The Elephant in the Room in the Google Search Case: Generative AI
Kate Brennan / Nov 4, 2024Kate Brennan is the Associate Director at the AI Now Institute. She is a former Google employee and a graduate of Yale Law School.
Earlier this month the US Department of Justice filed its proposed remedy framework in its antitrust case against Google. This kicks off the second stage of the major federal court case following the District Court ruling that Google maintained an illegal monopoly in the internet search and search text advertising markets. The filing provides crucial insight into how the DOJ is considering approaching its upcoming remedy recommendation, due to be released on November 20.
Building off of the framework set out in US v. Microsoft, the DOJ states that remedies adopted by Google must achieve four goals: (1) unfetter markets from Google’s exclusionary conduct, (2) remove barriers to competition, (3) deny Google the fruits of its statutory violations, and (4) prevent Google from monopolizing these markets and related markets in the future.
The elephant in the room is generative AI, which is rapidly transforming the search engine market landscape. Look no further than Google’s own generative AI chatbot Gemini, which provides users with an impossible-to-opt-out-of “AI Summary” at the top of Google’s own search results page. The DOJ is acutely aware of this potential, writing:
Further complicating matters, artificial intelligence—while not a substitute for general search—will likely become an important feature of the evolving search industry. It is, therefore, critical that any remedy carefully consider both past, present, and emerging market realities to ensure that robust competition, not Google’s past monopolization, will govern the evolution of general search and text advertising.
To effectively ensure that Google’s past monopolization does not govern the future of search —as well as deny Google the fruits of its statutory violation—the DOJ’s proposed remedies must solve for the interdependence between Google’s search market monopoly and its emerging dominance in generative AI. To that end, the DOJ must consider the following variables while crafting its remedy proposal.
1. Google directly leverages its data and market advantages derived from its illegally maintained search position to achieve and maintain dominance in the generative AI market.
Large language models (LLMs) like Gemini require access to massive amounts of training data to be effective. Simply put, Google is able to gain an advantage in training its own generative AI models because of the massive amounts of user data it derived from illegally maintaining a monopoly across Search. Real-time data about what, when, and how people search the internet every day is only the beginning. Google’s search text advertising business model incentivized relentless data collection and retention, meaning Google has spent decades investing in sophisticated data gathering, organization, and commercialization—giving it the machinery to generate high-quality AI training data.
The advantage Google achieved from its monopoly position in Search extends across its product portfolio, allowing Google to invest in bets on Chrome and Android that provided critical search access points to further cement Google’s dominant search position—and, ultimately, provide Google with more data to train its LLMs. Google is proud to admit that data it derived across its owned properties, including maps data and sports interests, contributed to its generative AI advantage.
Investing in generative AI is both expensive and risky: it requires significant upfront capital without a clear path to profitability. The next phase of AI will be shaped by those who have the pockets to withstand the cost and revenue risk—exactly where Google’s monopoly profits come in handy. Google is able to move considerable capital derived from its search monopoly revenue (estimated at $175 billion in 2023) to invest in building and thus controlling AI infrastructure.
Strategic capital investments into new business areas are not, in isolation, a problem. But Google stands apart for its vertical integration across the entire AI stack: Google has poured billions of dollars into controlling core compute infrastructure with Google Cloud, chips with Google Cloud TPUs, and data centers across the country. Google controls AI research by funding PhD fellowship programs and courting scientists to work at its in-house AI research lab, DeepMind. In fact, Google allegedly recently spent an incomprehensible $2.7 billion to poach back key AI scientists. Google also controls third-party training data, paying exorbitant fees for access to private training data for its LLMs. In each of these cases, Google can draw on search’s endless well of monopoly profit.
New entrants simply won’t be able to strike the same deals for IP access, limiting the diversity of their training data. They won’t be able to pay billions of dollars to poach academic fellows, build data centers, or control cloud infrastructure. Taken together, Google’s capital investments in AI do not paint a picture of a company making lucky “bets” on different areas; they reveal a strategic push to move capital into critical inputs, ensuring they are one of a few large companies capable of leveraging structural advantages to withstand the staggering capital required to stay the course.
2. Google is deploying the same illegal behavioral playbook from Search to achieve dominance in the generative AI market.
In Judge Mehta’s decision finding Google maintained an illegal monopoly, a key piece of evidence was Google’s exclusive contracts with Apple and other browser providers to set Google Search as the default search engine across products. Yet Google is already striking similarly exclusionary deals in the AI market, including its $60M annual deal with Reddit to receive exclusive access to Reddit’s training data. Reddit’s recent shareholder letter revealed that “Reddit” was the sixth most Googled word in the US, demonstrating just how valuable Reddit is for search users. This exclusionary move shuts potential generative AI competitors out from this deeply valuable resource.
And it’s potentially not just training data. It is easy to imagine a near future where Google strikes exclusive deals with emergent chatbot startups to ensure that Gemini is the exclusive LLM partner, reinforcing its dominant market position and ensuring that nascent competitors have embedded dependence on Google technology.
Cristina Caffarra and Robin Berjon previously wrote on Tech Policy Press that one important way Google controls search is by controlling the broader publisher ecosystem. Google does this by controlling how publishers organize information on their own web pages. Google gives preference within its search results to websites that are built on the Accelerated Mobile Pages (AMP) web framework, which is, in theory, an open-source web components framework. In practice, Google establishes the rule and uses it as coercive leverage to ensure publishers structure web pages in the way Google prefers—and, ultimately, in a way that enables Google to serve up publisher content within the search engine results page so users never have to click away from Google.
