI in the past ten years and sort out the company’s relationship with AI, we seem to be able to realize why Google is currently in an awkward situation in the AI era.
Just as Page and Brin met many years ago in a research project at Stanford University, because PageRank was too prominent, the two independently founded Google.
In 2011, Jeff Dean, Gray Corrado, and Andrew Ng participated in Google X, a joint research project between Stanford University and Google. One of its subprojects is called Google Brain. Their goal is to study deep learning and neural networks . To improve the performance of Google products and services. They firmly believe that artificial “neural networks” can actively build understanding of the world like babies.
Soon, Google Brain showed amazing effectiveness and success. Eric Taylor, the former head of Google X, once revealed that the money earned by Google Brain at that time exceeded the cost of the entire Google X department .
So in 2011, Google Brain independently became Google’s artificial intelligence project.
The next year, Google Brain demonstrated its application potential in the field of computer vision. Jeff Dean and Andrew Ng showed 100,000 YouTube videos to an artificial neural network composed of 16,000 computer clusters. A week later, in Without ever being “told” what a cat is, this system cluster accurately recognized the concept of “cat .”
Subsequent interviews with the New York Times, Google Brain papers, and National Public Broadcasting reports… the numerous exposures not only made Google Brain famous, but in the public eye, this may be the first time that people intuitively experience artificial intelligence. application.
Now we all know that the nourishment of AI is “data”. The success of Google Brain relies on Google’s years of data reserves in text, images, and videos, its huge computing infrastructure, and sufficient economic investment. In the field of AI, Google took the lead in realizing “Connect the dots.”
During the same period, Amazon, Facebook, Microsoft and even Baidu and Alibaba on the other side of the ocean began to pay attention to and invest in the research and development of AI. Even though the main businesses of major technology companies have little overlap, in the face of the menacing AI with unlimited potential, an arms race for talents and teams is unavoidable.
In 2013, two years after the founding of Google Brain, Facebook reached out to DeepMind, a little-known AI startup team located in London, England. At that time, DeepMind was on the verge of bankruptcy, but it still insisted on independent operations and insistence on AI ethics. Faced with the “what can’t be done” rules given by the DeepMind team, Facebook backed down.
A year later, Google signed the “Ethical and Safety Review Agreement” proposed by DeepMind and acquired DeepMind for US$600 million .
The initial “facilitator” of this acquisition that caused a sensation in the AI industry was a common friend who introduced Google co-founder Larry Page and DeepMind co-founder Hassabis to each other. He was also an early investor in DeepMind and was an Elon · Musk. He was also the person most opposed to the acquisition.
Musk invested US$5 million in DeepMind when it was first established. The purpose was not for financial returns, but for “monitoring,” because DeepMind founder Hassabis once told him during a meeting that “the intelligence of machines may surpass humans in the future. Maybe even want to destroy humanity.”
In “The Biography of Musk”, Musk bluntly stated that he had discussed the AI threat theory with his old friend Larry Page many times. He believed that Google was the company most likely to let AI go out of control. It had good intentions but might “accidentally create something evil.” “Things” and then destroy mankind. But Page believes that AI is just a technology. Even if artificial intelligence robots surpass humans, it is just evolution. This difference in views put the relationship between the two on the verge of collapse after repeated discussions .
Musk even raised money to try to block Google’s acquisition of DeepMind, but after failure, he co-founded the non-profit organization Open AI in 2015 in an attempt to check and balance Google in the field of AI .
But things may not turn out as Musk feared. Google will become an oligarch in the AI field after acquiring DeepMind.
In 2016, after being acquired by Google, DeepMind came up with AlphaGo and defeated Lee Sedol, the world’s number one Go player at the time, 4:1 . This is the second time that artificial intelligence has become a “hot search” in front of the public after “cat recognition” in 2012. The protagonist still belongs to Google. This is also the last honeymoon period between DeepMind and Google. Since then, DeepMind has been “fighting for independence” ”.
Since its establishment in 2010, DeepMind has never achieved profitability before 2020, burning Google hundreds of millions of pounds every year. As a large commercial company, Google also tries to “force” DeepMind to do practical things, such as putting a “Powered by DeepMind” label to endorse Google Cloud, or undertaking medical business. This makes DeepMind even more rebellious and values its own independence. sex. What’s more, Project Maven, a collaboration between Google and the Pentagon in 2017, has already touched the moral red line set by DeepMind at the time of the acquisition.
From ethics to reality, DeepMind and Google are at odds with each other. The reason is that they have different development directions for artificial intelligence. DeepMind focuses more on technology research and development, while Google wants to see practical applications .
From its inception to its acquisition by Google, DeepMind’s goal has always been to create general artificial intelligence, which is what scared Musk back then. This goal not only requires grasping the boundaries of ethics, but also requires taking a long-term view and focusing on the study of technology. Ungrounded and difficult to implement, this is DeepMind’s embarrassing situation under Google. No matter how good you are at Go, it seems that it cannot solve real-world problems, and profits will be far away.
