We build a computational model of how humans actively infer hidden rules by doing experiments. The basic principles behind the model is that, even if the rule is deterministic, the learner considers a broader space of fuzzy probabilistic rules, which it represents in natural language, and updates its hypotheses online after each experiment according to approximately Bayesian principles. In the same framework we also model experiment design according to information-theoretic criteria. We find that the combination of these three principles — explicit hypotheses, probabilistic rules, and online updates — can explain human performance on a Zendo-style task, and that removing any of these components leaves the model unable to account for the data.
Expert Commentary: Understanding Human Inference of Hidden Rules
In this article, the authors present a computational model that aims to explain how humans actively infer hidden rules by conducting experiments. The key principles that underlie this model include the consideration of a broader space of fuzzy probabilistic rules, representation of these rules in natural language, and the updating of hypotheses after each experiment using approximately Bayesian principles.
The multi-disciplinary nature of the concepts discussed in this content is noteworthy. The model presented here combines elements of psychology, linguistics, and information theory to provide insights into human performance on a Zendo-style task.
The Role of Explicit Hypotheses
One crucial aspect of the proposed model is the inclusion of explicit hypotheses. By incorporating this element, the learner becomes more capable of actively and consciously formulating expectations about the hidden rules governing a particular task. This aligns with our understanding of human cognitive processes, where individuals tend to generate hypotheses to make sense of their environment.
Probabilistic Rules and Fuzzy Spaces
Another essential aspect explored in this model is the consideration of a broader space of probabilistic rules. While the underlying rule may be deterministic, allowing for probabilistic variations enables the learner to capture the inherent uncertainty present in many real-world scenarios. By representing these fuzzy probabilistic rules, the model captures the essence of human cognition, which often deals with imperfect information and varying degrees of certainty.
Online Updates and Bayesian Principles
The model proposed in this article also emphasizes the importance of online updates based on Bayesian principles. By continuously revising hypotheses after each experiment or new piece of information, the learner can refine their understanding and improve their performance over time. This iterative process mirrors human learning, where individuals update their beliefs and expectations as they acquire new evidence.
Overall, the combination of explicit hypotheses, probabilistic rules, and online updates provides a comprehensive framework for understanding human inference of hidden rules. Removing any of these components from the model would result in an inability to account for the data, highlighting their interdependence in explaining human performance on tasks such as the Zendo-style task.
This research serves as a valuable contribution to the field, bridging various disciplines to shed light on the intricate processes involved in human inference. By incorporating psychological, linguistic, and information-theoretic perspectives, this model provides a solid foundation for future studies exploring similar phenomena and further advancing our understanding of human cognition.