AI, new battlefield of Office 365 and G Suite

While Microsoft focuses on document creation support, Google is focusing on Gmail with help writing messages.

Historically, Office 365 has been the first productivity suite to integrate machine learning. This dimension was introduced with Delve in 2015. Objective of this brick? Offer the user a personalized and hierarchical view of his content (files, emails, conversations …) according to his relationship and documentary graph: his hierarchical history, collaboration, the mesh of his interests … Delve remains very little deployed to date. Feedback on the application is rare or non-existent. Less and less put forward, Delve is still maintained by Microsoft. It is also marketed in the French data centers of the American group. Four years later, Google is catching up. In mid-2018, the Mountain View company caused a sensation by equipping G Suite with a series of Gmail-based AI functions.

Office 365 G Suite
Help content creation x
Bots and team messaging x
Knowledge Management x
Smart messaging x
Unified and personalized document search x
Grammatical and semantic suggestions x x

At the heart of these new advances, Gmail has been given a possibility of smart reply offering predefined answers based on messages received. To an e-mail requesting a call “Wednesday at 11am or Thursday at 17h”, it will for example make several suggestions for replies: “Wednesday is perfect”, “Thursday suits me”, “The two of me go” or ” Neither of the two proposals suits me “. Simple and efficient.

“Google manages to decode fairly complex mail,” says Arnaud Rayrole, CEO of the French consulting firm Lecko, expert in collaborative solutions. “To a message asking to establish a new quote alongside other peripheral information, the smart reply will be able to put forward consistent suggestions: ‘Here is the modified quote’ or ‘excellent news, here is our new proposal'”. Thomas Poinsot, digital consultant at the French service company Spectrum Group, said: “The smart reply is a real plus, especially in a situation of mobility when there is little time to answer.” Google is already planning to extend it to its Hangouts IM.

Gmail: the AI ​​at the messaging service

Another lever of Gmail based on AI: the smart dials. For the time available only in English, this device “auto-complete” a message being input. By analyzing the typed terms, he finds the following most probable words given the context, and thus increases the speed of seizures. “This AI will expand to other languages ​​and get rich over time by identifying how you are addressing this or that person,” says the Mountain View giant (read the official post ).

Signed Google animation showing Gmail’s built-in email help feature. © Google

On the smart e-mail front, Office 365 is currently for absent subscribers. In terms of artificial intelligence, Office 365 R & D focuses on file creation support. Among the first affected Office bricks: PowerPoint. Based on the analysis of the presentations being created, the application recommends related content (images or texts) stored locally or on the web, which can be used for writing. It also uses image recognition to provide complementary photos (read the MSFT article ). 

Smart content in PowerPoint

Of course, G Suite is also developing this approach (read  the post ). But Microsoft pushes it further. In Excel, for example, the editor now includes a smart wizard called Ideas that identifies trends, patterns, or outliers in a table.  The automatic translation is also supported in Excel, PowerPoint and Word , with 60 languages ​​covered, but also in Stream for the automatic captioning of videos. In parallel, Microsoft continues to optimize the machine learning layer of SharePoint . On the program: decryption of images for character recognition and metadata extraction, or sentiment analysis and facial recognitionusing Azure Analytics Services. The same logic for Power BI, the data visualization application can use Azure Machine Learning to refine its processing, including identifying entities (organizations, people, locations). Via Azure ML, it even offers the possibility to build its own model of machine learning (read the post ).

Animation illustrating the capabilities of the intelligent assistant, Ideas, integrated into the right column of Excel. © Microsoft

Another battleground where AI comes into play: the search for content. Unsurprisingly, this is an area in which Google is showing a good head start. Called Google Search , ” the search engineGoogle Drive is very powerful, “says Arnaud Rayrole at Lecko.” It suggests results based on your favorite themes, the frequency of consultation and editing of a particular document. It gives access to a ranking of the results with regard to the authors with whom you exchange the most. “Only downside: Google Search does not go as far as managing skills.” It does not analyze the profiles of employees, their career and their a work history to unearth the knowledge available internally that can be useful for this or that project, “continues Arnaud Rayrole, an approach that Microsoft integrates via Delve or Yammer, the Office 365 enterprise social networking brick .

Google’s strategy more readable

In terms of documentary research, Microsoft still has a way to go before catching up with Google. The group has announced its intention to evolve in this field towards a unified experience. For now, several indexing enginescoexist within Office 365, each associated with an application (Yammer, Teams, Outlook, Drive ….) The publisher intends to consolidate them within a single base called Microsoft Search. R & D work that should be completed by the end of the first half of 2019. The promise? Provide an AI-based search environment that delivers personalized and consistent results across the entire platform. Eventually, the company of Satya Nadella even intends to tend to a common search interface covering not only Office 365 but also Windows 10 and the software package Dynamics 365.

For the consultants surveyed, these Office developments are generally in the right direction, but are less readable against the simple and pragmatic AI levers that draws Google in Gmail.

Last opposition ground: chatbots. Google and Microsoft both integrate this dimension into their respective suite through team messaging application: Hangouts Chat for the first and Teams for the second. Launched in early 2017, just over a year before Hangouts Chat, “Teams has significantly more conversational agents,” says Thomas Poinsot at Spectrum Group. “But like at Google, they are mostly limited to simple tasks, click-to-action.Rare are those equipped with a layer of conversational AI capable of providing the expected response by seeking the good applications. “

Algorithms that shape the business

“In the end, all these efforts have a common purpose: to help users manage ever-increasing volumes of data.” The goal is laudable, but AI is never neutral. no less than a vision of the world coded in algorithms, which ultimately gives Google and Microsoft the opportunity to guide the choices of an organization by prioritizing information according to their own rules, “warns Arnaud Rayrole. “Likewise, they encourage the evaluation of employee performance by following their logic.” On this point, the CEO of Lecko evokes the Office 365 MyAnalytics brick. “It provides KPIs inspired by American culture, for example the rate of users transmitting emails at non-working hours and not by time slots.