In 2021, the Wall Street Journal revealed that DeepMind had been negotiating with Google for many years, hoping to gain operational autonomy and requesting an independent legal structure. Then DeepMind failed and Google rejected their request.
DeepMind and Google have maintained an awkward and delicate relationship in this way. OpenAI, which was established to counter this deal, has developed in full swing, and subsequently caught up with Microsoft. Now it has become the leader in the AI industry , and continues to subvert the theory. The field has entered our digital life in terms of application.
Can Hassabis realize his construction of general artificial intelligence in DeepMind? Will Musk’s worries come true? Will DeepMind regret being acquired by Google, and vice versa? How long can Google tolerate DeepMind? These questions may not be able to be answered in the first decade of AI’s boom.
In 2015, Larry Page announced on Google’s blog that Google would be reorganized into Alphabet.
The next year, Google’s new CEO Sundar Pichai made a statement at the I/O Developers Conference: Google has become synonymous with search, and Alphabet will become an AI-first company in the future .
But reality has overturned Alphabet’s assumptions. Today, online advertising still accounts for more than 80% of Google’s revenue, and all of Alphabet’s innovative businesses have continued to make ends meet for many years. Self-driving company Waymo, smart home company Nest, and AR glasses Google Glass… all went from a mentality of subversion when entering the market to now quietly existing. Even Google _
The Loon project, a signal balloon that continues to fly and crash, is like the epitome of Google’s pursuit of disruptive innovation so many times. On the contrary, the implemented technologies have actually improved Google’s products.
In 2016, Google released Google Assistant, a virtual assistant based on speech recognition and natural language processing technology. All core technologies of Google Assistant revolve around AI, and it has subsequently been integrated into the smart home system Google Home, the smart wearable system Wear OS, Systems such as the intelligent driving system Android Auto are now an indispensable part of the Google ecosystem.
In 2017, following “Cat Recognition”, Google Brain made another major breakthrough in image processing. They used neural networks and deep learning to restore the mosaic-like original image to a high-resolution image by first identifying and then guessing the pixels of the image. A clear picture of the rate. This technology has evolved to this day, and we can see various “high-definition remake” videos of private repairs on the Internet.
In the same year, Google released TensorFlow, an open source framework for building and training machine learning models . Researchers, engineers, developers and even ordinary people who are curious about AI can use it for free to build their own machine learning models. TensorFlow also supports multiple programming languages and operating systems.
With its open source nature, cross-platform regardless of hardware, and good community atmosphere, TensorFlow is like Android to smartphones and Chrome to browsers. It may not make much money for Google, but it has undoubtedly reduced the cost of getting started with AI. The threshold has rapidly promoted the development of machine learning and deep learning.
From search to translation, from voice assistant to image search, all of the mainstream Google services have been infiltrated into AI technology. While the product experience is getting better and better, the application of AI has returned to those of Google. “Old guy” , not Alphabet, not the disruptive innovation that Google X wants.
“Make machines smarter to improve the quality of human life.” This is the slogan of the Google Brain team. Simply change it to “making software services smarter to improve the quality of human digital life.” This is what Google has been doing for 25 years. Jeff Dean, co-founder and former head of Google Brain, has said he wants to avoid a “success disaster” in which a company’s theoretical research capabilities exceed its ability to actually develop products.
This is Google’s frustration in the AI era. The huge moat advantage it built in the company’s first fifteen years. When AI comes, innovation dares not be used for subversion, but can only be used for repairs. Long-term theory is not The business that has been implemented and monetized will gradually decline in the company until the talent is lost. Google had been ahead in many areas for too long, and it just wanted to maintain its lead because it thought there was no other competitor ahead of it .
Soon, its truly disruptive innovation rival arrived.
Forced to Challenge
On November 30, 2022, a Silicon Valley startup called OpenAI released ChatGPT, igniting discussions on AI in the global public sphere.
From composing poetry to writing code, ChatGPT delivers actionable advice in seconds. People are surprised to find that AI’s capabilities are far beyond their imagination and may even completely change the way humans create and consume information.
Three months later, Google rushed to the challenge and launched its own chatbot, Bard .
It can be seen from the simple layout of the conference that Google’s preparations were somewhat hasty. One speaker even lost the mobile phone used for demonstration. The most problematic part of the day was an answer from Bard, which made a factual error in saying that the James Webb Space Telescope, launched in 2007, had taken the first picture of an extrasolar planet. In fact, the first photo of an extrasolar planet was taken in 2004.
At the opening of the next day, Google’s stock price plummeted 7%, and its market value evaporated by more than 700 billion yuan.
In fact, Bard is not that bad. After all, ChatGPT also often talks nonsense seriously. But since the release of ChatGPT, everyone has been waiting for Google’s response. The mistakes at the press conference shattered Google’s image as a leader in the field of AI over the past decade. As if overnight, the giant Google was surpassed by the small company OpenAI in the field of AI .
It looks like a David and Goliath story: an innovator takes down an entrenched giant in an industry with a disruptive technology. But if you look at OpenAI’s financing history, you will find that this is actually a competition between the two giants.
