Microsoft has launched an AI-powered update of its Bing search engine. It’s not surprising, as we’ve lived with the same algorithmic approach to search since the launch of Google more than 20 years ago, so any innovation is going to get attention.
Microsoft CEO Satya Nadella has called it “a Mosaic moment” — referring to the early internet browser and pointing to a very interesting future for software. Before the Mosaic browser, the World Wide Web was simply a way to explore and share documents; after the early browser launched in 1993, it became the user interface to the wider internet. AI in Bing is much the same, as it’s the start of a series of experiments that could change the way we interact with software.
How Microsoft is delivering AI-enhanced search
At the heart of Microsoft’s AI-enhanced search is the same co-pilot assistive AI concept it’s using with its developer tools. The AI isn’t what drives the process — the user does. Switch to the chat feature, and what you get is a way to refine your queries and manage how they’re displayed.
For example, I used it to generate a set of instructions for a relatively complex feature I’ve been struggling to use on a new camera. Not only did it give me a clear explanation, but I was quickly able to transform that result into a step-by-step guide and then into an email. Results are backed up with references, so you can go straight to primary sources.
Possible use cases for this AI-enhanced search
There’s something important here for organizations looking to use tooling like this for customer support. Users can ask the questions that matter to them and get data the way they want it. Instead of an FAQ, the interface to your manuals and support database becomes a browser sidebar or a search query.
It’s the same for shopping and e-commerce, with the search tooling building comparison and recommendation tables in response to user prompts, delivering results that come close to Bing’s original mission as a tool to help you make decisions.
Microsoft’s secret sauce: Prometheus
Key to Microsoft’s approach is what it calls its “Prometheus model.” Named after the mythological hero who stole fire from the Greek gods, it’s intended to ensure that humans are in the loop, building on Microsoft’s work on responsible AI. A bad early experience with a chatbot taught Microsoft a lot of lessons, and its work on ensuring AI is safe has been baked into the company from the board-level down.
Prometheus is more than a new version of existing tools like ChatGPT: It mixes OpenAI’s large language models with Microsoft’s experience from Bing and with its responsible AI framework. Where OpenAI’s demo services only take input from a text box, Microsoft injects additional context, including location, date and time, while splitting queries up into multiple parallel searches, aiming to speed up requests and improve the quality of results. Results are parsed and used as inputs into the model, looking for fresh insights that can drive additional searches, an approach that Microsoft described as a “virtuous loop.”
The initial implementation of this AI tooling is in both its Edge browser and Bing search engine. Microsoft gave a picture of how the service will eventually appear in the rest of the company’s productivity tooling. In Edge, a new Bing sidebar lets you take searches outside the traditional browser, giving a clue to how it could be implemented in the familiar Office applications. The new sidebar design Microsoft has been rolling out across Office is very much like the Bing AI tooling that’s being trialed in the developer release of the Edge browser.
While Microsoft has yet to announce any plans to provide API-level access to this tooling, it would be surprising if it wasn’t on the company’s radar. Microsoft is first and foremost a platform company, and its use of generative AI is clearly setting a pointer to what it thinks its next platform should be — assistive AI.
SEE: Artificial Intelligence Ethics Policy (TechRepublic Premium)
An AI platform for business
What would an assistive AI platform working across Microsoft 365 and Dynamics look like? More importantly, how would it work? Microsoft’s Prometheus tooling is designed to provide a set of guardrails around AI tooling, with a focus on what the company called “prompt engineering.”
If you look into early platform tools like the recently launched Azure OpenAI, the API to a generative AI like GPT is very simple: It’s a way to send a string of text to the service. What’s important is getting that string — the prompt — correct.
It’s also interesting to see how the platform helps you find that prompt. With Bing’s OpenAI tools, it’s better to start with a relatively open-ended query and then refine your results. It’s an interactive platform that is designed to help you find what you need, and the best results come from those interactions. Simple queries are handed off to the traditional search algorithms, as there’s no value in using the compute resources required for an AI-powered search.
Then there’s the larger question of training data. While large language models like GPT are trained on public data, that’s less important when it comes to working with the data stored in your own applications. Again, Azure OpenAI offers some clues, with the option to use a custom model that blends the existing GPT LLM with a selection of your own data. Both Dynamics and Microsoft 365 have their own data models, which would work well as part of an expanded Prometheus model, especially as Bing already uses that data in its enterprise search mode.
A contextual computing future
Towards the end of his tenure as CEO of Microsoft, Bill Gates talked about the idea of contextual computing, where systems responded to the context around a user. It’s not much of a stretch to describe Bing’s AI co-pilot as a contextual computing tool, using it to refine queries and results without leaving the context of the original query. Assistive tooling needs context to work, and while for now that context is limited to the scope of a query, it’s clear that there’s an opportunity to go a lot further.
Asking a question about a place might give you a traditional travel output, but refining it might give you an itinerary for a stay and a list of places to eat each day. Links in the answers let you open new pages, while references in Bing’s new chat interface help show where those results came from. All through the chain of queries, there’s no loss of context — you’re focusing the answers from the AI to get the results you want. The service might also reformat data, producing tables to help compare results.
It all adds up to a very different way of working with both search and the web, prioritizing the information users are looking for and presenting it the way they want. Like the first graphical web browsers, it’s still very new and not yet fully formed, but it’s clearly not something we can ignore. This technology will change a lot about how we build and structure content in the future.
SEE: Hiring Kit: Artificial Intelligence Architect (TechRepublic Premium)