Machine teaching is a new, complementary approach to machine learning that can be used by those without AI expertise. With machine teaching, a complex problem is broken into individual skills and give the AI important clues about how to learn faster, by focusing on extracting knowledge from experts, rather than only from data.
While traditional reinforcement learning is a time-consuming approach with lots of trial and error, machine teaching accelerates and improves the training process and even allows engineers to reuse the individual steps.
Because companies can’t afford to take critical equipment offline or risk damaging a system while the AI learns, the Reinforcement Learning process takes place in safe and cost-effective simulated environments, that can replicate millions of different real-world scenarios, including edge situations like a sensor failure, so the AI can learn how to adapt.
Machine teaching also makes it easier to understand and audit the autonomous control system’s behavior once it’s been deployed, which is crucial for safety-critical applications.