Advanced Computing in the Age of AI | Thursday, March 28, 2024

Microsoft, Google Set AI Paths for Search Engines in Historic Week 

This week could go down as a disruptive moment in technology and the first major transition in close to 30 years in the way people search for information on their devices.

Over back-to-back days, Microsoft and Google announced their next-generation search engines with artificial intelligence as a foundational layer. The search engines are adding a human element by answering questions with conversational responses, and this will apply to text and visual searches.

The fundamental change is an algorithmic approach to answering user queries, which is an addition to the conventional directory and crawler approach. The search engines will be able to handle more complex queries, and generate full answers to questions, in addition to linking to websites.

On Tuesday, Microsoft announced its AI-powered Bing, in which users can also ask the search engine to create content or summarize a financial report from a press release. The underlying AI technology, called Prometheus, is based on an advanced version of GPT-3, which has been customized by Microsoft for implementation in its search engine.

Users can join the Bing chatbot waitlist here. A new Edge browser with the AI-powered experiences can be downloaded from Microsoft's website.

Microsoft in late January announced it had infused an additional $10 billion in OpenAI with the goal to put its AI technology across product lines. Microsoft's Satya Nadella compared the launch of ChatGPT late last year to be as important as the launch of the first Internet web browser, Mosaic, close to 30 years ago.

Satya Nadella, Microsoft CEO

"It's a new day in search," Nadella said, later adding "rapid innovation is going to come. In fact, a race starts today in terms of what you can expect and we're going to move fast."

Microsoft's race is with Google, which on Monday preempted Microsoft by announcing that its next-generation AI technologies would be integrated in its search engine and apps. In a blog entry, Google's CEO, Sundar Pichai, announced the availability of Bard, a conversational AI, which is powered by its internally developed large language model called LaMDA.

Google on Wednesday shared more details on how it plans to use Bard with its text-based search engine. Users can ask questions on finding the best car based on safety features, location, and nature of use. "You'll be able to interact with Bard to explore complex topics, collaborate in real time, and get creative new ideas," said Prabhakar Raghavan, who runs the search business as a senior vice president at Google.

The search giant's presentation also covered plans on large-scale AI implementations in visual search, but didn’t cover specific enhancements to its image or video search tools. The company also talked about plans to combine text, image, video, sound and other AI modalities spanning across applications such as Maps, Translate and Lens.

Bard search example (Google)

Google will make the generative language API powered by LaMDa available to developers in the coming months, and then subsequently release other APIs. Google has many AI models in research including Imagen, which can generate images and video based on text descriptions, and PaLM, which is a large language model with 540 billion parameters.

Analysts agreed that Microsoft's presentation was more substantive with actual product announcements and a real-world outlook on bringing AI to its search engine and offerings. Google's presentation, by comparison, was lighter on product announcements, but richer in vision of how it wants AI to look across product lines.

Google's announcement appeared more iterative, whereas Microsoft's announcement was more transformative, said James Sanders, principal analyst at CCS Insight.

"I wouldn't characterize one as necessarily ahead of another, but the priorities are clearly different: Microsoft is working to redefine search as a way to differentiate Bing. Google's new features spread across the product portfolio, but prioritize visualization rather than plain text," Sanders said.

The announcements are also statements of intent, as integration between AI and search is still in its very early days. AI in search can be viewed as a disruptor, and like every new technology, could take time to mature.

"The problem we have had with so much tech recently is that the advances are incremental – like ‘this year's iPhone is 10% faster.’ And then something like ChatGPT comes along, and you are like, 'I have never seen anything like this. We can apply this to something that every person with a digital device does every day multiple times, which is search," said Bob O'Donnell, principal analyst at Technalysis Research.

Bing search example (Microsoft)

Microsoft's Prometheus engine behind Bing was characterized as a more advanced version of ChatGPT. It combines a user interface and other features in a front-end to present the results to users. Prometheus has a trained front-end – typically called 'human in the loop' – to better understand the way humans ask questions of machines, which is then connected to Microsoft's version of the GPT large-language model.

"Prometheus is the intermediary piece that figures out, 'do I just hit the traditional Bing search index that has all the latest news and information? Do I go straight to the ChatGPT portion? Or more likely, do I do some combination thereof?' That is where the real magic is," O'Donnell said.

Nadella considered the human-in-the-loop process as being a core part in its AI offerings.

"When you think about AI, you can have the human in the loop, ... human on the loop or ... human out of the loop. Those are decisions we as product makers build into products. Whenever we have come up with some new things and new models, we put a premium on human agency," Nadella said.

Prometheus can figure out how to create the appropriate queries, and use ChatGPT-style technology to generate answers. If the ChatGPT-style tech does not have a substantive answer, it can turn to Bing search to find and generate new learnings, and then present it in a natural language method to users. Prometheus is an orchestrator of the process.

Bing's summaries have citations that can take search engine users back to the source of the information, which is something that Google has not shown with Bard, O'Donnell said.

"That is a big deal for people who depend on a huge percentage of the traffic coming from search engines. We have not even talked about what Amazon, Apple or Facebook are going to do. There is a ton going on," O'Donnell said.

Google's Raghavan said his company wasn't rushing to push AI into search because it wanted to do so responsibly.

"From Bard, to the new AI powered features in search, to image generation, APIs and beyond, when it comes to AI, it's critical that we bring these experiences, the models to the world responsibly. That's why we've been focusing on responsible AI since the very beginning," he said.

The infusion of transformative AI technologies like large-language models into search and other offerings is a long-drawn process as it could impact advertising revenue, which is a source of income for the companies, analysts said.

Both Microsoft and Google presented examples of how users could get better product recommendations based on their specific needs with the AI-powered search engines.

Microsoft is building its AI product line around technologies developed by OpenAI, whose portfolio includes DALL-E 2, which generates images through a text description, and CODEX, which is a coding automation tool that Microsoft is already using in its co-pilot feature of Github code repository platform.

Google’s AI architecture is called Pathways, which was introduced in 2021 to “handle many tasks at once, learn new tasks quickly and reflect a better understanding of the world,” the company said in a blog entry.

EnterpriseAI