The conditional independence assumption in neurosymbolic learning models biases them towards deterministic solutions, hindering their ability to properly express uncertainty over multiple valid options.
