Author(s): Nicholas McQuire
Winning developers and the trust of enterprises are two of the biggest arenas of competition between major cloud providers, and for this reason, Microsoft Build has been a determining event for the company for the past six years.
Build 2019 was no different, as a steady stream of news focussed on many of the key technologies (and buzzwords) that have become vital to developers: cloud, open source, serverless and edge computing, blockchain and mixed reality, to name a few.
But at the heart of everything at the event was artificial intelligence (AI), which has become fundamental to Microsoft's positioning with developers, in its post-PC and Windows-centric era. AI played a central role in the future scenarios for interaction between humans and machines showcased by CEO Satya Nadella during his keynote address, and in most of the event's big announcements.
Let's take a closer look at the highlights and what they mean for Microsoft's strategy and the market.
The Keynote: AI Underlies Microsoft's Developer and Cloud Strategy
Mr Nadella used his keynote slot to frame Microsoft's strategy, which, unlike previous years, is being delivered with unprecedented clarity and precision. The strategy's four main platforms — Azure, Microsoft 365, Dynamics 365 including Power Platform, and gaming, which wasn't covered at the event — provide ample opportunity for developers to build products and create an enviable flywheel effect across Microsoft's core cloud businesses. In my view, these pillars are a great way to look at the entirety of the company's cloud offerings.
An important enabler spanning this picture is AI, as shown in the image below.
Microsoft focusses on three main audiences for AI: developers, organizations and industries, and people. Its capabilities can also be broken down between three categories, based on customer maturity:
- Incubation, aimed at helping businesses to bring AI to every application through its Azure developer tools spanning infrastructure, platform and developer services.
- Transformation, which centres on bringing AI to every business process and concentrates on its packaged business solutions in Azure as well as Dynamics 365 and Power Platform.
- Productivity, which brings AI to every employee through the infused AI in Microsoft 365.
This clearly articulates the breadth of Microsoft's capabilities and sends an unmistakable message to developers that Azure is the best cloud for AI. Customers on stage during the keynote presentation such as Walgreens, Starbucks and BMW reinforced this view.
So too, for the first time, did students.
In a clever move ahead of Build, Microsoft announced that attendees could bring their children to the event. The timing coincided with the final leg of its annual student developer competition, the Imagine Cup, which Microsoft held just before the opening keynote session. An 18-year-old freshman from the University of California, Los Angeles took the prize for his work developing a smartphone-based blood glucose monitor, using Azure Machine Learning, to tackle the global diabetes epidemic. It was an inspiring showcase of Microsoft's technology and the keynote's "new and better world" message.
A New Vision for Conversational AI, Cortana and Intelligent Agents
The keynote presentation also saw Mr Nadella articulate Microsoft's renewed vision for conversational AI and intelligent agents. This area is becoming more prominent as Microsoft expands its collection of assets including its Bot Framework, the incorporation of its XOXCO and Semantic Machines acquisitions made in 2018, Cognitive Services, its Xiaoice social chat bot in Asia and, above all, Cortana. These assets have been growing in popularity with developers looking to build bots, which often represents their first experience in building AI.
Microsoft said that more than 1.2 million developers are using its AI tools, with 350,000 of them using its Bot Framework to create on average nearly 3,000 bots per week. A notable example is BMW, which is putting Microsoft's speech AI into the next generation of its cars.
Build showed how Microsoft, by bringing together Microsoft Research with the expertise of Semantic Machines, is quickly developing new conversational interfaces that support multiturn dialogue, multiple skill domains and multiple third-party agents to deliver richer, more context-aware interactions.
A highlight of the keynote speech was a video showing a more natural, functional and conversational Cortana being used as a workplace assistant, scheduling calendar appointments, sending driving directions to a car and booking tables at a restaurant. It led some people to question how real the technology was, but after a private demonstration of the technology at the event given by Dan Klein, co-founder of Semantic Machines and professor at the University of California, Berkeley, I can confirm the progress Microsoft is making here.
Microsoft has been feeling pressure to show progress with Cortana over the past year as it continues to lag behind rivals. In our most recent survey of employees, for example, Cortana landed in fourth place as the preferred assistant, behind Google Assistant, Apple's Siri and Amazon's Alexa.
Although speech assistants have a long way to go before they gain widespread acceptance within enterprises, Microsoft must continue to invest in Cortana. The platform is likely to become the next user interface and front end to the Microsoft 365 experience. But above all, Cortana must counter the early lead Google has with Google Assistant, which is now being integrated into G Suite, Microsoft's fiercest rival in cloud productivity.
