Demystifying AI at Scale
Fast-moving artificial intelligence (AI) capabilities are empowering organizations big and small, helping them to stay resilient and fuel growth. However, many organizations face barriers when it comes to tapping AI’s full potential, including access to massive-scale infrastructure, and the large amounts of data and data science resources needed to train AI models.
A revolutionary new approach to AI, that we refer to as “AI at Scale,” removes these obstacles. It does so by giving organizations access to state-of-the-art large-scale AI models, training optimization tools and supercomputing resources, which together can lead to differentiated products and services, and accelerate time to market.
Inspired by early breakthroughs in large language models, AI at Scale is a marked shift from the creation of many smaller, purpose-built AI models to accomplish individual tasks, such as language translation and object recognition, to a new class of powerful large-scale models that are becoming more general purpose.
For the past two years, Microsoft has been training large state-of-the-art AI models, such as those in the Turing family, for understanding language and the combination of language with other dimensions of data, including images and video. In addition, we have complemented our portfolio of AI models through exclusive licensing of models from our partner OpenAI, i.e., GPT-3 for language generation and Codex for language-to-code generation.
Microsoft is already using these models in a broad range of scenarios across its services (Bing, Office, Dynamics 365, Power Platform, GitHub and LinkedIn). Now, we’re making these transformational capabilities, built on cutting-edge advancements in AI, available for organizations to build upon and customize.
Below, we look at how companies can access AI at Scale and why it’s relevant to your business. Plus, we share some innovative examples and how to get started.
How to access AI at Scale
Companies can benefit from the capabilities of AI at Scale in three ways.
First, organizations using Microsoft products automatically experience the myriad productivity and creativity benefits of our large AI models, such as richer semantic search and question-answering, more natural conversations, automated knowledge extraction and suggestion and generation of code.
Second, organizations can leverage the capabilities of our models through Microsoft Azure services such as Azure Cognitive Services, which support a range of common language tasks of interest, and Azure Cognitive Search, which supports semantic search and question answering. Customers can also leverage the new Azure OpenAI service (in private preview), which makes OpenAI models like GPT-3 directly available to customers to support their own specific tasks with the enterprise-grade capabilities built into Microsoft Azure. We are committed to making all our large state-of-the-art AI models directly available through Azure for customers to use as a basis for any task of interest. In the meantime, customers can request access to the Microsoft Turing models for their specific tasks through the Turing Private Preview.
Third, organizations wanting to leverage the same infrastructure and platform that Microsoft and OpenAI use to train and serve their state-of-the-art large AI models can do so at any level of scale desired. The Azure AI Infrastructure is available from only 8 GPUs in one virtual machine to 6,000 interconnected GPUs across many hundreds of machines (which at this scale qualifies as a top 10 supercomputer in the world). Azure Machine Learning Service supports the lifecycle of machine learning processes like training and inferencing of models while enabling the efficient utilization of the underlying AI infrastructure.
Why AI at Scale is relevant to businesses
Early-adopting organizations are starting to see AI at Scale transform business functions. For example, in product development, companies are using it to track millions of data inputs from social and online activity to discover new consumer trends. Combining these insights with their internal product catalogues they’re able to use AI to identify potential gaps in their product offerings and make recommendations to expand into new markets altogether based on the latest trend data.
Likewise, we’re seeing successful use cases in customer service, where virtual agents can handle complex questions, resolving issues faster and doing so 24/7. In marketing, our large-scale AI models’ nuanced grasp of text, video and images create trustworthy, original and impactful content, including blogs, ad copy and emails. Also, other scenarios include finance, where models detect fraud across fraud protection networks; strategy, where they’re used to summarize content to assess competitors; and software development, where these models help developers code faster.
AI at Scale in action
Some early deployments of AI at Scale have been in industries tackling significant business challenges. Take, for example, healthcare, where current projections show medical knowledge doubling every 73 days compared to every three-and-a-half years in 2010. AI at Scale is being used, among other things, for clinical research and information management, with organizations customizing our large natural-language models with the latest data from clinical trials and medical publications. This significantly improves their ability to search and gain insight over vast amounts of information, accelerating the research and discovery process.
PhactMI, a non-profit collaboration of over 30 pharmaceutical companies including Novartis, GSK and AstraZeneca, uses the technology to improve document search and summarization. This helps medical information professionals produce faster response documents to healthcare providers’ pharmaceutical inquiries. It’s part of phactMI’s mission to support healthcare professionals provide quality patient care.
And software company AvePoint is using the Microsoft Turing model to create a personalized learning experience for its team members, extracting key knowledge across its product guides, release notes and customer support history, and unstructured data. The result: AvePoint is creating unique learning guides and testing materials for each employee. Given today’s business climate, onboarding new employees has never been more important, and the latest advancements in AI at Scale can help by creating and updating training processes, summarizing learning resources and customizing training plans for new joiners.
Whether you’re looking to improve product development, speed up research and reasoning over vast amounts of information or identify your company’s future growth areas, there are three simple steps to take.
First, think about where these capabilities can have the most immediate impact on your business. Maybe it’s in your product lifecycle, or in optimizing your marketing campaigns. Perhaps it’s in helping with your growth strategy. Microsoft AI Business School, an online series designed to help leaders develop a holistic approach to AI, provides frameworks and guidance to identify AI use cases, evaluate AI investments and more.
Next, formulate a strategy with a clear business outcome for leveraging advanced AI models to power natural language understanding and generation within your products and services. You can do this with a pilot project by accessing the models through Azure Cognitive Services, Azure Cognitive Search and the new Azure OpenAI service.
And third, no matter where you are in your AI and innovation journey, it all starts with learning. Learn more about how businesses accelerate innovation and build competitive edge with AI at Scale.
Take advantage of next-generation AI and start unlocking innovation opportunities for your business today!
- Read about the AI advancements and technology stack behind AI at Scale with insights from Microsoft Research
- Learn more about AI innovation powered by AI at Scale
- See how we help computers more fully perceive the nuances of our world
- Explore Microsoft Research on AI at Scale
- Put AI into action with Microsoft AI Business School
- Try the AI Lab demo of AI at Scale
- Learn how it works with a deep dive into AI at Scale technology