Intel Consolidates its AI Teams (Mar 23, 2017)
Intel is announcing that it’s taking its various AI teams and consolidating them under Naveen Rao, who ran the AI company Nervana which Intel acquired last year. This feels like a sensible move from a company which has dabbled in AI in various parts of the organization but hasn’t become known as an AI powerhouse. Where things get slightly less credible is where Intel talks in the announcement about rallying the industry around a set of standards for AI as it has with past computational trends. Whereas Intel was a major player in personal computing, one of the examples it cites, it’s not nearly in the same position of influence with regard to AI, and so this feels like hubris rather than realism on what Intel’s role will be. Intel also talks, though, about bringing AI to more people, which sounds a lot like the “democratization of AI” message we’ve been hearing a lot from Microsoft lately, and which others including Google have also started parroting lately. This feels like it’s going to become an increasingly important theme in AI: less about individual companies owning capability and more about packaging up and making that capability available to anyone who wants to use it.
A.I. Expert at Baidu, Andrew Ng, Resigns From Chinese Search Giant – The New York Times (Mar 22, 2017)
This story is notable for two reasons. Firstly, Baidu especially and Chinese companies in general are often overlooked completely in discussions of who’s making big investments in AI and machine learning, and yet Baidu has made massive investments in this area, and recently hired former Microsoft exec Qi Lu to be its COO and to oversee its AI efforts. Secondly, despite Qi Lu’s recent arrival, the trend of former Silicon Valley execs joining big Chinese tech companies still has fewer long-term success stories than short-term fizzles, as this article points out. Both Hugo Barra and Andrew Ng’s move to Chinese companies were seen as highly symbolic, and as such it’s inevitable that their departures should be too. The big Chinese companies are doing good work, and in some cases pioneering new product and service categories, across a number of different areas, but attracting and keeping high-profile talent from the US (even those with ties to Greater China – Ng was born in the UK to parents from Hong Kong) remains tough.
via New York Times
Samsung has somewhat unexpectedly taken the wraps off its virtual assistant Bixby ahead of next week’s Galaxy S8 launch, where I’d expected it to be the main event from a feature perspective. Based on how Samsung is describing the feature, though, I think it’s merely trying to defuse some hype by downplaying expectations of what Bixby will and won’t be. (The hype was fueled in part by Samsung’s acquisition of Viv, which was a more traditional virtual assistant that Samsung acquired last year, but Bixby appears to be something less.) The description from Samsung is somewhat vague, but I think the approach actually has a lot of merit: every other assistant promises to be just that, implying a broad-based ability to meet needs, which inevitably leads to disappointment and frustration when it falls short, over-promising and under-delivering. Samsung looks like it will come at this from the opposite end, starting small and building up functionality over time, app by app, in a way that the voice interface is able to handle everything the touch interface does in the same app. That, incidentally, should be good for accessibility, something Android devices have always done less well than iPhones. But the big limit there as with Bixby overall is that if third party developers don’t support it, it won’t be very useful, and it the S8 ships with the Google Assistant users may just choose to use that instead. I’m very curious to see next week exactly how Bixby is invoked and how it compares to the more traditional assistant model. Samsung doesn’t have a great reputation in software and services, and I’m skeptical that it will get this right.
via The Verge
This feels like an extremely stupid move for Google. Though Google claims this wasn’t an ad, that’s utterly disingenuous, and inserting ads this early in the Google Home lifecycle (if ever) is a huge mistake – this is just the kind of thing that will put people off buying a Google Home, especially because it fits a narrative of Google only being interested in advertising. This is a hardware product, for which users have paid a decent price, and it shouldn’t be playing ads, especially without an opt-out – there is no indication that users would hear ads in any of the marketing material. I just tried my own Google Home to see if it would play this message, but it didn’t, suggesting that Google may have stopped playing the message. If so, good, but it never should have happened in the first place, unless Google wants to kneecap its own product this early in its competition with Amazon’s Echo.
via The Verge
Amazon makes it cheaper to host Alexa skills on AWS – ZDNet (Mar 16, 2017)
This is clever tie-in by Amazon of two of its valuable assets: its Alexa skills engine and its AWS cloud infrastructure. It’s offering developers of Alexa voice skills a better deal on hosting through AWS as a way to remove the barriers to developing smarter and more sophisticated skills for its Echo devices (and the small number of third party devices using Alexa). Amazon has touted its number of third party skills repeatedly since launching them as a sign of Echo and Alexa’s capability, but the reality is that many of those skills are very basic, and the model is clumsy to use. If it’s able to attract better skills to the platform, those numbers will start to be more meaningful as signifiers of the platform’s capabilities.
