Microsoft sheds light on the role of AI and Machine Learning in assessing Windows 10 update rollouts

Microsoft has been under fire recently for messing up Windows 10 updates yet again. The company had to release a major cumulative update to patch all the bugs in the May 2019 Update. Later a former Microsoft employee released a video explaining how Microsoft used to test Windows and what has changed with Windows 10. He put most of the blame at Microsoft using AI to test Windows 10 instead of actual humans.

Microsoft now has published a blog post explaining the role of AI and Machine Learning in assessing Windows 10 upgrades. The company recently marked Windows 10 v1903 for broader deployment and they are using AI to determine what devices will be suitable for the next wave of Windows 10 updates. Microsoft noted that v1803 was the first update that used Machine Learning to determine which devices will be suitable for the updates. Microsoft then separated them from the PCs which might cause issues and sent out updates to the supported ones. They also noted how ML is actually helping the company combat issues and make sure the transition process is flawless. According to the numbers, PCs that got updates through ML have less chance of having kernel issues, crashes and five times fewer post-update driver issues.

Microsoft also shared an extensive ML graph showcasing the overall architecture of ML and how it’s used to nominate PCs that are ready for update rollout. Machine learning also provides two key capabilities:

• It identifies potential issues that result in safeguard holds to protect PCs that have yet to be updated so that those issues can be promptly investigated and fixed by Windows developers.
• It predicts and nominates PCs that will have a seamless update experience and should, thus, be offered the update.

Since Microsoft is using a mix of AI and Machine Learning, the model itself learns from the previous deployments and it predicts the future rollouts better. This ensures that any PC that has incompatible hardware or software doesn’t get the update until the hardware or software has been updated to support the new update.

Microsoft is also using Azure Databricks to ensure they could identify abnormalities and deploy safeguards to protect similar PCs. Apart from Databricks, Microsoft is also relying on old school channels like laborious lab tests, feedback, and support calls to ensure updates don’t roll out to incompatible hardware.

While the company now relies on Machine Learning, they did admit that the model is not perfect and they are working towards making it identify issues in seconds rather than hours. If latest 1903 update is anything to go by, I would say that Microsoft still has a lot of work to do in order to have a perfect ML model in place that could save users from the frustration of having to roll back to an older version because of certain issues. That said, the model does sound good and when perfected, could help Microsoft save a lot in man-hours as well as resources devoted towards identifying the outliers that might break the update on certain computers.