Google vs Microsoft (Part 2): Google's Bard May Prove Everyone Wrong
Google vs Microsoft (Part 2): Google's Bard May Prove Everyone Wrong
Caution could be the company's best friend—or worst enemy
This is part 2 of a three-part article series covering the recent news on AI involving OpenAI, Microsoft, Google, ChatGPT, the New Bing, and Bard, and how the events will unfold into a new era for AI, search, and the web (part 1 & part 3).
Last week I published a (very long) post on Microsoft, its relationship with OpenAI and ChatGPT, and the announcement of a new suite of products, including the new Bing and Edge. I also dove into the potential consequences at the technical and application levels; are language models (LMs) really apt for search? I finished with a short section on what Microsft really wants with all this—is this a move to leverage the latest generative AI technology or a warning to Google?
Today I bring you the second part of this series, featuring Google’s perspective. How Microsoft and OpenAI’s recent announcements have forced Google to act, what the search giant plans to do, why it has been so cautious until now, and whether it can save its stake in AI and search against these two opponents. It’s divided into four subsections:
How Microsoft and OpenAI have awakened a sleeping giant
Bard—Google’s ace up the sleeve to combat ChatGPT and the new Bing
Four keys help explain why Google has been so cautious—and why it may be the best strategy after all
Are we living through a paradigm shift in how companies do AI?
How Microsoft and OpenAI have awakened a sleeping giant
Google has been for years synonym with state-of-the-art AI research. Throughout the last decade, the company has been, by far, the most notable industry player. The transformer, arguably its most influential contribution, now serves as the basis for virtually everything relevant going on in the field of language processing and understanding (GPT models, ChatGPT, LaMDA, Bard, etc. all belong to that branch).
But Google has also been a synonym for caution, restraint—and disappointment—for consumers and employees alike. The company has repeatedly failed to convert those uniquely profound insights into tangible products that people could use and benefit from. For one reason or another Google has decided to keep most of its developments under chains (I’ll go into the “whys” later).
It was a matter of time before a younger, less risk-averse company would take those learnings and transform them into revolutionary technology (for better or worse). That’s OpenAI (the other AI startup that could’ve faced Google was DeepMind. The highly-regarded AI lab was acquired by Google in 2014 and contrasts sharply with OpneAI due to its research-focused practices, as I explained here).
Microsoft, seeing the potential in OpenAI decided to take the chance and it paid off. Google and Microsoft (OpenAI) have been in a sort of truce for a long time as far as AI is concerned, but that’s gone. Microsoft’s ambition to reimagine search (regardless of whether in intention or only appearance) is too high a threat for Google. As it was OpenAI’s surprising—and successful—decision to ship ChatGPT on Nov 30, which Google’s execs deemed a “code red”.
OpenAI and Microsoft’s latest movements set the final scene of a plot arc that began with their partnership in 2019 when Google was the absolute leader in AI by any metric. Now, at least in the world’s eyes, the tides have turned. It’s Google’s turn to move. A long-time advocate of responsible AI (not unlike Microsoft, as long as it’s in line with their financial incentives), Google is being forced to act—or “dance”, as Satya Nadella, Microsoft’s Chairman and CEO, said last Tuesday during the announcement of the new Bing. Despite the execs’ intention to “get this right” and avoid the potential “reputational risk” that an ineffective chatbot-enhanced Google Search could entail, they seem to be rehearsing their dance moves.
One day before Microsoft’s event last week, Sundar Pichai, Google’s CEO, published a blog post outlining the company’s next steps and unveiling their short-term goal to ship a ChatGPT rival called Bard. Although it’s unclear whether Google will copy Microsoft’s decision to merge the search engine with an LM (i.e. Bard) or choose a different strategy, you can be sure they won’t sit around waiting for their fate.
