Imagine interacting with a computer that truly understands your intentions, beliefs, and desires—a machine that can "read your mind." This seemingly futuristic concept is at the heart of Theory of Mind (ToM), a crucial cognitive ability that allows humans to navigate the complexities of social interaction. While humans develop ToM naturally, imbuing this ability into artificial intelligence, specifically Large Language Models (LLMs), has been a significant challenge. LLMs often stumble over the intricate logical chains involved in ToM reasoning, especially when dealing with higher-order beliefs (what one person thinks another person thinks). A new research paper introduces "TimeToM,” a novel approach that uses the concept of “temporal space” to help LLMs grasp these complex social dynamics. TimeToM works by creating a timeline of events within a story or conversation. This timeline allows the LLM to track each character’s beliefs at different points in time, constructing what the researchers call a ‘Temporal Belief State Chain’ (TBSC). The TBSC not only captures what each character knows about the world, but also what they know about other characters' knowledge—crucial for higher-order ToM reasoning. Inspired by cognitive science, TimeToM also differentiates between ‘self-world’ beliefs (what a character believes about the physical world) and ‘social-world’ beliefs (what a character believes about others’ beliefs). This distinction helps LLMs tackle different types of ToM questions, like where someone will look for an object versus where they think someone else will look. To conquer the challenge of higher-order ToM reasoning, TimeToM introduces a ‘belief solver.’ This tool analyzes the timelines of different characters to determine when they share information (belief communication periods). By pinpointing these moments of shared understanding, the belief solver simplifies complex higher-order questions into simpler first-order questions, making the reasoning process more manageable for LLMs. Results across various benchmarks show significant improvements in LLMs' ability to answer complex ToM questions, especially those involving higher-order beliefs. However, limitations exist, particularly with smaller LLMs that struggle to accurately construct the initial belief chains. The future of AI hinges on the ability to comprehend and respond to complex human behavior. TimeToM brings us one step closer to AI that can truly understand our intentions, beliefs, and desires—a future where computers aren’t just intelligent, but socially intelligent.