Google is faltering. After years of single-minded worship of the false god Virtual Assistant, the company is rushing its AI strategy as its competitors join hands and raise their pitchforks. The irony is that it’s all happening because Google thought it had cornered the pitchfork market.
See, in 2017 Google researchers published the paper “Attention is all you need,” introducing the concept of the transformer and vastly improving the potential capabilities of machine learning models. You don’t need to know the technical side of it (and indeed I’m not the one teaching it to you), but it’s been hugely influential and powerful; let it suffice to say it is the T in GPT.
You may be wondering why Google gave this wonderful thing away for free? While large private research firms have historically been criticized for withholding their work, the trend over the last few years has been toward publishing. This is a prestige game and also a concession to the researchers themselves, who prefer their employers not to hide their light under a bushel. There’s probably an element of hubris to it too: After inventing the technology, how can Google not make the best use of it?
The capabilities we see today in ChatGPT and other major language models didn’t follow right away. It takes time to understand and take advantage of a new tool, and every major tech company got to work exploring what the new AI era could offer and what it would take.
Assisting the Assistant
There is no doubt that Google devoted itself to AI work just like everyone else. In the years that followed, it made serious strides in AI computing hardware design, built useful platforms for developers to test and develop machine learning models, and published tons of articles on everything from esoteric model tweaks to more recognizable things like speech synthesis.
But there was a problem. I’ve heard this anecdotally from Google employees and others in the industry, but there’s a sort of feudal aspect to the way the company operates: getting your project under the auspices of an existing major product, like Maps or Assistant, is a reliable way to get money and staff. And so it seems that despite having gathered many of the best AI researchers in the world, their talent was channeled into the ruts of corporate strategy.
Shall we see how that turned out? Here’s a (admittedly selective) little timeline:
In 2018, they showed off incremental improvements to the Google Assistant stream, Photos (things like colorizing monochrome images), a smart display with a “visual-first version of Assistant” (have you ever seen it?), Assistant in Maps , AI-assisted Google News and (to their credit) MLKit.
In 2019 a rebranded and bigger smart display, AR search results, AR maps, Google Lens updates, Duplex for the web (remember Duplex?), a compressed Google Assistant that does more local, Assistant in Waze, Assistant in driving mode, live subtitling and live relay (speech recognition) and a project to better understand people with speech disabilities.
To be sure, some of these things are great! Most of them, however, were just an existing thing, but with a boost from AI. In retrospect, many feel a little cringe. You really see how large companies like Google are guided by trends and also stimulate them.
Image Credits: businessroundups.org
Meanwhile, in February of that year, we also had the headline, “OpenAI Built a Text Generator So Good It’s Deemed Too Dangerous to Release.” That was GPT-2. Not 3, not 3.5… 2.
In 2020, Google created an AI-powered Pinterest clone, then fired Timnit Gebru, one of the leading voices in AI ethics, in December over a paper pointing out the limits and dangers of the technology.
To be fair, 2020 hasn’t been a great year for many people – with the notable exception of OpenAI, whose co-founder Sam Altman had to personally mute the hype for GPT-3 because it had grown above sustainable levels.
In 2021, Google’s own major language model, LaMDA, debuted, though the demos didn’t really sell it. Presumably they were still casting for a reason other than making sure Assistant makes fewer mistakes.
OpenAI kicked off the year with the show of DALL-E, the first version of the text-to-image model that would soon become a household name. They had begun to show that LLMs, through systems like CLAMP, can perform more than just language tasks and rather acted as an all-encompassing interpretation and generation engine. (To be clear, I don’t mean “artificial general intelligence” or AGI, just that the process worked for more than a preset set of verbal commands.)
In 2022, more tweaks to Assistant, more smart displays, more AR in Maps, and a $100 million acquisition of AI-generated profile photos. OpenAI released DALL-E 2 in April and ChatGPT in December.
At some point, I suspect early 2022, Google executives opened their eyes and what they saw scared the hell out of them. I imagine the scene in Lord of the Rings where Denethor finally faces the assembled armies of Mordor. But instead of losing their minds and being made up by a wizard, these frenzied VPs sent emails asking why some cocky startup was running circles around the world leader in AI. Especially after they practically invented the means to do this.
Evidence for this is Imagen’s exit a month after DALL-E 2, though like practically every other interesting AI research Google published, it wasn’t available for anyone to test out, let alone connect to an API. When Meta released Make-A-Video in September, Google responded with Imagen Video a week later. Riffusion made waves for music generation, and a month later here comes MusicLM (which you can’t use).
