AI and ML at Microsoft, Google, and Apple
With machine learning (ML) and artificial intelligence (AI) being of particular relevance to this course as leading trends in IT, and also central themes in the annual developer conferences of 3 tech giants, I have chosen to compare & contrast strategies, products, and approaches in the field.
Microsoft at Build, Apple at WWDC, and Google at I/O all talked up their AI and ML capabilities. At the surface these all had similar buzz-words, but I use three figures of merit to compare and contrast: vision/strategy, product and roadmap highlights, and underlying infrastructure.
Vision/Strategy
Satya Nadella, CEO of Microsoft (MSFT), laid out the best vision. In the first 20 minutes of his Build 2017 Keynote he spoke about the unintended consequences of technology, and how he viewed AI as a means to empower people, not replace their jobs. Almost all company CEOs will have a slide with the company vision at the keynote, but it was unique and heartening to see Satya tie it back to how he wanted companies like MSFT to take accountability for their algorithms and build trust in future technologies. MSFT views the future as an interconnected intelligent cloud and intelligent edge, with your “conventional” devices utilizing the cloud while Internet of Things (IoT) devices like wearables form the edge (very similar to the concept of Fog Computing). Being MSFT, they envision this for both the consumer and the enterprise (apps in the cloud and factory sensors forming the edge).
Sundar Pichai, CEO, reiterated that Google (GOOG) was now an “AI first” company. While a grand vision was lacking, he described what I found to be the most comprehensive strategy in terms of how GOOG intends to take these lofty buzz-words and actually apply them to product. Several GOOG products heavily use AI, including Smart Reply, Street View, search in Photos, YouTube suggestions, and GOOG is just getting started. Their strategy is to infuse AI in every major product/platform, from search and maps to Chrome to Android. GOOG has invested heavily in ML frameworks like TensorFlow and infrastructure like TPU, and is probably ahead of the curve here.
Apple (AAPL), as it often does, chose a completely different approach. Tim Cook didn’t spend time talking about AI and its broader impact, neither did he list all the products that Apple was going to infuse with AI. Rather, there were subtle references to AI. Siri is the face of AI at Apple, and I expect more integration with Apple’s 4 major platforms: WatchOS, MacOS, iOS, and tvOS. For example, WatchOS now supports a new Siri watch face that tells you relevant information based on time of day (eg. lunch meetings) or activity (eg. laps during swim). ARKit, a Software Developer Kit (SDK) and framework for augmented reality experiences, and the HomePod smart speaker were some of the other AI-related announcements.
AI/ML Product and roadmap highlights
There is a lot that can be said here, since all 3 companies have dozens of innovative products, so I keep this section to a very high-level.
MSFT has identified 3 key areas of change in the AI/ML world: Windows, Office (O365), and Azure cloud. From a bird’s eye view, MSFT’s two big advantages are that a lot of consumer devices run Windows, and a lot of enterprises use Office products and the Azure cloud. This means that most of their products focus on pushing AI/ML to existing devices and software stacks: Microsoft Teams boasts integration with Cortana, new ML algorithms allow real-time monitoring and people & object detection, etc.
GOOG’s product roadmap is quite elegant: continue to make existing products smarter through the massive amounts of user data it already has. Also worth highlighting is the Google Assistant. Google has made a big push in the home with smart TVs and the Google Home, and has realized that a good way to unify the Google experience is through a smart assistant. AI has enabled speech to become a useable input option, and one that I find will be put to good use in the home. Expect to see more smart devices from kitchen appliances to strollers powered by the Google Assistant.
While Google Assistant has access to the vast amounts of data GOOG collects and puts to use in the Google Home, Siri seems to be falling behind (particularly because Amazon Alexa and Google Assistant are now available as apps on the iPhone). Yet AAPL has taken a different approach with the HomePod smart speaker. AAPL played down the AI/ML “smartness” of the product, instead focusing on how it was a much better speaker than competitors’ products, but also had Siri. Given the hefty price tag, I am interested to see how this product performs and how ML/AI capability evolves.
Underlying infrastructure
The underlying datacenter infrastructure that each tech giant is investing in is most interesting to me as a Hardware Engineer. Machine Learning requires a lot of data and a lot of compute, and the need for compute has sent NVIDIA’s stock soaring with no end for “GPU-hunger” in sight. GOOG and MSFT, however, have taken opposite approaches in building out their compute.
