Google’s Plan to Catch ChatGPT Is to Stuff AI Into Everything

Artificial intelligence was supposed to be Google’s thing. The company has cultivated a reputation for making long-term bets on all kinds of far-off technologies, and much of the research underpinning the current wave of AI-powered chatbots took place in its labs. Yet a startup called OpenAI has emerged as an early leader in so-called generative AI—software that can produce its own text, images or videos—by launching ChatGPT in November. Its sudden success has left Google parent company Alphabet sprinting to catch up in a key subfield of the technology that Chief Executive Officer Sundar Pichai has said will be “more profound than fire or electricity.”

ChatGPT, which some see as an eventual challenger to Google’s traditional search engine, seems doubly threatening given OpenAI’s close ties to Microsoft. The feeling that Google may be falling behind in an area that it has considered a key strength has led to no small measure of anxiety in Mountain View, California, according to current and former employees as well as others close to the company, many of whom asked to remain anonymous because they weren’t allowed to speak publicly. As one current employee puts it: “There is an unhealthy combination of abnormally high expectations and great insecurity about any AI-related initiative.”

The effort has Pichai reliving his days as a product manager, as he’s taken to weighing in directly on the details of product features, a task that would usually fall far below his pay grade, according to one former employee. Google co-founders Larry Page and Sergey Brin have also gotten more involved in the company than they’ve been in years, with Brin even submitting code changes to Bard, Google’s ChatGPT-esque chatbot. Senior management has declared a “code red” that comes with a directive that all of its most important products—those with more than a billion users—must incorporate generative AI within months, according to a person with knowledge of the matter. In an early example, the company announced in March that creators on its YouTube video platform would soon be able to use the technology to virtually swap outfits.

Some Google alumni have been reminded of the last time the company implemented an internal mandate to infuse every key product with a new idea: the effort beginning in 2011 to promote the ill-fated social network Google+. It’s not a perfect comparison—Google was never seen as a leader in social networking, while its expertise in AI is undisputed. Still, there’s a similar feeling. Employee bonuses were once hitched to Google+’s success. Current and former employees say at least some Googlers’ ratings and reviews will likely be influenced by their ability to integrate generative AI into their work. The code red has already resulted in dozens of planned generative AI integrations. “We’re throwing spaghetti at the wall,” says one Google employee. “But it’s not even close to what’s needed to transform the company and be competitive.”

In the end, the mobilization around Google+ failed. The social network struggled to find traction with users, and Google ultimately said in 2018 that it would shutter the product for consumers. One former Google executive sees the flop as a cautionary tale. “The mandate from Larry was that every product has to have a social component,” this person says. “It ended quite poorly.”

A Google spokesperson pushes back against the comparison between the code red and the Google+ campaign. While the Google+ mandate touched all products, the current AI push has largely consisted of Googlers being encouraged to test out the company’s AI tools internally, the spokesperson says: a common practice in tech nicknamed “dogfooding.” Most Googlers haven’t been pivoting to spend extra time on AI, only those working on relevant projects, the spokesperson says.

Google is not alone in its conviction that AI is now everything. Silicon Valley has entered a full-on hype cycle, with venture capitalists and entrepreneurs suddenly proclaiming themselves AI visionaries, pivoting away from recent fixations such as the blockchain, and companies seeing their stock prices soar after announcing AI integrations. In recent weeks, Meta Platforms CEO Mark Zuckerberg has been focused on AI rather than the metaverse—a technology he recently declared so foundational to the company that it required changing its name, according to two people familiar with the matter.

The new marching orders are welcome news for some people at Google, who are well aware of its history of diving into speculative research only to stumble when it comes to commercializing it. Members of some teams already working on generative AI projects are hopeful that they’ll now be able to “ship more and have more product sway, as opposed to just being some research thing,” according to one of the people with knowledge of the matter.

In the long run, it may not matter much that OpenAI sucked all the air out of the public conversation for a few months, given how much work Google has already done. Pichai began referring to Google as an “AI-first” company in 2016. It’s used machine learning to drive its ad business for years while also weaving AI into key consumer products such as Gmail and Google Photos, where it uses the technology to help users compose emails and organize images. In a recent analysis, research company Zeta Alpha examined the top 100 most cited AI research papers from 2020 to 2022 and found that Google dominated the field. “The way it has ended up appearing is that Google was kind of the sleeping giant who is behind and playing catch-up now. I think the reality is actually not quite that,” says Amin Ahmad, a former AI researcher at Google who co-founded Vectara, a startup that offers conversational search tools to businesses. “Google was actually very good, I think, at applying this technology into some of their core products years and years ahead of the rest of the industry.”

Google has also wrestled with the tension between its commercial priorities and the need to handle emerging technology responsibly. There’s a well-documented tendency of automated tools to reflect biases that exist in the data sets they’ve been trained on, as well as concerns about the implications of testing tools on the public before they’re ready. Generative AI in particular comes with risks that have kept Google from rushing to market. In search, for instance, a chatbot could deliver a single answer that seems to come straight from the company that made it, similar to the way ChatGPT appears to be the voice of OpenAI. This is a fundamentally riskier proposition than providing a list of links to other websites.

Google’s code red seems to have scrambled its risk-reward calculations in ways that concern some experts in the field. Emily Bender, a professor of computational linguistics at the University of Washington, says Google and other companies hopping onto the generative AI trend may not be able to steer their AI products away “from the most egregious examples of bias, let alone the pervasive but slightly subtler cases.” The spokesperson says Google’s efforts are governed by its AI principles, a set of guidelines announced in 2018 for developing the technology responsibly, adding that the company is still taking a cautious approach.

Other outfits have already shown they’re willing to push ahead, whether Google does or not. One of the most important contributions Google’s researchers have made to the field was a landmark paper titled “Attention Is All You Need,” in which the authors introduced transformers: systems that help AI models zero in on the most important pieces of information in the data they’re analyzing. Transformers are now key building blocks for large language models, the tech powering the current crop of chatbots—the “T” in ChatGPT stands for “transformer.” Five years after the paper’s publication, all but one of the authors have left Google, with some citing a desire to break free of the strictures of a large, slow-moving company.

They are among dozens of AI researchers who’ve jumped to OpenAI as well as a host of smaller startups, including Character.AI, Anthropic and Adept. A handful of startups founded by Google alumni—including Neeva, Perplexity AI, Tonita and Vectara—are seeking to reimagine search using large language models. The fact that only a few key places have the knowledge and ability to build them makes the competition for that talent “much more intense than in other fields where the ways of training models are not as specialized,” says Sara Hooker, a Google Brain alumna now working at AI startup Cohere.

It’s not unheard of for people or organizations to contribute significantly to the development of one breakthrough technology or another, only to see someone else realize stupefying financial gains without them. Keval Desai, a former Googler who’s now managing director of venture capital firm Shakti, cites the example of Xerox Parc, the research lab that laid the groundwork for much of the personal computing era, only to see Apple Inc. and Microsoft come along and build their trillion-dollar empires on its back. “Google wants to make sure that it’s not the Xerox Parc of its era,” says Desai. “All the innovation happened there, but none of the execution.”

© 2023 Bloomberg LP