Azure Cognitive Services Expand with New Features
A big investment area that came alive at Build was Microsoft's Azure Cognitive Services set of developer APIs. The company announced several enhancements to its portfolio, including a new Decision category that incorporates two existing services, Content Moderator and Anomaly Detector, as well as a new service called Personalizer, which uses reinforcement learning to deliver a relevant experience to each user. Additional new features include Ink Recognizer, which can read handwriting, Forms Recognizer, to help automate data entry by extracting data from forms, and real-time conversation transcription capabilities.
Microsoft also announced the general availability of Azure Cognitive Search, which integrates with Azure Cognitive Services to augment data as it's ingested, enhancing content understanding and creating a richer search index.
Cognitive Services are evolving into a premier collection of developer APIs. However, a criticism of cloud suppliers in this area has been that they lack domain relevance for businesses and few focus on improving business processes or solving more industry-specific problems. Microsoft should take advantage of its Azure AI accelerators and Azure AI Gallery to release solutions for businesses such as fraud detection, compliance monitoring, predictive maintenance and dynamic pricing. We expect more on this over the next 12 months.
Azure Machine Learning Pushes into MLOps
Microsoft also announced a host of important enhancements to its Azure Machine Learning platform. These included:
- A new, no-code visual interface for its automated machine learning service. This is geared at improving data scientist productivity by automating the testing of multiple models in parallel.
- High-speed inference from the cloud to edge, with the general availability of hardware-accelerated models that run on field-programmable gate arrays in Azure or — in preview — Data Box Edge. Microsoft also announced that Project Brainwave has entered preview and that it now offers ONNX Runtime support for Nvidia TensorRT and Intel nGraph.
- A machine teaching platform based on reinforcement learning. This enables domain experts in companies with limited AI experience to train computers and create autonomous systems for business and industrial scenarios such as smart buildings, industrial machinery and robotics. The technology is based on the Bonsai acquisition Microsoft made in 2018.
Arguably the most important announcement in this area, however, was the release of MLOps, a DevOps environment for machine learning. Integrated with Azure DevOps as well as GitHub, MLOps is an end-to-end life-cycle management solution for machine learning operations, with capabilities to improve processes for model creation, deployment and most importantly, performance in production environments. It also offers many advantages in terms of model reproduction, auditing, validation and observation.
In my view, MLOps' most important benefit is that it helps businesses improve the governance of machine learning through features such as tracking bias in data pipelines, explainability in model outcomes, monitoring data drift and assessing overall model performance and quality assurance. It does this through continuous monitoring of the feedback loop process that helps firms improve algorithmic quality and live computing operations. These areas have become critical as businesses look to move AI projects from experimentation to operationalization in their companies.
Key Insights from Build 2019
Altogether, Build's AI announcements reveal a few important insights. Firstly, Microsoft is clearly demonstrating some strength with enterprise developers, especially in the areas of speech and bots. There were also early signs of success from Microsoft's 2018 acquisitions in these fields, namely Semantic Machines and XOXCO. The company still has work to do to consolidate all its properties for a clearer go-to-market strategy, including incorporating the Xiaoice technology, now with 660 million users, into other areas. However, Microsoft is beginning to show market leadership in this area.
Secondly, with the expansion of its Azure Cognitive Services, the incorporation of "everyday AI" features in Microsoft 365 and the improvement of its inference and network-edge deployment capabilities, Microsoft continues to increase the breadth of scenarios for its AI. I believe this scope is why Microsoft was chosen by IT decision-makers as the top brand for advancing AI in the enterprise, ahead of Google, Amazon Web Services and IBM, in our recent survey.
Finally, the Azure AI platform, which includes Azure Machine Learning tools, Azure AI infrastructure and Azure Cognitive Services, continues to expand with the incorporation of machine teaching, MLOps and inferencing capabilities. The Azure AI platform now rivals Amazon SageMaker in terms of the breadth of its machine learning life-cycle capabilities for developers and data scientists.
Next Steps for Microsoft in 2019 and Beyond
The sheer number of announcements and future scenarios presented by Mr Nadella in his keynote highlight the progress Microsoft is making in a wealth of cloud assets and developer platforms. Few players have such a flywheel effect throughout their businesses producing the same level of growth and scale for developers.
As the role of the developer changes significantly over the next few years to work closer than ever with data scientists and business leaders, as well as with privacy, security and compliance departments, Microsoft's AI will be crucial to help them make the transition. The release of MLOps is an example of a good step in this direction.
But Microsoft will have to improve its AI strategy. It needs more production-level features, rather than research, to improve AI governance, including more instruments in Azure Machine Learning for bias detection, explainability and security. It also should look to build a more comprehensive marketplace strategy for machine learning models and data sets.
I believe customers will expect a trusted environment in Azure, rather than GitHub to transact intellectual property in AI. Microsoft should also prioritize the expansion of more packaged business solutions and the promotion of industry-led partners for specific business processes that can help organizations implement AI in their operations more quickly.
Successfully addressing each of these areas must now be Microsoft's priority as it presses forward in 2019.