Microsoft Presses Lead in Integrating Generative AI in Search
At an event I attended last week in New York, Microsoft showed off its accelerating investments in making generative AI more accessible and more useful. Might the timing have had something to do with Google’s expected announcements at Google I/O this week? I put the question to Bing Chat, and it agreed that “Microsoft announced the next wave of AI innovation with Microsoft Bing and Edge on May 4th, 2023,” but claimed that, “The reason for this announcement was to tackle a universal problem with traditional search – that nearly half of all web searches go unanswered, resulting in billions of people’s searches falling short of the mark.” OK, but what about Google? “I couldn’t find any information that suggests that Microsoft’s announcement was related to Google’s I/O event.” There you have it: either Microsoft’s Bing product marketing managers are oblivious of the competitive environment, or Bing Chat still doesn’t quite understand corporate marketing.
(Google certainly understands that Microsoft’s lead in incorporating AI into search and productivity apps is a competitive problem; at Google I/O it mentioned “AI” 140 times. Techsponential’s report on Google I/O is coming soon.)
The reason I could so easily but the question to Bing Chat is that the service is now in open preview, meaning anyone can use it without a wait list, but Microsoft is trying to keep expectations in check with the “preview” label. For now, Microsoft is keeping regular search results separate from Chat to ensure that searches for simple factual answers and websites aren’t confused with Chat’s more creative responses. As before, you can customize Chat’s “conversation style” algorithm between “more creative, more balanced, or more precise.”
“Chat” may not be the best name for Microsoft’s extensive generative AI capabilities. Chat is certainly capable of providing chat-like natural language responses, but it can also generate images and artwork, summarize text in long documents, compose poetry, and much more. (Ironically, the one thing that Chat can’t do at the moment is actually chat using voice – Cortana may have been shelved prematurely. Microsoft assures me that voice control is something that is being considered, but that it will not be Cortana.) Chat is not even limited to text or image responses. I was surprised to learn that Chat in Microsoft’s Edge browser can actually manipulate the browser itself, exposing buried settings and grouping your tabs together.
Not all of this functionality is new for AI; Samsung’s Bixby has been doing the ‘help me find this buried setting’ trick for years, and ChatGPT and DALL-E have been through multiple iterations outside of Microsoft products. All the usual caveats about today’s generative AI apply as well (see “Limitations” below). Still, consolidating this functionality into a Search engine and browser preloaded on millions of PCs and exposing it to anyone for use without making them jump through hoops is significant.
But Wait, There’s More
In addition to opening up Bing Chat to more users, Microsoft is also opening it up to external plug ins. This effectively turns Chat into a platform, allowing third parties to add new resources – like technical information and formulas from Wolfram Alpha – or features and capabilities – like information and reservations from Open Table.
Chat responses are moving from just text responses and will now add multimedia like images or video when appropriate.
The reverse process is coming soon; instead of just asking with text and receiving images or video, you will be able to present Chat with an image or video and ask for information and context about it.
Finally, Microsoft recognized that after you make a query, refine it, and get a result you’re happy with, you probably want to do something productive with that data. The latest update lets you save chats and searches within Bing, or export and share them in other formats like Word, PDF, and email. Within the Edge browser, Chats can automatically move over to the side and remain open, enabling you to use the results as a starting point for further exploration within new browser tabs without having to flip between tabs or start the query over.
Limitations
Microsoft is making generative AI more accessible and productive but this doesn’t magically cure AI’s known issues.
I asked Chat to create hieroglyphics of ancient Egyptians using smartphones and drones – something I could imagine as a lighthearted introductory slide for a Techsponential presentation. The results weren’t great, so to add whimsy I asked Chat to do it in the style of the Simpsons. This time, one of the images was hilarious and nearly perfect, but the hands had weird finger overlaps and multiple shoes in odd places. Matt Groening famously chose to give his characters only four fingers to make them easier to animate, but Chat couldn’t manage it.
What if you’re landing in Newark airport (airport code, “EWR”) and want to put an image up on social media that expresses your joy of being home sweet home? I asked Chat to create a framed image of a needlepoint with a picture of a home, with the text, “EWR sweet EWR” underneath. It absolutely nailed the frame, homemade needlepoint aesthetic, and house, but could not generate legible text. I’m sure some future generative AI model will allow text entry, but it does not appear to be possible today. I also tried a version where instead of a house I asked for a needlepoint depiction of Newark Airport. Despite there being hundreds of photos of Newark online, three out of four images produced did not look like airports, and the fourth might be San Francisco International or it might be a roller coaster. It’s definitely not Newark.
Factual inaccuracy continues to be a problem in text responses as well. Microsoft attempts to help you verify the sources of information with citations, but not everything has a citation, and some of the facts can be wrong. I asked Chat to generate 1000 words on the impact of Apple’s rumored VR headset on the VR industry, in the style of a Techsponential report. Techsponential has plenty of reports posted on the web, so the style constraint wasn’t an issue. However, mixed in with a list of companies that Apple has actually acquired, Chat said that Apple hired top talent – like 3D pioneer John Carmack – who absolutely does not work there and never has. There were no citations on any of these facts.
Chat absolutely nailed the happy panda playing Xbox. Nearly every variant was excellent, though on one image (not this one) the controller looks suspiciously like a PS2…
The report itself offered pretty generic analysis, with little of the insight that Techsponential clients expect, but, like the slide images, it would make a great reference to hand to a human analyst or graphic designer as a starting point. Generative AI, when used as a somewhat flawed tool, can be breathtakingly powerful. Not all responses were faulty. I was able to quickly find out how large a business needs to be for it to be listed as a Fortune 100 company without having to read through the entire list. Chat created a cute name for a Star Wars-themed fruit smoothie Microsoft served at its event. And it turns out that Chat is spectacularly good at creating pictures of happy pandas playing Xbox.
Competitive Implications
Even though generative AI is expensive to provide, every point of market share that Microsoft gains for Bing from Google represents billions of dollars. Microsoft reports that Chat is driving improved market share for both Bing and Edge.
Microsoft 365 dominates the office productivity software market, so adding tools and assistants that help to create PowerPoint slides and generate inter-office memos is more about protecting its existing franchise rather than entering new markets. However, Microsoft 365 is an enormous business that is highly profitable, and constantly improving the suite keeps that machine going. Google is adding generative AI features to Workspace, too, and is creating a separate subscription service to monetize it. There are also endless possibilities for Microsoft to use AI features to drive market expansion into legal tools, salesforce automation, customer management, and much, much more.
In the Satya Nadella era, Microsoft is all about cloud and multi-platform, but Windows is still a critical platform for the company. The potential for deep integration of AI tools into Windows (not just the browser and web search) could help Microsoft fight Apple’s increasing success on the desktop.
To discuss the implications of this report on your business, product, or investment strategies, contact Techsponential at avi@techsponential.com.