Microsoft uses GPT-3 to add AI features to Power Apps

During its Build developer conference, Microsoft unveiled the first features in one of its products — Power Apps — powered by GPT-3, the natural language model developed by OpenAI. Microsoft says this will make it easier for users to build apps without needing to know how to write computer code or formulas, and the features are set to launch for North America customers in preview in English by the end of June.

Learning how to create complex data queries can involve a steep learning curve, particularly for data practitioners who don’t know to program. A study from Mendix found that 24% of customers have no previous experience using low-code platforms like Power Apps and that 40% come from a mostly business background. Still, even non-developers have to understand the logic of formulas like “FirstN(Sort(Search(‘BC Orders’, “stroller”, “aib_productname”), ‘Purchase Date’, Descending), 10).”

The idea behind the new GPT-3-powered features in Power Apps is to assist people in choosing the right formulas to get the result they need.

GPT-3 in Power Apps

Roughly a year ago, Microsoft announced it would invest $1 billion in San Francisco-based OpenAI to jointly develop new technologies for Microsoft’s Azure cloud platform and to “further extend” large-scale AI capabilities that “deliver on the promise” of artificial general intelligence. In exchange, OpenAI agreed to license some of its intellectual property to Microsoft, which the company would then package and sell to partners, and to train and run AI models on Azure as OpenAI worked to develop next-generation computing hardware.

In the months that followed, OpenAI released a Microsoft Azure-powered API that allows developers to explore GPT-3’s capabilities.(OpenAI said recently that GPT-3 is now being used in more than 300 different apps by “tens of thousands” of developers and producing 4.5 billion words per day.) Toward the end of 2020, Microsoft announced that it would exclusively license GPT-3 to develop and deliver AI solutions for customers, as well as creating new products that harness the power of natural language generation.

Microsoft says GPT-3 will be integrated “deeply” with Power Apps, its low-code app development platform — specifically for formula generation. The AI-powered features will allow a user building an ecommerce app, for example, to describe a programming goal using conversational language like “find products where the name starts with ‘kids.’”

With the new capabilities, a person can get a formula by typing a plainspoken sentence like “Show 10 orders that have stroller in the product name and sort by purchase date with newest on the top.” Power Apps’ tailored GPT-3 model will offer choices for translating the command into Power Fx, Power Platform’s programming language, like “Filter(‘BC Orders’ Left(‘Product Name’,4)=”Kids”).

“The goal of Power FX is to enable people to build apps more quickly by using simple language. With these new features, customers can speak in natural language, and GPT-3 can understand that and put that into the syntax that they’re using,” Eric Boyd, CVP of AI Platform at Microsoft, told VentureBeat in an interview. “What our team did is took the GPT-3 model and produced a specialized model that is particularly appropriate for this particular use case.”

Microsoft says the Power Platform team worked closely with its Azure AI division to fine-tune a GPT-3 model that could translate between natural language and Power Fx expressions. Microsoft engineers used Azure Machine Learning-managed endpoints, a new capability announced in preview at Build, to deploy and manage the GPT-3 model used to deliver the new capabilities in Power Apps.

“Using an advanced AI model like this can help our low-code tools become even more widely available to an even bigger audience by truly becoming what we call ‘no code,’” Charles Lamanna, CVP for Microsoft’s low-code app platform, said in a blog post. “This will allow people to query and explore data in ways they literally couldn’t do before, and that will be the magical moment. In all cases, there is a human in the loop. This isn’t at all about replacing developers — it’s about finding the next 100 million developers in the world.”

Going forward

Despite the potential of natural language models like GPT-3, many blockers exist. The models can’t always answer math problems correctly or respond to questions without paraphrasing training data, and it’s well-established that they amplify the biases in data on which they were trained. That’s problematic in the language domain, because a portion of the data is often sourced from communities with pervasive gender, race, and religious prejudices. AI research firm OpenAI notes that this can lead to placing words like “naughty” or “sucked” near female pronouns and “Islam” near words like “terrorism.” A separate paper by Stanford University Ph.D. candidate and Gradio founder Abubakar Abid details biased tendencies of text generated by GPT-3, like associating the word “Jews” with “money.”

To address these concerns, Microsoft says it has added filters to help detect sensitive or inappropriate content in any results that might get returned in Power Apps. In Power Apps, GPT-3 offers multiple suggestions for Power Fx formulas, and users can choose which to apply. And Microsoft argues that because the model in this circumstance is generating prescribed formulas, unintended outcomes are less likely than if GPT-3 were asked to answer open-ended questions.

“GPT-3 is the most powerful natural language processing model that we have in the market, so for us to be able to use it to help our customers is tremendous,” Power Apps product marketing manager Bryony Wolf says. “This is really the first time you’re seeing in a mainstream consumer product the ability for customers to have their natural language transformed into code.”

Going forward, Microsoft plans to infuse Power Fx into other tools within Power Platform, at which time the natural language features powered by GPT-3 will expand to those products as well. “We’re finding ways to bring it into Azure and our mainstream products,” Boyd said. “We think there are a whole bunch more things that GPT-3 is capable of doing. It’s a foundational new technology that lights up a ton of new possibilities, and this is sort of that first light coming into production.”