Microsoft AI news: Making AI easier, simpler, more responsible

Today is a big day for AI announcements from Microsoft, both from this week’s Build conference and beyond. 

But one common theme bubbles over consistently: For AI to become more useful for business applications, it needs to be easier, simpler, more explainable, more accessible and, most of all, responsible. 

Microsoft works to makes responsible AI easier to understand

Responsible AI is actually at the heart of a lot of today’s Build news, John Montgomery, corporate vice president of Azure AI, told VentureBeat. Most notable is Azure Machine Learning’s preview of a responsible AI dashboard, which brings together capabilities in use over the past 18 months, such as data explorer, model interpretability, error analysis, counterfactual and causal inference analysis, into a single view. 

Azure Machine Learning will also now offer a responsible AI “scorecard” in preview to summarize model performance and insights so that all stakeholders can easily participate in compliance reviews. 

“The question was how to make AI work more accessibly and easier to understand,” said Montgomery. “Users need to create something compliance officers can leverage.” The responsible AI dashboard and scorecard provides a “very straightforward set of visualizations, a wizard-driven flow that makes it easier for people with less data science experience to use these tools and export the scorecard into a PDF.” 

UK’s National Health Service uses Azure’s responsible AI dashboard

For example, a team of medical professionals at one of the largest National Health Service trusts in the UK is exploring how AI could help reduce waiting times, support recommendations from healthcare teams and provide patients with better information so they can make more informed decisions about their own care. Two orthopedic surgeons developed an AI model to help consultants give their patients a personalized risk assessment of upcoming hip or knee operations. 

The model is hosted in Microsoft’s Azure cloud and uses the responsible AI dashboard in Azure Machine Learning. Medical professionals can see how the model works and get a clearer understanding of why the AI has reached those conclusions. This boosts confidence that their advice to patients is based on accurate and reliable data.

“Machine learning models are kind of a dark art, so a data scientist can explain to another data scientists why they behave a certain way, but you need to be able to explain to business owners, CEOs, auditors and compliance teams – that’s where the responsible AI tooling comes in, to help explain why models are making the decisions they’re making,” said Montgomery. “That’s the direction we’re heading in.” 

Making developers’ lives easier with AI 

At Microsoft Build, Microsoft’s chief technology officer, Kevin Scott, also highlighted how AI-powered software development tools are allowing developers to build software using natural, everyday language, by translating natural language into the programming languages that computers understand. 

“That’s a fundamentally different way of thinking about development than we’ve had since the beginning of software,” Scott said. Microsoft highlighted this paradigm shift as driven by Codex, a machine learning model from AI research and development company OpenAI that can translate natural language commands into code in more than a dozen programming languages. 

Codex descended from GPT-3, OpenAI’s natural language model that was trained on petabytes of language data from the internet. According to Scott, the increase in productivity that Codex brings to software development is a game changer, allowing developers to “accomplish many tasks in two minutes that previously took two hours.”  

AI-powered low code and no code tools bring AI to the masses

At Build, Microsoft also highlighted AI-powered low code and no code tools such as those available through Microsoft Power Platform, which they said: “are poised to enable billions of people to develop the software applications that they need to solve their unique problems, from an audiologist digitizing simple paper forms to transform hearing loss prevention in Australia to a tool that relieves the burden of manual data-entry work from employees of a family-owned business and an enterprise-grade solution that processes billions of dollars of COVID-19 loan forgiveness claims for small businesses.” 

According to Charles Lamanna, Microsoft corporate vice president of business applications and platform, if you work with formulas in Excel, you can build AI-powered software applications. 

“The hundreds of millions of people who are comfortable working with formulas in Microsoft Excel could easily bring these skills into Power Platform, where they can build these types of software applications,” he said.  

More mainstream moves from Hugging Face with Microsoft partnership

In other Microsoft news, today Hugging Face — which recently announced a $2 billion valuation thanks to its rising status as the main repository for all things related to machine learning models — continued its march to the mainstream with an announcement of a new collaboration with Microsoft that makes the Hugging Face machine learning platform accessible to Microsoft Azure customers.

A new service, Hugging Face Endpoints on Azure, makes Hugging Face models, according to the company, “simple,” “scalable” and “secure.” 

“We’re deepening our partnership with Microsoft and offering this new service which is a very easy way for people to turn their Hugging Face model into a scalable, secure production that lives within their own Azure tenancy,” Jeff Boudier, product director at Hugging Face, told VentureBeat. “We’re really providing the last mile of the machine learning development cycle.” 

Machine learning projects, he pointed out, especially in the enterprise, have low completion rates – 80% never make it into production. They also take months to complete. According to Boudier, with “a few clicks” and “10-20 minutes” developers can “have an endpoint running within your own Azure environment that’s compliant with everything within your company.” Hugging Face hopes that this can turn 20% completion rates into 80% completion rates, and three to four months of delays into a finished effort in less than half an hour. 

“Companies have been reinventing the wheel and scratching their heads on the same sort of pain points about what kind of instance can I fit this model in? How many replicas? How do I scale this?” said Boudier. “We’re making that so simple that the machine learning engineer can go all the way to production.”