The agent-based model is the principal scientific instrument of generative social science. Typically, we design completed agents-fully endowed with rules and parameters-to grow macroscopic target patterns from the bottom up. Inverse generative science (iGSS) stands this approach on its head: Rather than handcrafting completed agents to grow a target-the forward problem-we start with the macro-target and evolve micro-agents that generate it, stipulating only primitive agent-rule constituents and permissible combinators. Rather than specific agents as designed inputs, we are interested in agents-indeed, families of agents-as evolved outputs. This is the backward problem and tools from Evolutionary Computing can help us solve it. In this overarching essay of the current JASSS Special Section, Part 1 discusses the motivation for iGSS. Part 2 discusses its goals, as distinct from other approaches. Part 3 discusses how to do it concretely, previewing the five iGSS applications that follow. Part 4 discusses several foundational issues for agent-based modeling and economics. Part 5 proposes a central future application of iGSS: to evolve explicit formal alternatives to the Rational Actor, with Agent_Zero as one possible point of evolutionary departure. Conclusions and future research directions are offered in Part 6. Looking 'backward to the future,' I also include, as Appendices, a pair of 1992 memoranda to the then President of the Santa Fe Institute on the forward (growing artificial societies from the bottom up) and backward (iGSS) problems.