Bard—Google’s ace up the sleeve to combat ChatGPT and the new Bing
Pichai’s Monday blog post and Google’s Wednesday demo event covered pretty much the same news on Bard, but there’s not much public info about it yet (people who have access are most likely under NDAs), so I’ll focus on the big announcement: Google will make Bard, “an experimental conversational AI service, powered by LaMDA … [that] seeks to combine the breadth of the world’s knowledge with the power, intelligence and creativity of our large language models … more widely available to the public in the coming weeks [emphasis mine],” wrote Pichai. “Next month, we’ll start onboarding individual developers, creators and enterprises so they can try our Generative Language API, initially powered by LaMDA with a range of models to follow.”
If Google proceeds as promised and opens Bard to the general public, it’d be the first LM product the company ships in years that’s directly usable by consumers. Google announced the (in)famous LaMDA in 2021 but only allowed constrained access to certain groups via the AI Test Kitchen. A broadly available Bard would be a milestone for Google—and strong evidence that things are changing for them and the AI industry as a whole.
People’s perception of Google: Boldnessless
But the public may not see this as a win for Google. I’ve heard people summarize the company’s event last Wednesday as a mix of “coming soon” and “in the next weeks” announcements. James Vincent wrote for The Verge that “[Google’s] features pale in comparison to Microsoft’s announcement yesterday of the ‘new Bing,’ which the company has demoed extensively to the press and offered limited public access to.” He’s spot on. Google’s long-time reluctance to transform research into products contrasts with the attractiveness of Microsoft’s recent releases and with OpenAI’s policy on shipping things first and fixing them later.
For now, Bard is only “open to Trusted Testers” because they want to ensure it meets “our high bar for quality, safety, and groundedness before we launch it more broadly,” said Prabhakar Raghavan, Google SVP in charge of Google Search. Google falls on the side of caution whereas Microsoft is bolder (people like it—at least when that boldness results in a new toy they can use).
This decisional disparity shapes public discourse: Google feels like it’s going down (and it is by some metrics, more on that soon) while Microsoft is ascending fast. But that feeling dies in the superficiality of the public narratives that now, because AI has gone mainstream, dominate the spaces of opinion, like social media and the press, and influence stock prices irrationally.
That’s precisely why I don’t think current views on Google vs Microsoft are based on the quality or performance of the companies’ tech. Bard is probably as good—or bad—as Microsoft’s Prometheus (if Google doesn’t finally integrate Bard with Search, it’ll make no sense to compare it with the new Bing directly. Bard vs Prometheus would be more appropriate except we know pretty much nothing about them). In the end, both Google and Microsoft have similar resources, knowledge, and talent. If they decide to play the same game, they’re pretty much tied up.
Google’s newfound willingness to make a state-of-the-art AI product and the undeniable fact that it has nothing to envy or fear Microsoft in the spaces they’re now fighting on (regardless of what people claim), makes Bard a big deal.
Bard’s “fatal” mistake
There’s another reason why people are hesitant to consider Bard a significant step forward for Google. The most commented part of the demo wasn’t the supposed capabilities of the incoming product nor the abilities of the underlying technology (I don’t blame the press, Google really didn’t try to show off so there wasn’t much to comment on) but a factual mistake Bard made during the demo. Astrophysicist Grant Tremblay was among the first to realize the error and posted this on Twitter:
The press wasted no time echoing the spicy news. Reuters reported it first: Google promised Bard “would help simplify complex topics, but it instead delivered an inaccurate answer.” The Washington Post entitled the article “Google showed off its new chatbot. It immediately made a mistake.” And CNN began its reporting with “Bard, which has yet to be released to the public, is already being called out for an inaccurate response it produced in a demo this week.” Although most outlets pointed out that unreliability is a generalized problem for LMs, not just Bard, they fixated on Google and not Microsoft—which is very far from being free of guilt.
