Good morning,
Alas, it is time for me to say goodbye to my Milwaukee Bucks for the year; I have to say, I much prefer closing the season with a championship, like last year!
On to the interview:
Google I/O
From the New York Times:
There was a time when Google offered a wondrous vision of the future, with driverless cars, augmented-reality eyewear, unlimited storage of emails and photos, and predictive texts to complete sentences in progress.
A more modest Google was on display on Wednesday as the company kicked off its annual developer’s conference. The Google of 2022 is more pragmatic and sensible — a bit more like its business-focused competitors at Microsoft than a fantasy play land for tech enthusiasts. And that, by all appearances, is by design. The bold vision is still out there — but it’s a ways away. The professional executives who now run Google are increasingly focused on wringing money out of those years of spending on research and development.
The company’s biggest bet in artificial intelligence does not, at least for now, mean science fiction come to life. It means more subtle changes to existing products.
It’s kind of cliche at this point to look to the New York Times for the hater’s take on any set of new technology announcements, but what makes this article particularly striking is that first sentence: Google literally demo’d augmented-reality eyewear — it seems weird to cite that as something the company no longer talks about when they literally talked about it! Still, as John Gruber noted, it was vaporware:
It’s kind of funny watching them go from things definitely shipping in a few months (Pixel 7 phones) to something they plan to ship in “2023” (a Pixel tablet, which looks a lot like an iPad but has the front-facing camera on the long side) to something with no timeline of actually shipping (AR glasses that show live translations of what people are saying to you in languages you don’t understand, like subtitles for the real world). Never a good sign when your most interesting demo is the most nebulous.
It’s a fair take, but in this case, I think Google deserves the benefit of the doubt for reasons that the New York Times article stumbled upon: Google I/O was compelling precisely because it was rooted in the present, and what made the future products compelling was you could see the path from here to there.
This clarity started with last year’s Google I/O; I noted a month later in response to another New York Times’ piece casting doubt on CEO Sundar Pichai’s leadership:
And, in fact, I would argue that Pichai has done better than not screw up: I think that Google’s increase in focus — what some disgruntled former executives might call a lack of ambition — has been good for the company. Google today has five clear product lines — Search, YouTube, Android, Workspace and Cloud — all of which still have clear growth opportunities. Moreover, as I noted in a Daily Update after I/O, I think that Pichai has clearly put his imprint on Search in particular, pushing Google towards an ambitious goal that only Google could achieve: being a true answer machine, not simply an index.
At last year’s I/O Google announced its new “Multitask Unified Model” for search, which combined search’s knowledge graph, language models, translation capabilities, and had the ability to work across different mediums; this video of Senior Vice President of Search Prabhakar Raghavan is from last year:
What makes this technology ground-breaking is its ability to multitask in order to unlock information in new ways. Here are a few tasks it can handle at the same time:
- It can acquire deep knowledge of the world
- It can understand language and generate it to
- It can train across 75 languages at once, unlike most AI models which train across one language at a time
And then, what makes MUM even more amazing, is that it is multimodal, which means that it can simultaneously understand different forms of information like text, images, and videos.
Raghavan followed up that video with a stylized demonstration of MUM’s capabilities; this year’s I/O had much more tangible examples:
We’re combining our understanding of information across multiple modes to help you express your needs more naturally than ever before. Just last month we launched multisearch, one of our most significant updates to search. In the Google app, you can now search by taking a photo and asking a question at the same time. You can snap a pic of a spill-proof water bottle and ask for one with rainbows on it to brighten your kid’s day. Or in my case, I was able to take a picture of my leaky faucet and order the part to fix it; the funny thing is I still don’t know what the part is called.
This is pretty futuristic stuff, even though it’s subtle; what is also noteworthy is how this capability went from demo a year ago to shipping today. That lends more credence to other new technologies, like this explanation from Pichai of how Google might explain New York pizza preferences in Bengali:
Just like before, we prompt the model with two examples of question with both Bengali and English answers. That’s it. Now we can start asking questions in Bengali.
“What’s the national song of Bangladesh?” The answer, by the way, is “Amar Sonar Bangla”, and PaLM got it right too. This is not that surprising because you would expect that content would exist in Bengali. But you can also try something that is less likely to have related information in Bengali, such as, “What are popular pizza toppings in New York City?” And the model again answers correctly in Bengali, though it probably just stirred up a debate amongst New Yorkers about how correct the answer really is.
This capability is just now being developed, but it’s a clear next step beyond what Google is already doing today; that, by extension, makes the company’s AR glasses demo, which only came at the very end of the keynote, more compelling than it might have been in isolation:
This was, in my estimation, the best AR (or VR) demo any tech company has put out by far. It’s not because the glasses look cool (they’re not bad, but almost certainly years away from shipping), but because the use case was so compelling. It’s not just that the use case was compelling, though: it’s because Google has, for two years in a row now, both described a path to this use case, and, more importantly, actually shown that it is progressing down said path.
“Near Me”, Competition, and Innovation
To jump back to Raghavan’s demo, there was one more bit of capability he showed off — adding “near me” to multisearch:
Later this year, we’ll add a new way to search for local information with “Multisearch Near Me.” Just take a picture or long-press one you see online, add near-me, and find what you need from the millions of local busiensses we serve with Google. “Near me” will work on multisearch for everything from apparel to home goods to my personal favorite, food in local restaurants.
So, let’s say I spot a tasty looking dish online. I don’t know what’s in it, or what it’s called, but it’s making me hungry. With this new capability, I can quickly identify that it’s japchae, a Korean dish, find nearby restaurants that serve it, and enjoy it in no-time.
