Artificial Intelligence (AI) continues to be a key area of investment for Microsoft, and we’re pleased to announce that for the first time we’ve leveraged AI at scale to greatly improve the quality and reliability of the Windows 10 April 2018 Update rollout. Our AI approach intelligently selects devices that our feedback data indicate would have a great update experience and offers the April 2018 Update to these devices first. As our rollout progresses, we continuously collect update experience data and retrain our models to learn which devices will have a positive update experience, and where we may need to wait until we have higher confidence in a great experience. Our overall rollout objective is for a safe and reliable update, which means we only go as fast as is safe.
Our AI/Machine Learning approach started with a pilot program during the Windows 10 Fall Creators Update rollout. We studied characteristics of devices that data indicated had a great update experience and trained our model to spot and target those devices. In our limited trial during the Fall Creators Update rollout, we consistently saw a higher rate of positive update experiences for devices identified using the AI model, with fewer rollbacks, uninstalls, reliability issues, and negative user feedback. For the April 2018 Update rollout, we substantially expanded the scale of AI by developing a robust AI machine learning model to teach the system how to identify the best target devices based on our extensive listening systems.
To me, it sounds like simple if-else statements would have implemented the whole thing without touching the AI; they mentioned that positive experiences include fewer rollbacks, uninstalls, and so on, so we may use these as a criterion of a positive experience.
I am just wondering if the word 'AI' is being misused, or can be misleading in this context? Could anyone point me out on this or give any insight on how AI can be used in this context? In my experience, I have only seen AI mostly being used in speech recognition, image recognition and other sort-of classifying problems, with a training and consequently a computer can "learn" from the data, not like an if-else statement. Today, AI seems to be everything that is considered "smart"?