 Today's AI systems provide us useful recommendations and predictions, but their accuracy depends on lots of annotated data to train neural networks for every new task. To address this, IBM Research is championing a new approach called Neurosymbolic AI. It combines the statistical, data-driven learning capabilities of neural networks with symbolic reasoning techniques. For example, a Neurosymbolic system could use a neural network's pattern recognition capabilities to identify objects in videos or images, then rely on symbolic AI programs that apply logic and semantic reasoning to identify relationships among these objects. The MIT-IBM Neurosymbolic Concept Learner learns by simply looking at images and reading paired questions and answers. The goal is to create explainable AI systems that can tackle more complex tasks while increasing accuracy, learning from fewer examples, and using less data.