Table 2. Revised fitness function.

Lists of the multiple objective/criterion of the fitness function applied to each simulated group. refPCA is 1, 0 and −1 for STs, IGs, GTs respectively. This function is combined with NSGA-II to retrieve parameter values that best reproduce the experimental results. It results in a Pareto front of parameters from which we select by hand the solution that is consistent (no agent being misclassified) and that best visually fits the observed data (between engaging with the lever versus engaging with the magazine).

Objective	Formula
Best fit magazine engagement	$\min \left({\sum }_{s_{i\in \text{sessions}}}{\left(p_{\mathit{Sim}}^{s_i}\right(\mathit{engM}\mid \mathit{Group})-p_{\mathit{Obs}}^{s_i}(\mathit{engM}\mid \mathit{Group}\left)\right)}^2\right)$
Best fit lever engagement	$\min \left({\sum }_{s_{i\in \text{sessions}}}{\left(p_{\mathit{Sim}}^{s_i}\right(\mathit{engL}\mid \mathit{Group})-p_{\mathit{Obs}}^{s_i}(\mathit{engL}\mid \mathit{Group}\left)\right)}^2\right)$
Penalize parameters that lead to misclassification	min(∑ai∈animals ∣ref PCA (Group) − IndexScoreai (Group)∣)