Others, like James Vincent, looked at the bigger picture. In a more level-headed take, he noted that “[Bard’s] mistake highlights the biggest problem of using AI chatbots to replace search engines—they make stuff up,” but he didn’t just jab at Google. “Microsoft … has tried to preempt these issues by placing liability on the user,” he added, underscoring that the problem isn’t Google’s but inherent to the technology both companies are using—and that it’s much more worrisome when integrated into search engines, which Google may not do after all.
Bard’s mistake cost Google more than $100B (later estimates put the number at $170B) although I wouldn’t be surprised if the press’ extensive reporting on the issue had a critical role here. When I read all those articles on Bard’s mistake and the subsequent drop in Google’s stock price, I realized something: most people don’t really know what we’re talking about when we say generative AI, LMs, or ChatGPT. As Matt Maley, chief market strategist at Miller Tabak + Co, told Time, “for a stock like Google to get knocked down this much, it just shows you that people aren’t even looking at the fundamentals.” People don’t know how these tools work or how they fail, which aggravates big tech companies’ questionable decision to integrate them with products that could reach hundreds of millions of users.
Just in case, let me say it again: All current LMs make stuff up. It’s a well-known and thoroughly documented issue. ChatGPT does it. Bard did it. And the new Bing will do it (already has, actually). Companies use techniques (e.g. RLHF) to mitigate the problem but so far there’s no way to prevent it completely. Using LMs to retrieve information reliably or as a source of knowledge is a bad use case.
I’m commenting on Bard’s mistake not because I think it’s worthy of being in the news (it isn’t in the sense “hey, Bard made a mistake,” although it may have some relevance in the sense “Google didn’t take enough care to catch a mistake they know very well these systems make”), but because it’s interesting how superficial narratives strongly override the underlying technical reality as well as people’s critical thinking.
As I explained in the previous section, Google and Microsoft are playing with the same tech (part of which was invented by Google engineers), so their products will incur the same flaws. That something like an LM’s factual mistake could make Google lose $100B+ in market value reflects how AI going mainstream is having wild effects no one could’ve predicted and, in many cases, driven not by the value of the tech or the product, but by people’s emotional and irrational perceptions. Today it’s Google’s loss, let’s hope tomorrow isn’t everybody’s.
Four keys help explain why Google has been so cautious—and why it may be the best strategy after all
Excluding Bard, the most notable aspect that permeates all of Google’s decisions about AI products is the caution the company has displayed throughout the years. People mock Google as being unable to ship anything anymore, too afraid to make something that could be unsafe, or too focused on AI ethics—well…
Google’s employees who created the technology that now OpenAI and Microsoft have turned into an appealing product decided to leave the company to run their AI startups instead. Aidan Gomez co-founded Cohere, Ashish Vaswani and Niki Parmar co-founded Adept, and Noam Shazeer is Character.ai’s CEO—all of them are co-authors of the famous “Attention is all you need” 2017 transformer paper.
But why is Google acting like this? Is it really an unbreakable commitment to responsible AI practices? The inability to ship AI products at a scale no other company has to face? Maybe the unwillingness to do so with products they can’t ensure will work out well enough? Maybe they didn’t want to disrupt the space where they’ve been leaders for so long? Or maybe this is all part of a bigger strategy and they’re simply waiting patiently to strike hard. Let’s unpack all this.
A two-faced relationship with AI ethics
Google's actions reveal the company’s priorities. Even if they’re also pressured by other, more interested forces, the end result is that Google hasn’t made a search engine powered by LMs at a time when they’re completely unreliable for the task—they’re acting more responsible than Microsoft and OpenAI (even though the latter also adhere—or so they say—to strict AI ethical and responsible principles).
Google execs prioritize doing things right over doing them fast or first. OpenAI had little to lose and a lot to gain with ChatGPT, however, the decision to ship the chatbot may not align with the startup’s long-term mission. It’s hard to marry an idealistic ultimate purpose (i.e. building AGI) with the financial necessities to survive during the journey, but that only explains—not excuses—OpenAI’s decision to do things less ethically than they purported to. The same applies to Microsoft. In their attempt to take some market share from Google, they’ve shipped a product that not only fails a lot, they knew it did: During the Q&A last week Sarah Bird, Microsoft’s responsible AI Lead, acknowledged Bing’s shortcomings but she made it sound like the problems were minimal. It doesn’t look like it. So much for adhering to AI principles.
