Description
The ongoing environmental risks put urgency in the use and development of AI tools to support and improve ecosystem management. Still, rushing the development of such a tool may unintentionally cause more harm if not done properly. We discuss two challenges that should be carefully considered when planning a fully functional EcoWeaver model.
First, training AI systems risks incorporating biases to the system, potentially perpetuating them. E.O. Wilson shifting perspective about kin selection as “key” for the evolution of altruism to later suggest that the theory has “fallen” is one of many cautionary tales about how evidence can change theoretical perspectives. Despite Wilson’s latter reassessment of the relative value of group and kin selection, initial ideas inspired by him provided fertile ground for significant research under the assumption that group selection was wrong and kin selection was right. By dismissing a theory, research in those years disregarded certain causal paths, which were not explored or tested.
Second, the simplicity of an idea can make it powerful, but if misunderstood or misused it can generate wrong information that could also bias a knowledge graph. An example is the general representation of “A causes B” as 'A -> B'. The arrow visualization of causality has been popularized by structural equation models (SEM), linking visual representations, mathematical formalization, and statistical analyses. But the simplicity hides conventions and assumptions that may not always be adequate, like the assumption that A and B are numeric or that the statistical representation implies a linear model. Inadequate uses of the arrow notation risk misapplication of the tool, potentially confusing interpretations during the research, and further complexities during synthesis processes.
This work will present ongoing ideas and discussions in the EcoWeaver team about how to address these challenges, looking to explore and expand the debate about them.
| Topic | Topic 4: Can knowledge graphs and AI-based synthesis improve ecological synthesis? |
|---|