Wisam Hirzalla, director of product marketing for Microsoft Azure, at Microsoft’s Redmond, Washington campus. Photo by Dan DeLong for Microsoft.
Just a few years ago, artificial intelligence was largely relegated to universities and research labs, a charming computer science concept with little use in mainstream business. Today, AI is being integrated into everything from your refrigerator to your favorite workout app.
“It’s really exciting, because there’s a new breakthrough every month, or every week,” said Lance Olson, director of program management for applied AI at Microsoft. “Increasingly, the conversations are switching from discussing the art of the possible to getting to the next level of implementation on a specific project.”
Still, many companies are struggling to achieve their AI goals, as the supply of data scientists and AI experts has failed to keep up with surging demand. Creating AI models is difficult work. And then comes a struggle to get them into production – and keep them running. Data ages, much more quickly than code, making models less accurate as the world changes around us.
At its 2019 Microsoft Build conference, the company says it’s focused on helping all developers – even those without an AI or data science background – use its tools and services to deliver the big benefits that more and more customers expect.
“AI and machine learning can turn developers into heroes, for their ability to deliver really personalized, super-immersive experiences to customers,” said Wisam Hirzalla, director of operational databases and Blockchain product marketing at Microsoft. “We want to make it easy for any company to use the technology.”
Simplified and automated machine learning
Toward that end, Microsoft is announcing new capabilities for its cloud-based Azure Machine Learning service, with a goal of enabling developers and data professionals of any skill level to build advanced machine learning models.
We can think of AI practitioners in three categories, according to Bharat Sandhu, director of artificial intelligence at Microsoft. First, we have developers and data scientists who like to write code. They want to build machine learning models using tools and processes they already know. For them, Azure Machine Learning offers a “code first model,” where they can use the development tools they like.
A second group, including business domain experts, may know a lot about data, but they don’t know much about machine learning or code. For those customers, Azure Machine Learning’s automated machine learning experience is a “no code” option, accessible without having to write any code.
“A third category of people, who are learning machine learning concepts, they want to make their own models, but they are not coders. This could be IT professionals, or folks with background in statistics or mathematics,” Sandhu said. “For those customers, we’re offering a drag-and-drop experience to make models visually.”
Sandhu noted that no matter which way the machine learning models are created, they all use the same back end, meaning all the models can easily be integrated together.
Of course, developers and data scientists have a number of platforms to choose from when they build AI models. To make sure companies can adopt AI advances as quickly as possible, Microsoft says it’s important to overcome platform mismatches, which can delay the rollout of those models into production.
One way Microsoft promotes interoperability among the various AI frameworks is a standard called ONNX Runtime, or Open Neural Network Exchange. This joint effort with other tech companies creates deployment models that work across multiple platforms.
That frees up developers and data scientists to use whatever framework and hardware target is best for them. And it frees up the operational team to focus on deploying and getting results, instead of having to translate as they move from one to the other.
At Build, Microsoft is announcing support for ONNX integration with leading hardware accelerators.
The company also is announcing that it is now an active contributor to the MLflow project, an open source platform for managing the machine learning lifecycle.
Azure Cognitive Services updates
More than 1.3 million developers, many without specific AI or data science skills, currently use Azure Cognitive Services to build intelligent apps that can see, hear, speak, understand and even begin to reason.
At Build, Microsoft is announcing a new category of Azure Cognitive Services called Decision, which gives specific recommendations to help people make decisions. This new category includes Personalizer, which uses a branch of AI called reinforcement learning to help technology glean knowledge from its own experiences and then offer informed recommendations.
“We are able to take reinforcement learning and ship it in a way that’s accessible to developers and doesn’t require a data scientist,” Olson said. “That will be very impactful for customers.”
At Build, the company is announcing many other updates to Azure Cognitive Services, including Ink Recognizer, which can learn to read handwriting, Form Recognizer, which identifies forms, and other new conversation transcription capabilities and other speech, vision and language advances.
Just getting started
To date, Microsoft’s customers have created almost 400,000 digital agents through its Azure bot service, and more than 3,000 come on line each week. Companies of all sizes are looking to AI to give them a competitive edge.
That includes Cheetah Mobile, a leading mobile app maker building AI-enhanced hardware, including the hand-held CM Translator. Rather than developing the entire speech system from scratch, the company used Azure Cognitive Services, leveraging its text-to-speech API to provide rapid, high quality translations.
The development cost savings helped keep the device affordable, with no compromise in the natural speech flow.
Other companies say one of the chief benefits of using Azure data and AI tools is that they can take advantage of other attributes built into the tools. For example, the digital asset management company MediaValet relies on the security and privacy safeguards Azure provides to reassure customers that the images it processes will be handled properly.
“We’re not a big company, but we can actually play ball with big enterprise players, because we can leverage the information security and privacy attributes, the trust-ability of Azure,” said MediaValet chief technology officer Jean Lozano.
In the coming months and years, Microsoft expects more and more customers to start using AI, both because they see the business benefits and because the tools are more accessible.
“AI opens up so many possibilities. And the limits are very few, generally limited only by your imagination,” Olson said. “It doesn’t need to be overwhelming for people. We are getting to the point where we can now make AI accessible to a much broader set of customers.”