Machine Learning

Machine Learning

Microsoft’s new Bing AI chatbot could be coming to the enterprise, too
Microsoft's ChatGPT-based Bing bot is all about consumers (and advertisers) for now. But the technology could find its way into business software and services, maybe in the not-too-distant future.
A stylized image showing Bing, Search and Chat from the new Bing search

Microsoft's new Bing chatbot is a consumer-focused technology. To gain access to the preview, users need to sign in with their Microsoft Accounts; it doesn't work with Work or School accounts. But there are signs that Microsoft won't take long to integrate the new Bing bot technology into some of its business products and services, as well.

A little more than a week after its splashy introduction, the new ChatGPT-powered Bing chatbot is showing up in more places. As of today, February 22, it's now available in preview in Bing on iOS and Android devices, as well as an integrated part of the latest Skype client release.

Why start with consumer, rather than business scenarios? Microsoft often uses consumers as unofficial testers before making products available to business customers. And given that the new Bing Bot has been behaving badly and got put in a timeout, caution is warranted.

Then there’s the money angle. Microsoft officials no doubt are hoping that the new "AI-powered answer engine," as they're referring to the new Bing, will help the company grow market share for Bing and Edge, and pull more Microsoft Advertising dollars along with it.

Microsoft already enables customers to roll their own OpenAI integration via its paid Azure OpenAI service. And it has integrated OpenAI's GPT-3+ technology—minus the chatbot part—into Power BI, Viva Sales and Teams Premium. It also offers developers access to Bing programming interfaces for their own custom use, but given the huge price hike for these APIs that is coming this May, that option is looking less viable than ever. Add to that the complexity and overhead required to generate large-language models, many customers will be looking to Microsoft to do the heavy lifting here rather than attempting to do it themselves.

In creating the new Bing bot, Microsoft built on top of "the next generation" OpenAI GPT model (which is something more recent than GPT 3.5, but seemingly not GPT 4.0). Microsoft integrated this model with its Bing indexing, ranking and search results, along with Microsoft's Responsible AI Framework, and christened the combination "Prometheus." The new model powers the updated version of Bing search, as well as the new Bing chat mode.

What might be next for the Bing chatbot? Given that Teams and Skype share the same underlying foundation, maybe the Bing bot will show up in not just the consumer/free version of Teams, but in the Microsoft 365 one, too, now that Microsoft is previewing the new Bing bot integrated into Skype.

There also are rumors that Microsoft is planning to add ChatGPT capabilities to its Office apps in March, according to The Verge. (It's not clear if Microsoft will limit this, at least initially, to a subset of Office apps for families/individuals by making this capability reliant on a Microsoft Account, rather than a Work/School one.)

Microsoft already offers AI integration in Office via a variety of other AI technologies like Editor in Word and Designer in PowerPoint. But as the new Bing bot told me when I asked, the current AI features in Office "are mostly focused on enhancing the user experience and productivity, such as suggesting captions, improving acoustics, or auto-completing code." The coming ChatGPT-powered in Word, PowerPoint and Outlook would be more focused on "new and advanced" capabilities like "generating graphs, graphics, and text from user input."

(To be frank, I wouldn't put too much faith on these Bing-bot-generated answers, since even though they sound authoritative, they change every time I ask the same query.)

There are more hints about where Microsoft is dabbling with Bing-bot integration in other business areas.

A highly likely candidate: Integration of the ChatGPT bot technology with Microsoft's Power Virtual Agents (PVAs). PVA is the unofficial successor to Microsoft's Bot Framework. It's meant to help people create chatbots. It seems like a no brainer to make creating chat bots even simpler by adding a chat-centric front-end. A number of people already are finding ways to do this integration themselves.

It's also worth pointing out that Microsoft is testing a number of "modes" for the new Bing chat bot that could extend its usefulness. In addition to a Game mode, there's also a Friend mode, the regular Bing Chat "Sydney" mode and an Assistant mode (hello, Bing Concierge Bot 2.0!), as Bleeping Computer recently discovered. After the author unearthed these modes, Microsoft cut off public access to them.

There are possible new vertical/domain-specific uses, as well. On the research front, Microsoft is experimenting with integrating GPT and natural language processing capabilities and training a new "BioGPT" model on biomedical literature. Its researchers also have done work in using ChatGPT as a front-end for robotics tasks.

I'm seeing more and more reports of businesses discouraging, if not outright banning, their employees from using ChatGPT at work. But Pandora's box has been opened. More rules and tools are needed before enterprise customers will begin to trust ChatGPT and its progeny. But there's no turning back now.

Subscribe to Machine Learning

Illustration shows an image created by OpenAI’s generative AI technology; text explains how generative AI differs from traditional approaches.

OpenAI’s ChatGPT, along with related services such as DALL-E and Codex, are based on a new technology called generative AI that heralds a series of changes for the Microsoft portfolio.

Cognitive Services and AI Applied Services are collections of prebuilt AI components that help speed development, compared to building capabilities manually.

The Azure Machine Learning service can now be deployed anywhere, which could eliminate compliance, security, and latency barriers for using the technology.

Screenshots show how Microsoft’s AI Dashboard, in preview, can help identify potential errors and risk items in training data and models.

Sidebar explains how bias can creep into AI-enabled applications.

A new AI dashboard, in preview, can help organizations understand potential errors and bias in their models.

Illustration shows a screenshot of Azure ML Studio.

Sidebar discusses Microsoft machine learning offerings outside of Azure Machine Learning.