MSFT Azure has taken the reconfigurable route with FPGAs. On Day 2 of Build, the CTO of Azure spoke about Project Catapult v2 where MSFT put Altera Stratix V FPGAs in “almost all servers in our datacenters worldwide”[1]. This is likely over a million servers[2] that have redesigned to form a hypermesh of FPGAs and CPUs. The FPGAs are used for deep learning among other things.
GOOG has chosen custom, hardened, silicon: the first mention of TPU came last year, and a paper published this year mentions a 15-month design to deployment cycle (fast for hardware). TPU2 claims 1.5x the performance of NVIDIA’s latest Volta GPU, and I have no doubt that a TPU3 is already in production.
AAPL is also investing heavily in custom hardware, but mainly chips for its mobile and consumer products than datacenters. Imagination Technologies, a British company that used to supply graphics chips for iOS devices lost 70%[3] of its value after APPL announced it would move design in-house.
Note that all opinions are my own, and I have no first-hand knowledge of the inner workings of these 3 companies, so I look forward to your thoughts and comments.
[1] Microsoft: https://channel9.msdn.com/Events/Build/2017/B8063
[2] https://www.nextplatform.com/2016/09/26/rare-tour-microsofts-hyperscale-datacenters/
[3] https://9to5mac.com/2017/06/22/imagination-technologies-sale-apple/
Users who have LIKED this post:
10 comments on “AI and ML at Microsoft, Google, and Apple”
Comments are closed.
Well covered points on AI hardware!!!!
I agree with you that Siri is lacking behind in this game but i believe only Google has the most access to data for running effective AI function in todays environment, eg Amazon Echo works wonderful when it comes to the shopping experience or using it in the American markets but fails to function with that effectiveness in other markets due to lack of local data… while Google Home performance is more or less the same in all markets as the data points collected via various Google products across the globe are similar.
Innovation in machine learning is just starting at Google and they are making it truly open to the developer community like with the launch of there new “Video Intelligence API” which allows the users to search information in videos. we should have some new disruptive apps in the video editing and research space soon.
Users who have LIKED this comment:
Hi Gaurav,
You did a good job covering all the relevant points from the keynotes and providing your own interpretation as well. From a different perspective, I believe Apple has a major advantage in the coming future when you compare it to the other two companies. As technology expands and in the process intrudes more into people’s lives, the fear of the amount of data being collected is only likely to increase. But Apple does not rely on data to make profits. Tim Cook calls “privacy a fundamental human right”. The speakers launched by Apple retain speech data only for a few months; Amazon Alexa keeps it forever. Although, I can’t really comment whether this approach will be sustainable but I do believe that as more advances are made in the field of AI, consumer demand for transparency about how their data is being used and where will increase. Apple, then has a chance to succeed.
Users who have LIKED this comment:
I was very intrigued by your post and your thoughts about what Microsoft, Google, and Apple are all doing within the AI and ML space. I believe that Amazon and Facebook are also major players and are revolutionizing how organizations will move into the future. With said, Google, Microsoft and Amazon AWS have the most information and are the only major players in the Cloud.
http://www.zdnet.com/article/aws-microsoft-seen-rated-top-dogs-in-iaas-in-gartners-magic-quadrant/
I was reading an article on ZDnet about Google, where they discussed the “arms race” among the giants. Google has been using Machine Learning for over a decade. For instance, “the Google app uses speech recognition and natural language processing to understand speech in 55 languages; Google Search uses RankBrain, a ranking signal that uses deep learning to improve results; and Google Photos taps into the company’s latest image recognition system.”
http://www.zdnet.com/article/should-google-be-your-ai-and-machine-learning-platform/
Based on my research, it seems to me that Google probably has a greatest investment in AI and ML, but I am not a huge fan of their Google Cloud Platform. As a Network and Solutions Architect, Microsoft and Amazon have the best cloud offerings with Azure and AWS respectively. Google does have great technology, and a great deal of information, but the UX is lacking in opinion.