Apple’s Siri learns Shanghainese as voice assistants race to cover languages – Reuters (Mar 9, 2017)
One of the things that’s often missed by US writers covering Amazon’s Alexa and its competitors is how limited it still is in language and geographic terms. It only speaks English and German and the Echo range is only available in a handful of countries. Siri, meanwhile, just got its 21st country and 36th language, which reflects a long-time strength of Apple’s: broad global support. Apple News is a notable exception, which is only available in a few countries and one language, but almost all of Apple’s other products are available in a very long list of countries and territories, often longer than for other competing services. The article here is also interesting for the insights it provides into how each company goes about the process of localization, which is quite a bit more involved than you might surmise.
This is a good counterpart to the Time article from last week about Amazon working on voice identification in their respective home speakers. It points out the complications in providing such a feature, not least that heavy processing to make voices clearer will also tend to distort them and therefore make it harder to recognize and distinguish speakers. The article also makes clear, though, that these challenges are far from insurmountable, which leads me to believe that Amazon or Google or both will eventually figure this out. In fact, whichever does figure it out first could have a big advantage, because for a lot of the most useful features (calendar, emails, etc) individual profiles are critical. So much so that Google misleadingly included that exact use case in its I/O launch video last year.
via Fast Company
I’m actually linking to two articles here. The Atlantic article explicitly argues that there’s AI-washing going on and that lots of things are being described as AI which are in reality just computer programs. The Axios article is a little more neutral, pointing instead to a rapid rise in mentions of AI on earnings calls, and taking the upward trend more or less at face value, while attempting to explain what is enabling all the activity in the AI field at the moment. I think they’re both worth reading, but there’s no doubt in my mind that the AI term and its cousin machine learning are being over-applied at the moment, which risks devaluing true AI and real achievements it’s enabling. At least part of the rapid spike in AI mentions Axios cites is down to this over-application and a bandwagon-jumping mentality that serves no one well (least of all investors suckered into believing their investment is in AI and not just bog standard software). Unfortunately, there’s no way to stop this at this point, and I think the problem will only worsen, which will make it that much harder for companies to truly differentiate on the basis of AI claims. But it also reinforces my argument that companies really need to show and not just tell when it comes to AI – real user benefit and not AI capability itself is the key.
Yet another story about using either AI or deep learning (or both) to solve a real-world problem, from Google. This time, it’s an application miles away from any of Google’s current businesses (though perhaps a little relevant to some of the Other Bets), but the point is that Google is finding a very broad set of applications for its capabilities here, which can of course be applied back to lots of things which are relevant to the core Google business (as well as providing tangible human benefits if adopted by other organizations).
This feels like a somewhat gratuitous use of AI here by Netflix – maybe this is technically AI, but it’s hard to see how it’s not just image analysis. But the broader point here is that this is an often overlooked aspect of Netflix’s differentiation: its technical capabilities in video delivery. Yes, its investments in original content and its massive and rapidly growing scale globally are huge advantages over the competition, but its content delivery networks, compression techniques, and a host of other technical capabilities are also key to making its user experience better. And this is another area where it often feels like it will take competitors a long time to catch up even if they ever decide that’s strategically important.
The number in the headline refers to the acquisition price of Viv, a virtual assistant startup which Samsung bought a few months back and is expected to integrate into the Samsung S8 launching later this month. To put that number in context, it’s around the same amount Apple was reported to have paid to acquire Siri, and tiny in the context of Samsung’s overall business – it generated $180 billion in revenue last year, along with $25 billion in operating profit. So Samsung can far more easily afford this investment than, say, Xiaomi can afford its comparably-sized investment in in-house chip capability. But it’s still a decent chunk of money from Samsung in a year when it also announced the much larger Harman acquisition. Far more importantly, we haven’t yet seen what Viv will do when integrated into a Samsung phone, and whether it’ll be as good as the early hype around the standalone product suggested.