Raghavan explained how it worked:
Here’s what’s happening under the hood: Google’s multi-modal understanding recognizes the visual intricacies of the dish and combines it with an understanding of my intent: I’m looking for local restaurants that serve japchae. It then scans millions of images and reviews posted in web pages and from our active community of maps contributors, to find results about nearby spots.
Back in 2013 Google reached a settlement with the FTC in which it avoided an antitrust lawsuit about favoring its own properties in search; one of the more egregious actions uncovered in that investigation was how Google has been scraping sites like Yelp and Trip Advisor to populate its own local search product.
This was, to be clear, very not cool (that’s a legal term 😂); but then, once Google stopped, Yelp in particular would point out to anyone who would listen that Google was being a bad actor because they were serving worse results than they might have had they loaded Yelp content instead (Yelp has long proposed that Google have some sort of modular search interface where they are forced to load content from third parties; the local recommendations company has yet to convince me why this is necessary given that consumers can simply go to yelp.com or download the Yelp app if the content is so much better).
Two things have happened over the last decade, though. First, Google local has gotten a lot better, at least in my experience. That noted, this should be taken with a very large grain of salt, because not only is it anecdotal, but it’s also mostly based on my experience in Taiwan, where Yelp has almost no penetration. The second thing, though, is this new capability for Google search: the advantage of Google having its own local search product for the past decade is that it has slowly but surely accumulated its own collection of content, including reviews and photos. Thanks to that content, the company can now offer a completely new kind of search that combines image recognition, text, and location. That’s pretty cool!
I get that this doesn’t make Yelp or its vertical-search peers feel any better; Google obviously leveraged its position as the most-used search engine to build up a corpus of content that lets it do things that its competitors are not yet capable of. At the same time, this also seems like a pretty obvious example of how consumer welfare can significantly increase from the innovation the flows from such an arrangement, and, by extension, why regulations like the E.U.’s Digital Markets Act — which may result in something similar to Yelp’s proposal, wherein Google is only allowed ten blue links — will result in worse consumer experiences in the long run.
Pixel and Fitbit
Two points on Google’s hardware announcements.
First, from The Verge:
After weeks of rumors and leaked photos, Google finally confirmed that the Pixel Watch is real. Today’s announcement is more of a tease than a full reveal, however, with the watch arriving later this fall alongside the Pixel 7.
I talked a bit about the Pixel Watch on last Friday’s Dithering (I don’t think that a round display works for digital content, for one); what I wanted to focus on here was the Fitbit integration. From that article:
The Fitbit integration is the most significant collaboration between the two companies we’ve seen since Google bought Fitbit for $2.1 billion≥According to Rick Osterloh, Google’s senior vice president of devices and services, this Fitbit integration will go beyond customizing watch faces and be “imbued throughout” the Pixel Watch experience. Users will be able to sync their data with a Fitbit account, meaning they’ll be able to view it within the Fitbit app and on the web. The watch will use all of Fitbit’s latest algorithms for health and fitness.
As for data privacy, Osterloh said in a briefing that Fitbit and Google data will stay private and separate due to promises it made to regulators during the Fitbit acquisition, meaning that any health data collected on the Pixel Watch will remain under Fitbit’s purview, separate from Google. Osterloh also added that while the Fitbit team was deeply involved with the Pixel Watch, Fitbit still has plans for its own Wear OS watch and will continue making its own products.
Those regulators were European, and, well, this sounds pretty crappy for users! Does anyone actually want to have to maintain distinct accounts for their Watch overall and the fitness component specifically? Doesn’t this seem to give Apple a big advantage in the space? I can understand the privacy argument that Google should not have been allowed to acquire pre-acquisition Fitbit data, but once consumers know that Google owns Fitbit why shouldn’t consumers be able to decide if they are ok with that arrangement? This seems like another example of how design by regulation just ends up with a worse experience for consumers.
Secondly, while the substance of Google’s announcements wasn’t hugely different from past years — there was a new low-end Pixel phone, a sneak peek of a new high end Pixel phone, a Pixel tablet, and the Pixel Watch — I did like how the devices were placed in a higher level framework of “ambient computing”; David Pierce noticed the same thing at The Verge:
Google I/O feels unusually… coherent this year. Google is trying — harder than I can remember — to build products that not only work well but work well together. Search is becoming a multisensory, multi-device proposition that understands both who’s searching and what they’re really looking for. It’s also extending the search experience far beyond just questions and answers. It’s making Android more context- and content-aware so that your phone changes to match the things you do on it. It’s emphasizing natural interactions so that you can get information without memorizing a rigid set of commands. It’s building the hardware ecosystem it needs to make all that work everywhere and the software to match.
I’m still pretty skeptical about how much Pixel hardware Google is actually going to sell in the long run. What is much clearer is how much better Google can tell its broader search story because it has its own hardware. Moreover, that story isn’t just about today, but also the future: the other reason why the AR demo was compelling was because it came at the end of a presentation that not only included a path in software but, if you squinted, a path in hardware as well. Time will tell if the company will successfully traverse both; I certainly think they have a better chance at a design that works than do regulators.
This Update will be available as a podcast later today. To receive it in your podcast player, visit Stratechery.
The Stratechery Update is intended for a single recipient, but occasional forwarding is totally fine! If you would like to order multiple subscriptions for your team with a group discount (minimum 5), please contact me directly.
Thanks for being a subscriber, and have a great day!