But I don’t believe for a second that Google is acting more responsibly than OpenAI or Microsoft because they’re more altruistic. We can’t ignore Google execs’ past history of conflicts with the company’s AI ethics team. AI ethics leads Timnit Gebru and Margaret Mitchell were fired in 2020 and 2021 respectively when they raised the very concerns about LMs that we’re now seeing appear in ChatGPT, Bing, and Bard. They warned Google. And they were right, both in terms of the tech’s shortcomings and in terms of the implicit consequences that would eventually come if a big tech company shipped a half-baked LM-enhanced product. That doesn’t sound very responsible to me.
Maybe AI ethics isn’t really Google’s top priority. Maybe it simply happens that Google’s responsible AI principles align now with the other, stronger incentives (if we look at the records when financial needs have clashed with ethical issues, the former have always trumped the latter). I think there’s some honesty in Google’s cautious approach to AI products, but I don’t buy the idea that the company’s commitment is the only factor at play—or that it’s the main driver of its decisions.
Google has strong reasons to research and develop AI without going to production beyond ethics. The company holds a hegemonic position in search; why would they disrupt the space where they’re leaders? Also, why would they risk appearing as the bad guy publicly in case the product flops? As far as I can tell Google’s inaction may respond more to financial and reputation needs rather than to ethical principles—without denying the existence of the latter.
As it turns out, Google was right to hold back all this time. John Hennessy, Alphabet Chairman, told CNBC that “Google was hesitant to productize this because it didn’t think it was really ready for a product yet.” He has a point. It’s not been a week since Microsoft released the new Bing and we're already seeing all the things “critics” and “skeptics” have warned about for years. The first one below is quite telling. That’s not how a search engine should work:
If it’s working, don’t fix it
One other reason that better explains Google’s actions (or, better, inaction) is, as I wrote above, that they’re leaders in the space Microsoft is trying to disrupt. Why change anything when it’s working just fine as it is? And, more importantly, when the change may threaten the company’s main source of revenue?
Combining Google Search with Bard before Microsoft announced the new Bing may have resulted in significant user attrition (everyone is already using Google Search—they can only go elsewhere). People wouldn’t have seen Google as a first-mover positively—more so given that the product doesn’t work well and Google knows it. Microsoft had the incentive to move fast with a half-baked product to try and take some market share from Google Search (despite the rise in Bing downloads, it hasn’t happened—Google stays strong at 93% of the market share).
Google's decision to keep a conservative stance and avoid inadvertently wiping out its main source of revenue was the logical move. But they knew others would try—Microsoft’s $1B investment in OpenAI in 2019 was a clear sign. They had to prepare to face any potential new player that could eventually threaten to dethrone them by means of sufficiently innovative technology (like generative AI). And they did.
The innovator’s dilemma
The innovator’s dilemma—how large, established companies can lose market share to smaller, less risk-averse companies that can dare leverage innovative technologies against which established players can’t do anything fast enough—was what Google feared most, and what it would have faced had it not prepared adequately.
Clayton Christensen, who coined the term, recommends that “large companies maintain small, nimble divisions that attempt to replicate this phenomenon internally to avoid being blindsided and overtaken by startup competitors.” Google did. And they didn’t just “replicate” others. They set up the Google Brain + DeepMind unbeatable combo. They did it first and better than anyone else. It wasn’t Google trying to replicate OpenAI, but OpenAI replicating Google. No one would’ve predicted Google was badly positioned to face a ChatGPT-like threat.