Amazon Echo May Get Voice ID Feature – Time (Feb 28, 2017)
From the first time I heard about Google Home at I/O last year, I assumed it would have multi-user support, and yet it didn’t. Now it sounds like it’s Amazon that may bring this feature to its home speaker first, which is yet another example of how Google seems to be punching below its weight in this fight. Google is all about individual user accounts: email, calendar, to-do lists, YouTube subscriptions, Android device identities and lots more are all tied up in personal accounts. Amazon, by contrast, probably works mostly at the level of the household, with families sharing Prime shipping and video accounts. So it’s ironic that Amazon would be the first to market with something that provides individual identification by voice. At the same time, I think there are going to be severe limitations around voice identification that may well make it inappropriate for anything security related – voice recordings are much easier than fingerprint cloning, for example. And in both the household I grew up in and my own home now, there were several people with very similar voices – it will be very important for Amazon (and Google) to be able to tell apart even voices with shared genes.
This, to my mind, is one of the bigger announcements coming out of MWC – that Google will finally allow other smartphone makers to use the Google Assistant, after several months of keeping it exclusive to its own Pixel smartphone. I described that decision at the time as representing a big strategic shift for Google, and probably a mistake, and the evidence since has borne that out. The Pixel has sold in small numbers, Amazon’s Alexa has extended its lead considerably as the voice platform of choice for hardware makers, and even at MWC itself Android vendors announced Alexa integration despite Google’s shift here. The good news is that it’s only been a few months, but the bad news is that this change in policy will come too late to hit the new flagships debuting at MWC, including the new ones from both Samsung and LG. It will likely become available later, but shipping as an integrated part of these new smartphones would have been much better. I’m betting that Google will continue to pay for this strategic misstep for some time to come – even once it’s available, OEMs will want to offer more differentiation than the Google Assistant allows them, which will continue to make Alexa an appealing alternative.
I love the term “Google cousin” to describe the non-Google companies under the Alphabet umbrella (though confusingly Jigsaw’s website makes it seem as if it’s actually part of Google despite no longer being called Google Ideas). The bigger point here is that this is a clever use of machine learning to solve a real problem, which I’m always a big fan of. Online comments can be horrible and very time consuming to moderate, and this API can be used by publishers to filter out the most “toxic” of those moments. Having said that, the sample comments Jigsaw shows to demonstrate the tool highlight just how inane most online comments are regardless of whether they’re actually toxic, calling into question for me at least whether they’re worth having at all. But this Perspective tool seems to be part of a broader push around technologies for increasing “safety” in various scenarios – that’s definitely the message you get at the Jigsaw website.
Inside Facebook’s AI Machine – Backchannel (Feb 23, 2017)
Backchannel (and Steven Levy in particular) seems to be becoming the default outlet (I was going to say channel) for these access-y pieces on AI and machine learning. Levy previously did something very similar for Apple last August, Amazon in November, and Google in October. And there continues to be a perceived need for this kind of thing because AI continues to be something that’s mostly talked about rather than seen by consumers. That’s not to say that it’s not in products – it clearly is, and the money quote from this article is that “Facebook today cannot exist without AI” – but that it’s not intuitively obvious to consumers that AI is behind a lot of what they use. Companies still need to tell their AI stories, particularly because narratives have emerged about Google being ahead or Apple being behind, and those narratives need to be countered. There are several interesting things in this particular article, but as that quote indicates the biggest thing that comes out of it for me is how central AI and machine learning are becoming to almost everything at Facebook. Secondarily, it’s interesting to see Facebook in some cases do complex processing on the phone itself, something Apple has pioneered but which others have largely eschewed in favor of cloud processing.