But it was certainly ChatGPT that caused Google’s leadership to quickly move from fear to full sweat.
It would have been obvious to everyone involved that this kind of conversational AI was categorically different from the Assistant products Google had invested in for a decade, and was Actually do what everyone else’s pseudo AIs are (basically just natural language front ends to a collection of APIs) pretended Unpleasant. That is what is called an existential threat.
Fortune or foresight?
Image Credits: Microsoft/Open AI
Now it was bad enough that someone else, a new entrant immune to acquisition, had triggered the next stage of evolution for the search engine, and they had done it in a very public way that captured the imagination of everyone, from industry leaders to the tech-avoidant. The real twist of the knife came out unexpectedly Microsoft.
Calling Bing a “rival” to Google Search might be too generous – with about 3% of global searches compared to Google’s 92%, Bing is more of a wealthy gadfly. Microsoft seems to have given up any illusions about Bing’s ability to improve its standing and has looked outside their own homes for help. Whether their investment in OpenAI was supernatural foresight or happy serendipity, at some point it became clear that they had backed a fast horse.
Perhaps Satya Nadella and Sam Altman conspired in a smoke-filled room to exclude Google from their new world order, but in public the conversation took the form of money, and lots of it. Whatever the backstory, Microsoft had secured its allegiance to the innovative newcomer and with it the ability to put its technology to work where it would do the most good.
While we’ve seen some interesting ideas about how generative AI can help with productivity, coding, and even management, they’ve yet to be proven, due to copyright concerns or AI’s tendency to be a bit too “creative” in its reactions. But with the right guardrails, it was clearly very good at synthesizing information to answer almost any question, from simple factual questions to complex philosophical questions.
Search combined Microsoft’s need to innovate to move forward with a core competency of major language models, which by chance or common sense had just established as a partner the world’s foremost maker. The move to the latest GPT model (some call it GPT-4, but I suspect OpenAI will reserve that name for its own first-party model) with Bing and Edge is a kind of forced hail mary, the last and greatest game in the search engine world.
Google, clearly upset, attempted a spoiler campaign with an empty blog post the day before Microsoft scheduled its big event to announce the OpenAI-powered Bing. Bard, apparently the name of Google’s LaMDA-based ChatGPT competitor, was revealed in what is now typically parsimonious. Promises of possibilities and no hard dates or access plans.
This attempted announcement seems to have been made so quickly that its content was barely mentioned at Google’s “Search and AI” event two days later, and indeed it also escaped the sort of fact-checking you’d want to do if you were promoting the future of the knowledge graph. The image used to illustrate Bard contained a non-trivial error and said the James Webb Space Telescope “took the first-ever pictures of a planet outside our solar system.” This is not true, and the fact that this vaunted machine intelligence was wrong and that no one at Google noticed or cared enough to check, seems to have spooked investors.
ChatGPT certainly has issues, and indeed immediately after the rollout of Microsoft’s improved Bing, businessroundups.org was able to get the supposedly safe and appropriate AI to improvise an essay by Hitler and then regurgitate vaccine disinfo that an earlier version of itself wrote last month. But these are blemishes on an established record of billions of prompts and calls made, much to the satisfaction of users.
Google rushing its shot and stumbling so visibly speaks of a lack of preparedness, even on a narrow, experimental level — let alone a global rollout like the one Microsoft has already begun.
In his call to investors, CEO Sundar Pichai said, “I think I see this as an opportunity to rethink and reinvent and drive Search to also solve more use cases for our users. It’s still early days, but you will see that we are bold, get things out there, get feedback and iterate and improve things.” Does that sound like a man with a plan?
Understandably, Google wouldn’t want to slaughter the golden goose by prematurely merging Search with the half-baked general-purpose LLM they have lying around. They have become experts at deploying highly specialized AI, task models that do one or two things. But when it comes to making a big move, their comfortable position has saddled them with inertia.
Is it the downfall of Google? Of course not, it remains the standard and a fabulously profitable, slightly ridiculous venture for the foreseeable future. But investor confidence has been damaged as it appears that Google’s failure to innovate meaningfully in recent years may have stemmed not from wisdom and confidence, but from restraint and pride. (The FTC and Justice Department making another attempt to improve their advertising efforts can’t help either.)
However, this turn of the worm is only in its first few degrees and we shouldn’t speculate too far when the technology in question has yet to prove itself as valuable as everyone wants to believe. If not, the entire tech industry will suffer the consequences, not just Google.
