Build 2020: Microsoft ‘Goes Big’ on AI and Demonstrates Thought Leadership

R. Bhattacharyya

Summary Bullets:

• AI was featured prominently during Microsoft Build 2020, with a tagline of ‘Putting AI Into Action’ and a goal of bringing state of the art AI to all developers.

• Microsoft made several announcements that supported this vision, including updates on Microsoft Project Turing model, investments in infrastructure for AI processing, and the preview of Project Bonsai.

Microsoft’s annual developer conference, Microsoft Build 2020, was held as a virtual event on May 19 and 20, with more than 200,000 people registered. When kicking off Microsoft Build 2020, CEO Satya Nadella noted that the technology industry is being called upon to address the world’s most acute needs, and that developers are now more important than ever. He pointed out that organizations need the ability to remote everything at a moment’s notice, and to simulate and automate everywhere to enable more agile responses.

AI was featured prominently with a tagline of ‘Putting AI Into Action’ and a goal of bringing state of the art AI to all developers. The company noted that businesses are ready to move beyond narrow pilot programs to broader AI deployments that connect AI to business outcomes. Microsoft highlighted key themes driving its AI vision: Deploying AI at scale, AI for everyone, and Responsible AI, It made several announcements that supported this vision, including updates on its massive Microsoft Project Turing model, investments in infrastructure for AI processing, and the preview of Project Bonsai for machine teaching. As always, Microsoft also emphasized the tools and investments it has made in support of Responsible AI, a topic that’s been near and dear to the tech giant for quite some time.

Project Turing

At 17 billion parameters, Microsoft claims its “Project Turing” AI model is the market’s largest. Project Turing will soon be open sourced, thus giving developers access to the natural language generation model Microsoft enables across its products. The move gives developers access to the same model used by Microsoft to enable natural language generation across its products. Project Turing is trained via ‘self-supervised learning’ using billions of publicly available documents. Microsoft claims that the mega-model has the potential to change how AI is developed: it moves the market towards multipurpose models that developers can customize for their specific purposes. The benefit of developing such large models is that they only need to be trained once (using large amounts of data) and can then be customized for specific tasks using smaller data sets. Developers enjoy the advantages working with more complex and ideally more accurate models (given the large training sets) without the need to develop them from scratch, which would be a daunting endeavor.

AI Processing

Microsoft announced that it had built a supercomputer hosted in Azure that ranks among the top five supercomputers in the world. It was developed with and exclusively for OpenAI, and consists of more than 285,000 CPU cores, 10,000 GPUs, and 400 Gbps of network connectivity for each GPU server. Again, Microsoft is looking to bring state of the art AI processing to all developers by making massive AI models and the infrastructure necessary to train them available as a platform for others to build upon.

Microsoft also announced a new version of DeepSpeed that reduces the amount of computing power needed for large distributed model training, and added support for distributed model training on ONNX Runtime (previously focused on inference).

Project Bonsai

Microsoft’s new autonomous systems platform (Project Bonsai) is available in public preview. Project Bonsai targets professionals lacking deep expertise in machine learning. The platform combines human intelligence with a technique called machine teaching to build intelligence into control systems. Engineers supervise AI agents that address problems in a simulated environment, providing feedback so that the agent dynamically adapts within the simulation. The announcement is part of Microsoft’s larger vision to bring AI to a broader audience, in this case engineers without AI expertise. The company also announced Project Moab, an open source machine teaching robotics hardware kit that allows users to experiment with simulations.

Responsible AI

Microsoft devoted several sessions at Build 2020 to Responsible AI. The company already offers several tools in support of Responsible AI, such as Fairlearn, which assesses the fairness of a model and can be used to mitigate unfairness, and Interpret ML, which helps users understand the factors influencing model results. The company also offers best practices based on its own experiences implementing AI. It has established an Office of Responsible AI that develops internal policies, supports customers, evaluates sensitive use cases of AI, and participates in public policy discussions. Microsoft’s AI, Ethics, and Effects in Engineering and Research (Aether) Committee advises the company’s internal leadership on the challenges raised by AI innovations. In April Microsoft launched an AI Resource Center that informs customers’ understanding of multiple aspects of Responsible AI, including Practices (guidelines), Tools (for understanding models, data protection, and model management), and Insights (podcasts and webinars). Its AI Business School offers classes for business leaders with modules on the need to put Responsible AI into practice by establishing guidelines and governing practices.

Microsoft announced innovations in responsible machine learning that can help developers understand, protect and control their models throughout the machine learning lifecycle. Going forward, the company plans to release tools to make it easier for customers to ensure they are adhering to corporate ethics policies. It plans to offer a Harm Model framework and to expand its guidance related to best practices. It will enhance Interpret ML with new interpretability techniques, including explanations as to why a model may have errors.

Conclusion

Many organizations are focused on bringing AI to a larger community. They prioritize tools to enable non-data scientists to incorporate AI into their applications more easily or provide operationalized solutions. Microsoft is also doing this, but has stepped up the message with ‘bigger is better’. Whether it’s with a bigger supercomputer, or a massive Turing model, the company is communicating that there is much to be gained by scaling up, and that it can offer the tools that allow the developer community to take advantage of Microsoft’s size. Does Microsoft need to make this kind of splash? It doesn’t hurt, especially at a time when competitors such as IBM and Google are playing up their investments and successes with quantum computing. To counter competitors, Microsoft wants to communicate that it isn’t just offering AI – it is offering ‘state of the art’ AI. (Microsoft also announced that Azure Quantum, which offers access to quantum software, solutions, and hardware from Microsoft and partners, is now in preview.)

Most companies probably do not really need this kind of power and scale but would likely benefit from the thought leadership that comes along with this vision. For example, the company’s Responsible AI resources have expanded significantly and the experience Microsoft has in the area could greatly benefit its customers. Also, innovations that streamline adoption and reduce costs will be welcomed by customers. Going forward, offering toolkits such as Project Moab would do much to help non-AI specialists expand adoption of AI with new use cases, and industry-specific operationalized solutions would also help customers realize the value of AI more easily.

Microsoft is stepping up its game, not just in the tools it offers, but also in its processing capabilities as well as in thought leadership. Most customers may not require this kind of scale, but they can rest assured that Microsoft is investing in the technologies that will help them streamline future implementations.

 

What do you think?

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.