While likely not its initial intention, the AMP framework is a deceptively effective mechanism to organize the internet in a way that prioritizes LLM training data (take, for example, the countless websites structured entirely in question and answer format, just like the chatbots like it!) and exert control in the AI market. Write Berjon and Caffarra, Google can easily downrank any publishers within their search results who fail to play into their game or provide training data for Google’s LLMs. More than anything, this hypothetical illustrates how Google, by using its dominance in search to control the organization of the internet, can pull levers to gain cross-market advantages.
3. We should be wary of propositions that suggest the shift towards generative AI-enabled search engines may enable competition in the search market because the advent of generative AI does not change the structural chokepoints of the search market.
Search engine technology is rapidly changing with the advent of generative AI-enabled search, and new companies are able to license access to search indexes and LLMs to build their own chatbots. For example, search engine competitors like DuckDuckGo provide AI chatbots as a complementary way to search. This has created some optimism that generative AI may remove barriers to competition in the search engine market.
Yet this optimism is greatly misplaced. First and foremost, access to search index data required to build search products is conditional. Last year, Microsoft threatened to cut off access from search engine competitors using Bing’s data to train their own competitive chatbots.
Second, we should be wary of propositions that suggest that generative AI companies are capturing Google’s market share and will obviate the need for bold, court-sanctioned remedies. There has been much discussion over whether ChatGPT will replace Google. For example, The Economist writes that ChatGPT is the “go-to search engine” for 8% of Americans. And last week, OpenAI launched ChatGPT Search, building web search capabilities into ChatGPT’s existing interface. However, claims that ChatGPT is meaningfully stealing Google’s search market share fail to consider just how entrenched Google’s position is as a dominant search provider. While OpenAI loses $5 billion a year and struggles to find a business model, Google’s search ad revenue and its search market share continue to rise. Contesting market share at just one of search’s access points—text-based search inputs (rather than, say, voice)—ignores the meaningful and material ways Google has used the fruits of search dominance to vertically integrate the AI stack and gain market advantages in adjacent markets. Perhaps more fundamentally, we should be skeptical of any argument that solves one monopoly problem with another—after all, ChatGPT’s OpenAI is effectively controlled by Microsoft, another company leveraging its dominance to control inputs across the AI stack.
4. Effective remedies must divest generative AI from the fruits of Google’s illegal search monopoly, but this risks later consolidation within the AI industry. Accordingly, it is crucial for agencies to pursue a strong regulatory agenda beyond this case to investigate firms engaging in similar behavior as Google.
Many bold remedy recommendations have been offered, including divestiture, disgorgement of data, and treating Google like a public utility. Disgorging Gemini and other Google models of search data provides one of the strongest potential remedies for ameliorating the legacy effects of Google’s illegal search monopoly in the generative AI market. Under a data disgorgement remedy, the Court can order Google to delete illegally obtained data from products built using that data—in this case, Gemini or other AI models.
Strong regulatory moves like data disgorgement will be effective at denying Google the fruits of its illegal conduct. But make no mistake: the AI industry has a concentration problem writ large.
The appropriate focus of this case is to devise remedies that meaningfully address Google’s search monopoly, not create perfect competition in the AI market. However, it’s worth keeping in view that a weak regulatory agenda has historically allowed Big Tech companies to amass power, control key AI inputs, and position AI as a geopolitically strategic market where the government’s role is to enable rather than govern. Failure to take action against other companies who may wield outsized market power to shut out competition will set Google back without effectuating meaningful change in the broader search and AI ecosystems.
5. Finally, remedies must protect against Google using generative AI search as a means to evade regulatory scrutiny—including by pushing a broad-based and unsubstantiated “AI is good for innovation” narrative.
Google’s response to the DOJ’s remedy framework was exactly as expected: the company says regulating Google’s AI tools will harm American innovation. In its immediate response to the DOJ’s remedy framework, Google argued that “hampering Google’s AI tools risks holding back American innovation at a critical moment” and that there are “enormous risks to the government putting its thumb on the scale of this vital industry…all at precisely the moment that we need to encourage investment, new business models, and American technological leadership.” In other documents, Google is pushing the narrative that Google’s role in scientific AI advancements is so beneficial for society that bulldozing copyright law to generate more training data should be excused entirely. More than a PR tactic, this shows how Google is leveraging broad-based and unsubstantiated claims around innovation to evade regulatory scrutiny.
But this also begs the question: if AI is good for innovation in the search engine market, who is benefitting? Certainly not publishers—already thrown around by Google’s market power to optimize their web pages for search visibility—who now risk being completely cut out of Gemini search results. Certainly not for the energy or climate crises, given that an AI-enabled search result uses ten times the amount of energy as a normal search. Probably not users, who have watched search transform from a useful tool to an ad-laden, unescapable obstacle course. It may not even be good for the search engine market itself, considering Google can shift everybody’s search queries into their Gemini app with one smart rebrand. What good will even some of the most progressive data-sharing proposed remedies—such as providing nascent search engine companies with access to an API with all of Google’s search data—do if people are no longer searching through, well, search engines?
Ultimately, Google is a search advertising company with a brilliant eye towards preserving its own market dominance, whatever the eventual new product will be. By leveraging its search monopoly position to vertically integrate the AI stack and push the narrative that AI is poised to solve every social problem under the sun (including the very ones large-scale, generative AI models are actively contributing to), Google enjoys the fruits of its illegality while the rest of the marketplace struggles to compete.
Related Reading:
Authors