OpenAI was originally a non-profit organization established by a group of Silicon Valley tycoons such as Sam Altman, Reid Hoffman, Elon Musk, and Peter Thiel. The vision is to use AI to benefit mankind as a whole. However, as we all know, the cost of training large models is comparable to burning money, and the non-profit organizational form is not enough to carry such a grand vision. So OpenAI ushered in an organizational adjustment in 2019, transforming from a non-profit organization to a semi-profit organization. In the same year, OpenAI received a US$1 billion investment from Microsoft. In 2023, Microsoft added a second investment. The investment is said to be as high as US$10 billion, and Microsoft will hold 49% of OpenAI’s shares .
To some extent, the outside world can regard OpenAI as a subsidiary of Microsoft. In fact, just one day before Bard was released, Microsoft announced that it would integrate OpenAI technology into the Bing search engine and Edge browser. The form is the same as ChatGPT. As long as the user asks a question, Bing will retrieve the relevant web page information and directly tell the user the results by answering the question. This move directly touches on Google’s core business model-online advertising.
Although Google has been born for 25 years and is the most proud technology company on the planet, the company’s core business model is the same as in the first few years of its founding – “appropriate” insertion of merchant ads on search pages. and receive a commission. To this day, more than 80% of Google’s revenue comes directly from online advertising.
Just imagine, if in the future people no longer retrieve information through the search box, but directly ask questions to AI, where will Google insert ads ? Google can easily shut down a lot of projects (Google Reader, Google Buzz, Google Glass, Google+), but facing the challenge of Microsoft, this may be a war that loses everything if it is not careful.
Big Company Disease
To sum up Google’s failure in OpenAI in one word, “big company disease” should be the most accurate.
Today’s Silicon Valley AI companies almost all have a shadow of Google .
The T in GPT technology refers to the Transformer language processing architecture. If ChatGPT is compared to a person, Transformer is its brain. In fact, this technology comes from the paper “Attention is all you need” published by the Google Brain team in 2017 .
But as of August 2023, all eight authors of this paper have left Google. Aidan Gomez, one of the authors of the paper and CEO of the artificial intelligence company Cohere, said in an interview: “In a large company like Google, you can’t really have the freedom to innovate products. Fundamentally, the company structure does not support To innovate, you have to build the structure yourself.”
The most sustainable way for a business company to survive is to keep making more profits. In this process, resources will inevitably be tilted towards core business. In the past ten years, Google has indeed been the technology company that has invested the most resources in the field of AI. During this period, breakthrough technologies such as AlphaGo and AlphaFold were also born, but these technologies have never been successfully commercialized.
As early as 2017, Google developed a conversational robot LaMDA, but Google did not make it widely available to the public like OpenAI. There are two reasons:
First, AI technology is not suitable for Google’s core business . As mentioned above, Google’s core business model is to display advertisements in search results. The emergence of conversational robots will change the way users obtain information. How to continue to maintain advertising revenue will be a problem that Google needs to face. Without external pressure and its main business still running well, Google has no incentive to seek change.
Secondly, as a large company with billions of users around the world, every move by Google will have a huge impact. Compared with aggressively attacking cities and territories, making no mistakes is often the safest strategy. Microsoft could just add ChatGPT to Bing because its global market share is only about 2.8% . However, Google Chrome has occupied more than 90% of the market share all year round. On the one hand, the addition of chatbots will consume a lot of computing power; on the other hand, AI ethical issues force Google not to make mistakes.
Gaurav Nemade, a former product manager at Google, once revealed to the Wall Street Journal: “Google has a lot of concerns and is very afraid of damaging the company’s reputation… They tend to be conservatism.” Within Google, there is a central review team whose members include User researchers, social scientists, technologists, ethicists, human rights experts, policy and privacy consultants, legal experts. For any Google product, they will be reviewed according to the artificial intelligence guidelines set by Google to minimize ethical issues .
Just like a person who is barefoot and not afraid of wearing shoes, as a startup company, OpenAI does not need to bear such great moral pressure and does not need to be responsible for the stock price. Naturally, it can more easily bring ChatGPT to the market.
So will Google lose this AI war? It’s actually hard to tell.
In terms of technology, the Transformer language architecture was invented by Google. Today, Google is still the company with the deepest accumulation of AI technology on the planet. In terms of data, training large models requires a lot of data. Google has billions of users around the world and has accumulated a large amount of data through Youtube, Google Map, search, and Gmail. This is something that startups like OpenAI do not have.
Google CEO Sundar Pichai wrote in a blog post celebrating Google’s 25th anniversary: ”AI will completely rewrite technology and bring incredible advancement to human creativity. Let AI help everyone and deploy AI responsibly. This will become a Google’s most important mission delivery method in the next ten years.”
This is a declaration that “Google is all-in on AI.” People often use the metaphor of an elephant turning around to describe the transformation of a large company. This process may be difficult, but once the elephant turns around successfully and runs with all its strength, no other animal can stop it .