One offering that does interest me is the Google Machine Learning service that facilitates the creation of a neural network and creates algorithm models to run predictions at scale without worrying about the infrastructure. The service utilizes several of Google Cloud’s data analytic tools such as BigQuery, DataFlow, and Datalab.
https://cloud.google.com/products/machine-learning/
From a hardware perspective, it’s difficult to say for certain what is going to happen within the industry, and how long it will take to get there. During Cisco Live this past week, several Technical and Marketing Engineers seem to think that companies will eventually move away from proprietary routers, switches, firewalls and move all networking software and compute to a “White Box” model. (e.g. run code on commoditized networking and compute hardware.) Both Facebook and Google have been doing this for several years already.
Facebook Infrastructure
“As Facebook’s infrastructure has scaled, we’ve frequently run up against the limits of traditional networking technologies, which tend to be too closed, too monolithic, and too iterative for the scale at which we operate and the pace at which we move. Over the last few years we’ve been building our own network, breaking down traditional network components and rebuilding them into modular dis-aggregated systems that provide us with the flexibility, efficiency, and scale we need.”
https://code.facebook.com/posts/717010588413497/introducing-6-pack-the-first-open-hardware-modular-switch/
“At Facebook, we believe that hardware and software should be decoupled and evolve independently. We create hardware that can run multiple software options, and we develop software that supports a variety of hardware devices — enabling us to build more efficient, flexible, and scalable infrastructure.”
https://code.facebook.com/posts/1241394199239439/the-growing-ecosystem-around-open-networking-hardware/
With that said, you’re right… AI and ML capabilities require a great deal of compute. Most companies seem to be getting away from proprietary ASIC’s and moving toward commoditized FPGA’s due to increased flexibility, reduced cost and the need to improve “go to market” strategies among their competition.
https://www.altera.com/content/dam/altera-www/global/en…/wp-aab090205.pdf
One thing I find particularly interesting about these companies is their efforts to empower other people to use AI.
For instance, Microsoft’s Azure platform has its own Machine Learning Studio (https://azure.microsoft.com/en-us/services/machine-learning/) that attempts to allow its consumers to utilize machine learning algorithms for their own applications, even if they do not have any traditional background in AI/ML.
Similarly, Google has built tools that allow others to build their own AI applications. An example of this would be wit.ai (https://wit.ai/), which allows developers to build applications that users can talk or text to.
In that sense, these companies are not just working to incorporate AI into their products, but also empowering the community to use the latest technology in AI.
Users who have LIKED this comment:
In response to Aaron’s observations, it is true that companies are trying hard to empower people to use new technologies like AI. This is very interesting to me because it is clear that the majority of the time these companies are the ones who stand to profit from widespread use. Amazon will give college courses free credits so their students can use their AWS services, and companies like Nvidia will give grants and awards to startups in the AI space who use their hardware. This not only helps to create a market for the hardware (“picks and shovels” approach) but the resulting familiarity with specific software or the ease of integration with other products (apple …) reminds me of Gillette who makes money on new razors once they hook you with their handle or companies who lose money on consoles but make it back later on the games.
On the one hand, smoke and mirrors can leave those who buy in stuck with a poor product while someone gets rich. But if the technology is worth pursuing these efforts can be fantastic for both the supplier and the consumer.Sure companies are after the bottom line but that doesn’t mean what they are doing is nefarious. If you truly believe that AI (or whatever your hardware/software enables) can change the world for the better and empower people , and that your product is the best way to do it, then for profit companies can have their cake and it too.
Hi Gaurav,
I am also working as a hardware engineer, so the HW acceleration tools, especially for machine learning applications, are very interesting to me as well. In fact, I like that you mentioned the FPGA usage in Microsoft. In fact, back in the summer of 2014 when I was still pursing my BSEE degree, I saw an article about Microsoft incorporating FPGAs in their Bing servers to accelerate the intensive search workload. https://www.enterprisetech.com/2014/09/03/microsoft-using-fpgas-speed-bing-search/
Seeing this prompted me to focus even more in digital design and use of FPGAs/ASICs for acceleration, so it is great to continue learning about how the industry is evolving to incorporate and produce hardware accelerators. In fact, it is also very interesting to hear the trade offs between accelerator choices. Obviously an ASIC would be more efficient for speed and power (which is I’m sure why Google decided to create the TPU), but the cost and time to produce would be much greater than simply using FPGAs or GPUs. But even choosing between FPGAs and GPUs is a challenge in itself, so it is interesting to see which choice is used for each application and company.
