AI Ecosystem CourseLearn with Q
AI Ecosystem · In Development

Decisions, Forecasting, Optimization, And Causality

Use AI predictions to compare choices, plan ahead and ask what caused a result.

About This Course

Introduction

What will happen? What should we do? What would change because we acted? Those are three different questions. Prediction, decision-making, and causal analysis address them in different ways. A model can generate an explanation or proposal without resolving any of them. this course follows the steps between an output and a defensible action.

We will use small inventory, customer-work, and AI-adoption examples to explore probability, expected value, thresholds, forecasts, constraints, optimization, counterfactuals, confounding, experiments, and causal assumptions. You can calculate the examples by hand and see which evidence each question needs.

Revisit classification metrics in How Machine Learning Learns, evaluation in Post-Training, Adaptation, And Evaluations, and units and cost boundaries in AI Infrastructure Economics. Work through decisions, then causality, then forecasting and optimization. Your decision memo should state an action rule, a causal question where relevant, and how you would test whether the action creates value.

Planned Lessons

  1. Four Questions That Need Different AnswersSeparate describing what happened, predicting what comes next, explaining causes and choosing actions.
  2. Probability And Expected ValueCompare choices by combining possible outcomes with how likely each one is.
  3. Decision Thresholds Follow ConsequencesChoose when to act on an AI prediction using the consequences of a mistake.
  4. Base Rates, Bayes, And CalibrationUse how common something is, along with new evidence, to judge how likely it may be.
  5. Interventions And CounterfactualsAsk what would change if an action were taken, compared with leaving things as they are.
  6. Confounding And SelectionSpot other factors that could explain an apparent link between an action and a result.
  7. A Task-Mix Example That Reverses The StorySee how a change in the mix of tasks can make the same results tell a different story.
  8. Causal Graphs, Identification, And AssumptionsDraw the possible causes behind a result and state what must be assumed to make a claim.
  9. Randomized Experiments And The Assignment UnitSee how assigning people or groups to an experiment affects what its results can show.
  10. Experimental Analysis And Misleading SignificanceInterpret experiment results using uncertainty and the size of the real-world benefit.
  11. Observational Causal Methods And Their LimitsExplore when existing records can help answer a question about cause and effect.
  12. Forecasts Have Horizons And BaselinesJudge a prediction against its time period and a simple alternative way to forecast.
  13. Distributions, Scenarios, And Asymmetric LossPlan around several possible outcomes and the different costs of being wrong.
  14. Optimization: Objectives, Choices, And ConstraintsTurn a planning problem into a goal, choices and limits that a computer can work with.
  15. Sequential Decisions, Bandits, And Reinforcement LearningCompare ways to improve repeated choices using the feedback they produce.
  16. Human Authority And A Decision RecordRecord who decides, what evidence they used and who is responsible for the result.