Evolving from Windows Live Search, Bing’s 2009 launch introduced several notable machine learning features at its inception, including search suggestions during query input and a list of related searches, known as the “Explore pane.” These features leveraged semantic technology from Powerset, a company Microsoft acquired in 2008. 

A breakthrough from Microsoft Research, Deep Residual Networks, aka ResNet, was introduced in 2015. The residual learning framework significantly improved the training of deep neural networks, enabling the development of deeper architectures with better performance. It helped open the door for a wide range of practical applications, many of which are critical to our lives today.

“ResNet set the standard for computer vision that’s used everywhere,” Peter Lee, president of Microsoft Research, said. “If you have a self-driving car, it’s using AI based on ResNet. If you’re getting an MRI in the clinic, that MRI machine also has technology based on ResNet.”

With the goal of creating technology that increases accessibility, Microsoft introduced Seeing AI in 2016. This real-world example of AI uses computer vision and natural language processing to describe a person’s surroundings, read text, answer questions, identify currency and even describe people around you and facial expressions.

The Find My Things feature helps people who are blind or have low vision personalize their experience by teaching the tool to recognize the items they actually use and potentially misplace. These may include small items like house keys or earbuds.

Accessibility makes Microsoft tools and technologies easier for everyone and accelerates innovation for the world, including products such as Reading Coach, which helps students learn to read and supports emerging readers and learners with dyslexia.

In 2020, Microsoft released Turing-NLG (Natural Language Generation), one of the largest language models of that era at 17 billion parameters. This showed Microsoft’s leadership in the development of AI for natural language understanding and predated the development of other foundational models like Florence for visual recognition.

Also in 2020, Microsoft introduced a new supercomputer hosted in Azure for OpenAI, establishing new infrastructure to train large AI models and a first step in making them available as a platform. At the time of its release, compared with other machines listed on the TOP500 supercomputersin the world, Microsoft’s supercomputer ranked in the top five. In the following years, Microsoft expanded this infrastructure with AI supercomputers of unprecedented scale for major advancements in application performance for AI training and inferencing.

The partnership with OpenAI was formed on a foundation of shared ideals and a drive to responsibly advance the state-of-the-art in AI with more ambition than anyone else. Together the two companies built an unprecedented supercomputing infrastructure in Azure to accelerate OpenAI’s breakthroughs, and through that collaboration have been able to better understand not just how to put together the world’s best AI platform, but more importantly, how to deliver its benefits to customers and to the world.

In 2023, Bing and Edge became AI powered, helping create a seismic shift in the way people used search. It also led to rapid Copilot evolution and integration across the Microsoft stack over the next year-plus. Copilot continued to evolve, becoming not just an informational tool but also a true AI companion with the release of the updated Copilot app for consumers. The introduction of Voice and Vision allows Copilot to adapt to the user and provide answers, encouragement and advice as people navigate life’s everyday challenges.

Following the initial launch in Bing in 2023, Microsoft rapidly integrated Copilot across its product portfolio. Microsoft 365 Copilot helped people be more productive, while Dynamics 365 and Power Platform helped transform business processes.

Building on the work begun in 2019 with the introduction of the Surface Pro X, Copilot+ PCs introduced in 2024 an all-new system architecture to bring together the power of the CPU, GPU and a new high performance Neural Processing Unit (NPU). Leveraging the capabilities of the NPU, Copilot+ PCs achieved a level of performance never seen before – up to 20 times more powerful and up to 100 times as efficient for running AI workloads. This ushered in a range of new AI experiences that were previously impossible, making getting things done on a PC faster, simpler and more personalized.

This 2024 research-to-product innovation pioneered the category of small language models to unlock cost-effective, customized model innovation and make AI accessible to more people, for example, on smartphones and other mobile devices that operate “at the edge,” and not connected to the cloud. Since then, Microsoft has introduced more Phi iterations to the Azure AI Foundry model catalog including industry-specific Phi models to help customers address specific use cases.