Hi Gaurav,
It’s a great comparison and has been well jotted. Kudos on that.
Microsoft- Is doing some amazing work in the field of AI/ML and with advancement in OS, Office & Cloud services. They’ve taken the game forward with digital assistant and Chatbots.
It is a company that has always focused on or relied on professionals across the globe, hence it plays a major role in the development of other companies too.
It has improved on it’s development environment (tools to develop) and encouraged more and more developers to join the community, leading to a better growth.
Google-The company with it’s I/O this year made it very clear that it is going to incorporate AI/ML with all existing products.
The new TPU and GPU are path breaking as it helps developers to develop with great ease.
🍎- I feel they are in the game just for lucrative purposes and not for the overall development of the community. With the event this year they said they’re going to incorporate AI/ML with their respective OS.
Everything said, I feel all the 3 companies are exclusive and are doing some great work in their fields – which is why they’re able to hang on to their No.1 spot.
Source: Shail Shah – shailvshah@gmail.com
Great summary of the latest developments in AI/ML from Microsoft, Google and Apple and insights from their developer conferences – kiitos!
A few thoughts on how this enabling technology will shift consumer behaviour:
Voice will become the new touch screen and kill brands eventually. Apple will use voice for media, Google will use voice for search & information and Amazon will use voice for commerce. (Scott Galloway, NYU 2017) https://www.youtube.com/watch?v=3MOwRTTq1bY. Many of the tech giants are using AI/ML to automate and augment their current revenue streams, fortify their IT infrastructure to better serve their current business model and build an active developer community around it with easy to use APIs.
The rate of growth for Alexa voice apps has been growing and has reached over 15,000 already (Techcrunch, 2017) https://techcrunch.com/2017/07/03/amazons-alexa-passes-15000-skills-up-from-10000-in-february/?ncid=rss. This development is leading to greater convenience for customers, lesser friction in the shopping experience and reducing the amount of channels to only the home device or intelligent personal assistant. Amazon is hedging it’s bets by pricing products lower when bought via Alexa and it will be interesting to see how fast a bigger portion of the consumer markets will adopt this new behaviour.
Gaurav, you did a great job of summarizing the aspects of AI/ML – Vision, Roadmap and Infrastructure offerings from MSFT, GOOG, AAPL. Overall, I agree with your comments – companies are taking an “AI first” approach by either explicitly offering AI – as a service (NPL API, Vision API) or implicitly incorporating AI/ML into their products (Smart replies in Gmail, Google Photos).
The reality is that, companies will become obsolete if they do not adopt AI/ML as their competitive advantage. This is true for Auto industry (with Self driving cars) to Retail (ML based recommendations). Customer service departments are replacing humans with chatbots and Social media is becoming a data firehose for Marketing using ML based customer targeting algorithms.
The cloud vendors like – Amazon (AWS), MSFT (Azure), GOOG (GCP), IBM (Watson) are making it easier to build applications with a minimum barrier to entry (in terms of data science skills) by offering easy to build API’s and a low cost entry with pay-as-you-go models which have resulted in companies that have started or migrated to the Cloud. I believe that this allows startups and large companies to bring new ideas to market by rapid prototype and iterate on new ideas for a low cost and scaling easily for mass adoption. I can’t wait to see all the innovation that will emerge in the coming years!
Excellently organized article! It really gave me a direct overview of AI/ML used by the big three tech giants. It’s certainly interesting to see how the companies differ in their approach in employing those technologies in their products and services.
It seems as if Apple is falling slightly behind its competitors in the AI game, however I believe it’s not necessarily true. Apple has always put a greater emphasis on consumer electronics rather than pure technologies, and chose to “humanize” technical features into products. Such is the case in their operating systems, where Apple uses machine learning to observe the user’s usage behavior and carries out certain operations such as app suggestion based on location, thus making the experience more personal and convenient. Also, it is widely rumored that Apple is working on a car-related project, we don’t know whether it’s an actual car or simply an autonomous driving system, but we can be certain that there is a high level of AI technology involved. Along with other examples such as smart home, they show us that Apple sees AI and ML as the future and is committed to bring those technologies to the average consumer once development is mature and complete.