Multi-modal multi-objective model-based genetic programming to find multiple diverse high-quality models

Multiple models are better than one: Traditional machine learning, including genetic programming, typically produces a single "best" model. However, this paper demonstrates that domain experts often prefer having multiple diverse, high-quality models to choose from, since the mathematically best model may overfit, contain unwanted errors on specific data points, or simply not align with expert judgment about what makes sense in practice.

Explainability requires interpretable model design: The research emphasizes that genetic programming is particularly valuable for explainable AI because it can learn small symbolic models—essentially mathematical expressions—that humans can actually understand and interpret, rather than black-box models that make accurate predictions but offer no insight into their reasoning.

Trade-offs matter in real-world applications: The paper highlights that different accuracy measures (like mean squared error) can lead to different conclusions about model quality. By explicitly searching for models that balance multiple objectives rather than optimizing for a single metric, the approach better reflects the complex trade-offs that matter in real-world decision-making.

Domain expert judgment should guide model selection: Rather than automatically accepting whatever model performs best mathematically, this work recognizes that domain experts need options to evaluate which model is most "plausible" and trustworthy for their specific context—a crucial consideration for responsible AI deployment where stakeholder buy-in matters.