Imagine asking an AI, “Why did the player pass the ball?” and it responds not with text, but by highlighting the exact moment of the pass and the player receiving it in the video. This is the exciting frontier of Motion-Grounded Video Reasoning, a task that pushes AI beyond simply recognizing actions to understanding the complex interplay of movement and meaning in videos. Current AI models struggle with this. They can identify a “kick,” but may not grasp the subtle shift in a player’s weight before the kick or the strategic implications of passing the ball to a teammate in a specific location at a crucial moment. This research introduces a new way to train and evaluate AI's understanding of motion. Researchers have created a massive dataset called GROUNDMORE, filled with over 1,700 video clips and thousands of carefully crafted questions that probe an AI’s ability to reason about movement. These questions go beyond simple descriptions. They ask “why” and “what if” questions, exploring causal relationships and counterfactual scenarios. For instance, instead of asking “Who kicked the ball?”, GROUNDMORE asks “Who needed to be passed to in order for the goal to be scored?” This requires the AI to understand not just individual actions, but the chain of events and their underlying causes. The dataset also includes detailed pixel-level annotations of object movements, enabling AI models to provide visual answers by highlighting the relevant objects within the video. The initial results are promising, yet also reveal just how much room there is for improvement. A new baseline model called MORA, combining cutting-edge language models with object recognition capabilities, achieves state-of-the-art performance on GROUNDMORE. However, the scores reveal that truly grasping the nuances of motion remains a significant challenge. This research is a crucial step towards building AIs that can perceive and reason about the dynamic world around us. Imagine the possibilities: AI assistants that can analyze sports footage to provide insightful commentary, security systems that can understand suspicious behavior, or even robots that can learn new tasks simply by watching humans demonstrate them. While there are hurdles to overcome, including privacy concerns and ethical considerations, the potential of motion-grounded video reasoning to revolutionize how we interact with AI is undeniable.