Abstract This article introduces the Learning Ising Attitude Model (LIAM), which explains how attitude networks reduce entropy—the dissonance within an attitude. The LIAM extends the network theory of attitudes by adding Hebbian learning, thereby aligning it with classic neural network models like the Hopfield network and the Boltzmann machine. In the original theory, attention and thought temporarily reduce entropy in balanced attitude networks, but it remained unclear why such networks are balanced in the first place. We demonstrate that higher attention, when paired with Hebbian learning, causes the system to evolve toward balanced network structures that effectively reduce entropy. In addition, we model feedback between attitudinal instability and attention, incorporate the learning of dispositional tendencies of attitude elements, and illustrate how attitude change can emerge from the model. Finally, we discuss how the LIAM contributes to the unification of attitude research, its relation to previous connectionist models, and directions for future work.
Smal et al. (Mon,) studied this question.