In the field of Artificial Intelligence-driven healthcare systems, human motion detection is becoming increasingly popular as it can be applied to give remote healthcare for vulnerable people. This paper aims to develop a contactless AI-enabled Healthcare system, aimed to detect human motion using Channel State Information (CSI) from wireless signals, which can record patterns of human movements. Although human motion detection systems have been developed using wearable devices, this system still leaves many issues that cannot be solved. For some disabled and elderly people, it is difficult and easily forgotten to wear the devices. Thus, to tackle those issues, a novel method is proposed by using non-wearable methods. We first produced a dataset of CSI that contains patterns of human motion by using software-defined radios. Next, machine learning algorithms like Neural Network (NN), K Nearest Neighbors (KNN), Random Forest, and Support Vector Machine (SVM) were applied to processed CSI data to classify different human activities. Finally, we ensembled the three best-performed classifiers as the healthcare system to reduce the possibility of False Positive cases or True Negative cases. The ensemble classifier can achieve an accuracy of around 98% using 70% data for training and 30% data for testing. This is much higher in contrast with a benchmark dataset measured by accelerators of wearable devices with an accuracy of around 93%, proving the effectiveness of the non-invasive method.