How Facebook put the AI ​​at the heart of its social network

Detection of inappropriate content, ranking newsfeed, facial recognition … The platform uses massive machine and deep learning.

Artificial intelligence (AI) is present on all levels of the social network Facebook . At the heart of newsfeed, it prioritizes content based on users’ interests, their consultation history, their social and relational graph. Similarly, it allows them to push advertisements to which they have a high probability of joining, from the click to the purchase. More difficult, the AI ​​is also exploited by the platform to detect unauthorized connections or false accounts. Finally, it intervenes to orchestrate other tasks, less visible, but remain key in the daily social network: customize the ranking of search results, identify friends in photos ( by facial recognition) to recommend tagging them, manage speech recognition and text translation to automate captioning videos in Facebook Live …

Target function Algorithm family Type of model
Facial Recognition, Content Labeling Deep learning Network of convolutional neurons
Detection of inappropriate content, unauthorized access, classification Machine learning Gradient Boosting of Decision Trees
Customization of newsfeed, search results, advertisements Deep learning Multilayer Perceptron
Natural language comprehension, translation, speech recognition Deep learning Network of recurrent neurons
Matching users Machine learning Support vector machine
Source: Facebook search publication

The social network makes extensive use of standard machine learning techniques. Statistical algorithms (classification, regression, etc.) adapted to create predictive models starting from encrypted data, for example in order to predict changes in activity. Facebook uses it to find inappropriate messages, comments and photos. “Artificial intelligence is an essential component for protecting users and filtering information on Facebook,” insists Yann Lecun, vice president and scientific director of the group’s AI. “A series of techniques are used to detect hateful, violent, pornographic or propaganda content in images or texts, and, on the contrary, to label the elements likely to

Multiple neural networks

Alongside traditional machine learning, deep learning is obviously implemented by Facebook. Based on the concept of artificial neural network, this technique applies to both numbers and audio or graphic content. The network is divided into layers, each responsible for interpreting the results from the previous layer. The IA thus refines by successive iterations. In text analysis, for example, the first layer will deal with the detection of letters, the second of words, the third of noun or verbal groups, and so on.

Unsurprisingly, Facebook applies deep learning to facial recognition, particularly via convolutional neural networks, a particularly efficient algorithm for image processing. Via the multilayer perceptron method, ideal for managing a ranking, the social network uses it to customize the newsfeed or the results of its search engine. Finally, deep learning is also powered by Facebook for machine translation.

To recognize embedded text in images, Facebook uses a triple layer of neural networks. © Capture / JDN

Around image processing in particular, Facebook has built a deep learning platform called Rosetta . It is based on a triple network of neurons (see screenshot above). The first, of a convolutive nature, maps the photos. The second detects the zone or zones containing characters. As for the latter, he tries to recognize the words, expressions or sentences present within the identified regions. A technique known as Faster R-CNN that Facebook has slightly adapted to his needs. Objective: better qualify the images posted to optimize indexing , whether in the newsfeed or the search engine .

A deployment pipeline

To set its different AIs to music, the American group has a homemade pipeline. Code name: FBLearner (for Facebook Learner). It starts with an app store. For internal teams of data scientists, it federates a catalog of reusable functionalities as well for the training phases as for the deployment of the algorithms. A workflow brick is then added to manage the training of the models and evaluate their results. Training processes can be performed on computational clusters (CPUs) as wellonly clusters of graphics accelerators (GPUs), again designed internally. Last stone of the building, an environment motorized the predictive matrices, once these trained, in situ at the heart of the applications of Facebook.

Workflow built by Facebook to develop and deploy its machine models and deep learning. © Capture / JDN

Side libraries of deep learning, Facebook has historically chosen to develop two. Each is now available in open source. Designed for its basic research needs, the first, PytTorch, is characterized by its great flexibility, its advanced debugging capabilities and especially its dynamic neural network architecture. “Its structure is not determined and fixed, it evolves as learning and training examples are presented,” says Yann Lecun. The downside: the Python execution engine under the hood makes PytTorch inefficient on production applications. Conversely, the second deep learning framework designed by Facebook, Caffe2, was precisely designed to be the product deployments. In this context,

More recently, Facebook has set up a tool called ONNX to semi-automate the formatting for Caffe2 of models originally created in PyTorch. “The next step will be to merge PyTorch and Caffe2 into a single framework that will be called PyTorch 1.0,” says Yann Lecun. “The goal is to benefit from the best of both worlds, and result in a flexible infrastructure for research and efficient compiling techniques to produce usable AI applications.”

Towards the design of a processor cut for the AI

Following the model of Google’s Tensorflow Processor Units (TPU) cut for its Tensorflow deep learning framework, Facebook is also planning to develop its own optimized chips for deep learning. “Our users upload 2 billion photos a day to Facebook, and within 2 seconds each is handled by four AI systems: one manages the filtering of inappropriate content, the second the labeling of images in order their integration with the newsfeed and the search engine, the third the facial recognition, finally the last generates a description of the images for the blind, all consumes gigantic computing and electrical resources.With the real-time translation of video that we are now looking to develop, the problem will intensify.