How to Burn Calories... Setting Fire to a Hamburger Restaurant? [AI Error Notes]

⑪ Why Bing Can't Beat Google: Feedback Structure
Data Accumulation → Prediction Improvement → User Inflow Virtuous Cycle
Ensures Strong Advantage Hard to Catch for Followers

Editor's NoteExamining failures is the shortcut to success. 'AI Wrong Answer Notes' explores failure cases related to AI products, services, companies, and individuals.

Google is currently going through an unusually harsh winter. It has lost its image as the AI leader to latecomers like OpenAI, and it is facing a major crisis in the search market due to antitrust lawsuits.

The U.S. Department of Justice has claimed that “Google is monopolizing the online search market” and has requested the court to force the breakup and sale of Google’s web browser, Chrome.

Google, the overwhelmingly powerful monopolist... The secret is the power of the 'feedback loop'

A meme satirizing the performance difference between Google and Bing. When a user asks "how to burn calories," Google suggests 'running,' while Bing humorously answers 'burning down a fast-food restaurant.'

The search market is Google's core business. As of last September, it holds about 90% of the global online search market. Advertising revenue from search services accounts for 70% of its total sales. The advertising revenue earned in just one year is expected to be at least $250 billion, approximately 348 trillion Korean won.

Most Google searches are conducted through Chrome. Launched by Google in 2008, Chrome holds about 67% of the global web browser market share. In other words, Chrome is the key gateway through which Google’s search engine is used and the foundation that maintains Google’s monopoly position. The U.S. authorities aim to alleviate Google’s search market monopoly by separating this foundation.

How did Google build such a powerful foundation in the search market? What is the secret behind its absolute competitiveness that made the U.S. government consider splitting the company? The answer lies in the power of the 'feedback loop.'

In the late 1990s, when the internet was just expanding, the search engine market was dominated by AltaVista, Yahoo!, and Lycos. In January 1996, two friends and PhD students at Stanford University, Larry Page and Sergey Brin, challenged the search market. This was the birth of Google.

Google created a homepage that was strikingly simple for the time, featuring only a search box. It also gained users’ trust by providing clean search results without ads. The search results were differentiated by the 'PageRank' algorithm, which evaluated the importance of web pages. Cracks began to appear in the solid search market, and Google rode the steep growth curve to reach where it is today.

More users → retraining → better search results → more users: a virtuous cycle

As data accumulates, search engines provide better predictions (search results).

During this rapid growth, Google perfected a powerful 'feedback loop.' Positive reviews from early users → more users → more search data → improved algorithms based on data → higher search quality → increased user satisfaction → more users entering the cycle. The more behavioral data such as search queries, search volume, clicks, and dwell time accumulated, the more accurately Google could understand search intent and provide results. This, in turn, led to higher user satisfaction.

Microsoft (MS) launched the Bing search engine in 2009. Despite mobilizing company-wide resources and investing billions of dollars to catch up with Google, it has yet to succeed. DuckDuckGo also challenged Google in the search market by emphasizing the importance of privacy, but both Bing and DuckDuckGo lag far behind Google in search volume and advertising revenue. The reason why Bing and DuckDuckGo’s pursuit is so difficult is precisely because of the feedback loop.

Another meme satirizing the performance difference between Google and Bing. When a user asks "how to sleep when the neighbors are noisy," Google suggests 'earplugs,' while Bing answers with 'weapons,' which is the joke.

Latecomers could not secure enough data, limiting their ability to improve algorithms. The biggest characteristic of the feedback loop is that the leader’s advantage grows over time. The amount of data accumulated over time differed, leading to a gap in search quality.

When the gap with competitors expands exponentially, natural entry barriers are created. Even large capital investments are not easy to overcome.

Amazon, Coupang, Netflix, Facebook: pioneers who also perfected the 'feedback loop'

With the operation of the feedback structure, AI learns the advantages that humans have had, namely how to learn from results. Getty Images Bank

Google is not the only company enjoying the power of the 'feedback loop' that creates absolute advantages for pioneers. Distribution companies like Amazon and Coupang are the same.

They created a virtuous cycle of ‘many consumers → many sellers joining → more diverse products and competitive prices → more buyers.’ Even if they incurred short-term losses, they offered products at the lowest prices to secure market share first. Consumers flocked, and sellers, products, and shopping data accumulated.

Facebook saw more content and interactions among users as its user base grew, which in turn attracted new users. Netflix was able to invest more in content as subscribers increased, and it improved its recommendation algorithm with accumulated viewing data. Accurate recommendations increased content viewing, which led to further investment.

This is exactly why many AI startups rush to launch services and venture capitalists invest huge amounts of money in them.

The power of first-mover advantage well known to Google... rushing led to 'embarrassment'

Prabhakar Raghavan, Senior Vice President of Google, is introducing Bard at an AI demonstration event held in Paris, France, on February 8, 2023. Photo by Google YouTube

The AI market also grants overwhelming advantages to first movers. The earlier a product is deployed in the market, the faster it can collect data and start learning. Google, who knows the power of first-mover advantage better than anyone, made a big mistake.

In November 2022, OpenAI surprised the world by releasing ChatGPT 3.5. The term ‘ChatGPT shock’ even emerged. OpenAI and ChatGPT became synonymous with AI. This was unwelcome news for Google, which had invested more and longer in AI than anyone else. So Google launched its rival, Bard, just three months later.

Bard was officially announced on February 6, 2023, and its capabilities were revealed at a demonstration two days later. Bard was asked:

"Explain what the James Webb Space Telescope (JWST) discovered to a 9-year-old."

Bard answered:

"The James Webb Space Telescope was used to take the first pictures of planets outside the solar system!"

However, this was incorrect. JWST was not the first to photograph planets outside the solar system. That was the Very Large Telescope (VLT) of the European Southern Observatory in 2004. It was a moment of embarrassment for 'Google’s AI' in front of the whole world.

Internal criticism poured in that Google had rushed too much. On an internal Google site, one employee directly targeted Google CEO Sundar Pichai, saying, "The Bard launch was hasty and short-sighted." They urged, "Please return to a long-term perspective."

Many employees supported this post with numerous ‘likes.’ After the first demonstration, the stock price of Google’s parent company, Alphabet, plunged 10% in two days. Embarrassment from the wrong answer and a stock price crash ? it must have been an unforgettable day in Google’s wrong answer notes.

Next installment preview

⑫ The idea of replacing doctors with AI for '3-minute consultations' (25.01.04)
⑬ The 'Japanese AI genius' from the analog country (01.11)