A Probabilistic Hypotheses Generation

Can we make a probabilistic hypothesis generation? So, to improve the current hypotheses generation that tend to frame rules as deterministic binaries (e.g., "If she smiles back, then she is Interested"), I think we could try to generate hypotheses that explicitly estimate a probability distribution (e.g., "If she smiles back, there is a ~75% likelihood that she is Interested and ~25% chance that she is Scared"). I think this can bridge the gap between rigid theoretical definitions (from the literatures, from known facts) and the noisy, subjective reality of empirical examples.