𝗖𝗮𝘂𝘀𝗮𝗹 𝗔𝗜 𝗘𝘅𝗽𝗹𝗮𝗶𝗻𝗲𝗱: 𝗗𝗮𝘆 𝟲 - 𝗖𝗮𝘂𝘀𝗮𝗹 𝗗𝗶𝗮𝗴𝗿𝗮𝗺𝘀 & 𝗦𝗖𝗠𝘀 We’ve explored correlation vs. causation, the do-operator, confounders/colliders/mediators, interventions, and counterfactuals. Now, let’s discuss the backbone of Causal AI: 𝗖𝗮𝘂𝘀𝗮𝗹 𝗗𝗶𝗮𝗴𝗿𝗮𝗺𝘀 and 𝗦𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗮𝗹 𝗖𝗮𝘂𝘀𝗮𝗹 𝗠𝗼𝗱𝗲𝗹𝘀, also known as the map and engine powering it all. 𝗖𝗮𝘂𝘀𝗮𝗹 𝗗𝗶𝗮𝗴𝗿𝗮𝗺𝘀 (𝗗𝗔𝗚𝘀) Think of a Directed Acyclic Graph (DAG) as a treasure map. Nodes are variables (e.g., discount, purchase), arrows show cause-and-effect (𝗱𝗶𝘀𝗰𝗼𝘂𝗻𝘁 → 𝗽𝘂𝗿𝗰𝗵𝗮𝘀𝗲), and the layout reveals confounders (e.g. income) or mediators (e.g. interest). 𝘋𝘈𝘎𝘴 𝘷𝘪𝘴𝘶𝘢𝘭𝘪𝘴𝘦 𝘵𝘩𝘦 "𝘸𝘩𝘺" 𝘣𝘦𝘩𝘪𝘯𝘥 𝘥𝘢𝘵𝘢, 𝘨𝘶𝘪𝘥𝘪𝘯𝘨 𝘶𝘴 𝘵𝘰 𝘣𝘭𝘰𝘤𝘬 𝘣𝘪𝘢𝘴𝘦𝘴 𝘰𝘳 𝘴𝘪𝘮𝘶𝘭𝘢𝘵𝘦 𝘪𝘯𝘵𝘦𝘳𝘷𝘦𝘯𝘵𝘪𝘰𝘯𝘴. 𝗦𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗮𝗹 𝗖𝗮𝘂𝘀𝗮𝗹 𝗠𝗼𝗱𝗲𝗹𝘀 (𝗦𝗖𝗠𝘀) SCMs are the engine under the hood. They represent causal relationships as equations 𝗬 = 𝗳(𝗫, 𝗨) where Y is the outcome, X the cause, and U some random noise or unobserved factors. Note that X itself can be the outcome of a parent causal relationship i.e. X = g(Z,V). 𝘚𝘊𝘔𝘴 𝘳𝘦𝘱𝘳𝘦𝘴𝘦𝘯𝘵 𝘵𝘩𝘦 "𝘸𝘩𝘺" 𝘶𝘴𝘪𝘯𝘨 𝘮𝘢𝘵𝘩𝘦𝘮𝘢𝘵𝘪𝘤𝘴, 𝘰𝘱𝘦𝘯𝘪𝘯𝘨 𝘵𝘩𝘦 𝘥𝘰𝘰𝘳 𝘧𝘰𝘳 𝘮𝘰𝘳𝘦 𝘮𝘢𝘵𝘩𝘦𝘮𝘢𝘵𝘪𝘤𝘢𝘭 𝘵𝘰𝘰𝘭𝘴 𝘪𝘯 𝘧𝘶𝘳𝘵𝘩𝘦𝘳 𝘤𝘢𝘶𝘴𝘢𝘭 𝘴𝘵𝘶𝘥𝘪𝘦𝘴. 𝗥𝗲𝗮𝗹-𝗪𝗼𝗿𝗹𝗱 𝗘𝘅𝗮𝗺𝗽𝗹𝗲 Imagine a retailer: Sales dropped despite ads. A DAG shows "𝗮𝗱𝘀 → 𝘀𝗮𝗹𝗲𝘀", confounded by seasonality. The SCM does the maths: 𝗦𝗮𝗹𝗲𝘀 = 𝗮 * 𝗔𝗱𝘀 + 𝗯 * 𝗦𝗲𝗮𝘀𝗼𝗻 + 𝗨 We can then ask: "What if ads ran in spring instead of winter?", simulating outcomes with precision and proper visual tools. 𝗪𝗵𝘆 𝗜𝘁’𝘀 𝗣𝗼𝘄𝗲𝗿𝗳𝘂𝗹 • Clarity: DAGs untangle complex relationships in a glance • Precision: SCMs quantify causal effects with equations • Versatility: Together, they fuel every "what if" from interventions to counterfactuals 𝗧𝗮𝗸𝗲𝗮𝘄𝗮𝘆: 𝗖𝗮𝘂𝘀𝗮𝗹 𝗗𝗶𝗮𝗴𝗿𝗮𝗺𝘀 𝗮𝗻𝗱 𝗦𝗖𝗠𝘀 𝗮𝗿𝗲 𝗖𝗮𝘂𝘀𝗮𝗹 𝗔𝗜’𝘀 𝗿𝗲𝗽𝗿𝗲𝘀𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝘃𝗲𝗵𝗶𝗰𝗹𝗲𝘀, 𝗺𝗮𝗽𝗽𝗶𝗻𝗴 𝘁𝗵𝗲 ‘𝘄𝗵𝘆’ 𝗮𝗻𝗱 𝗰𝗮𝗹𝗰𝘂𝗹𝗮𝘁𝗶𝗻𝗴 𝘁𝗵𝗲 ‘𝘄𝗵𝗮𝘁 𝗶𝗳’ 𝗳𝗼𝗿 𝘀𝗺𝗮𝗿𝘁𝗲𝗿 𝗱𝗲𝗰𝗶𝘀𝗶𝗼𝗻𝘀. What’s a messy problem you’d map with a DAG? Share below! 𝘐𝘮𝘢𝘨𝘦 𝘤𝘳𝘦𝘥𝘪𝘵: 𝘊𝘢𝘶𝘴𝘢𝘭 𝘐𝘯𝘧𝘦𝘳𝘦𝘯𝘤𝘦 𝘌𝘹𝘦𝘳𝘤𝘪𝘴𝘦𝘴, 𝘍𝘢𝘣𝘪𝘢𝘯 𝘋𝘢𝘣𝘭𝘢𝘯𝘥𝘦𝘳 #CausalAI #DataScience #Causality #ArtificialIntelligence #AIExplained #CausalDiagrams #SCMs