Two key factors played a role here that Google couldn’t defend against—even with the strongest AI labs and talent in the whole world. First, ChatGPT’s unprecedented success. AI moves so fast and so unpredictably—even for the well-prepared Google—that not even OpenAI predicted it. The “code red” declaration was “akin to pulling the fire alarm,” wrote NYT’s Kevin Roose. “Some fear the company may be approaching a moment that the biggest Silicon Valley outfits dread—the arrival of an enormous technological change that could upend the business.”
And second, even if Google had a ChatGPT-like chatbot (to build one they only had to slightly tweak LaMDA, which existed since mid-2021), it’s not easy to protect the company from “startup competitors” when the very technology you’re pioneering to do so clashes frontally with your main source of revenue. The possibility of a working combination of LMs and search engines would’ve forced Google to choose between embracing AI fully, risking losing the fitness of the ad business model they’ve been milking for 20 years, and dismissing it, risking losing a big chunk of market share to someone else who could afford to run the product at a loss, without a viable business model—and without the worry to lose customers—with the sole intention to choke the leader to death (i.e. Microsoft).
But even now, with the danger of the innovator’s dilemma playing out, Google is still in a good position. They did everything up until the point of shipping a real product. Then they stopped. They waited. They still have cards left to play.
Risk big, win big
The obvious factor I’ve not yet mentioned is that Google and Microsoft, although similarly powerful overall, are extremely distant in terms of search market share. Google enjoys, at the time of writing, a ~30x market share dominance over Microsoft (93% vs 3%). By this metric, we shouldn’t even be comparing them.
Anything Google does in the space will have a Tsunami-like repercussion compared to the small breeze waves that Microsoft may cause (Microsoft’s shares rose and Bing’s downloads and users increased, but not significantly). This is important for two reasons: First, Google’s product, if shipped, would reach much more people than Microsoft’s. If the product is successful, this is an absolute win for Google even if they ship weeks or months after Microsoft (of course, they’d still have to rethink the business model).
But, second, if the product turns out to be problematic due to LMs’ lack of reliability, the outcome could be disastrous for the company. Microsoft is forcing Google to play a game of high risk, high reward in a space where Microsoft’s stake is minimal. Yet, by going second into the arena, Google can very closely watch Microsoft’s mistakes.
If the product works sufficiently well, Google could catch up fast and retain its market share and revenue almost intact. If the product flops, Google’s cautionary stance will pay off and the company will easily redirect its efforts, avoiding reputational damage and unnecessary costs. That’s why Google’s best chance was to prepare and wait.
Are we living through a paradigm shift in how companies do AI?
Now the question is, do people agree that LMs + search may not be a good idea yet? Do investors and shareholders agree? Maybe the AI industry hyped the tech too much and now Google’s dilemma is between shipping a half-baked product—with all the consequences it implies—and not satisfying those who keep the company’s pockets full.
Google faces a decision that Microsoft doesn’t: What if customers love the product even if it doesn’t work well? Although it sounds absurd, I can see it happening with generative AI. For instance, it happens with ChatGPT—although it’s not designed for search, people were already using it to replace Google—and, in the case of the new Bing, most people won’t stop to check the sources—even when there are many reports that claim it makes up stuff. If this is the case, Google may not have an alternative other than shipping a bad-working product just to satisfy users. In either case, the company could lose big.
We may be living through the beginning of a paradigm change not just for search but in the way companies do AI research, development, and production. Speed over caution. Attractiveness and superficial appeal over robust technology. In the next and final piece for this series, I’ll explore what all this means for the future of the AI industry as a whole—and the disastrous consequences that it could entail.
Very interesting post. I just happened to read a twitter thread from Steven Sinofksy which approaches this from a somewhat different angle: https://twitter.com/stevesi/status/1625622236043542530 . He seems to view the big tech companies (Google, MSFT, etc.) as inherently risk averse, and so he thinks that the most interesting use cases for LLMs (or AI more generally) won't come from them. Rather, he expects innovation to come from smaller, less risk averse, companies.
to be precise: the mistake of bard cost google shareholders, not google