How Messenger and “M” Are Shifting Gears — The Information (Feb 22, 2017)
Facebook’s M assistant in its original conception was a virtual assistant a la Siri or Cortana which lived in Messenger, but one which was being trained by humans while it was available to a very limited number of users. Over time, it became clear that the process of handing off from humans to AI for the broad set of tasks M was supposed to be able to handle wasn’t going well, and it appears Facebooks somewhat went back to the drawing board on that. At the same time, the bot strategy within Messenger hasn’t gone well either, with limited developer and user adoption. Facebook now seems to have decided to combine these two failing projects into a new one which it presumably hopes will go better – M will pop up from time to time in Messenger conversions between friends to offer to complete certain tasks based on context. That’s probably a better, narrower use case for an AI assistant, but it also has serious potential to be creepy to users having what they will perceive to be a private conversation. And herein lies one of the biggest challenges with AI and bots – in order to be useful, they need to insert themselves into private conversations, which means they need to listen in on private conversations, much like Google’s advertising within Gmail has always been context based. In theory, only computers are eavesdropping, but that doesn’t stop people from objecting. I’m not convinced yet that this is the right answer either for Facebook’s M or bot strategies.
via The Information
Zuckerberg manifesto removes reference to Facebook monitoring ‘private channels’ – Business Insider (Feb 17, 2017)
Kudos to Mashable, which first noticed that one paragraph in a 6,000-word manifesto had been changed from the original to the final version (I covered the manifesto itself yesterday). And kudos, too, to Business Insider for following up with Facebook to find out why it was removed. The official explanation is that the paragraph talked too specifically about a capability Facebook hasn’t finalized yet, but it’s at least as likely that Facebook worried it would cause major privacy concerns. The paragraph in question talked about using AI to detect terrorists in private channels, which rather flies in the face of Facebook’s commitment to encryption and protecting privacy. As with much else in the letter, I think it was likely intended to be mostly aspirational rather than specific, but the original paragraph was rather tone deaf about how such an idea would be received even in such high-level terms.
via Business Insider
When most of your news about AI comes from the tech world, it’s easy to imagine that big tech companies are the only ones doing interesting things in the field, but here as in autonomous driving there’s also lots of amazing work being done in academia, as in this case. Carnegie Mellon researchers have developed a poker-playing AI which combines three different methods for learning the game and ultimately beating human players. The piece is worth reading for the details of how this was done, but it’s also a good reminder that neither any single tech company nor the tech industry as a whole has a monopoly on big breakthroughs in AI.
Apple Officially Joins Partnership on AI (Jan 27, 2017)
I commented on the reports that Apple was about to join the Partnership on AI yesterday, so I won’t revisit all of this today. Two notable things from today’s announcement, though: Apple’s representative will be Tom Gruber, who runs Siri at Apple, and that may be indicative of where Apple sees ownership of AI residing within the company (it has no formal head of AI); secondly, Apple has been involved with the Partnership from the outset, but hadn’t formalized its membership until today. That might signify that there were some details of Apple’s membership which needed to be worked out before it felt comfortable joining -I’d love to know what those were. Separately from Apple’s involvement, it’s worth noting that the board now has representatives from a number of other organizations beyond tech companies including several universities. So the Partnership won’t just be about driving the agenda of the tech industry here.
Cloud was the big theme on Microsoft’s earnings call once again, with a $14 billion annual run rate and nearly 50% gross margins across its cloud businesses, and a 95% growth rate in the Azure business alone. Surface revenue was down a bit, predictably because the product line wasn’t refreshed as fully as in previous years, but not by much, and it seems commercial sales actually grew (probably a reflection of the long sales cycles in enterprise). The phone business continues to dwindle to nothing (just over $200m in revenue this quarter by my estimate, down 81% year on year), but that’s so small now it barely impacts results. Windows did well overall, with some revenue growth from slightly stronger shipments in the PC market, though the PC market overall was still down overall last quarter. Monetizing its consumer business continues to be one of Microsoft’s biggest challenges – its Office consumer subscribers appear to be plateauing at around 25 million, most of its other consumer apps are offered free, and gaming is performing decently, though overall gaming revenue was down year on year. Overall, the results feed the narrative that Microsoft is undergoing a comeback, though it’s a slow and subtle one from a financial perspective.
You might also be interested in the Microsoft Q4 2016 deck which is part of the Jackdaw Research Quarterly Decks Service.