In complex systems characterized by self-organized criticality, phase transitions emerge, impacting macroscopic system properties. These phase transitions exhibit behaviors that are independent of the scale at which they occur and can be classified as either desirable or undesirable. In this work, the Internet of Things (IoT) is utilized as a mechanism to intervene in these systems by avoiding or desiring the phase transition in this type of systems through the proposed intervention framework based on the Deep Q-Network algorithm. This facilitates the intervention of IoT devices in a decentralized manner using only locally sensed information. This approach’s efficacy is validated through a case study of vehicular traffic, a complex system with self-organized criticality, in which traffic lights act as an intervention element. The findings indicate that it is feasible to modify the system’s phase transition, thereby enhancing the metrics of average speed, average vehicle flow, and total vehicles flow per intersection in comparison to a conventional traffic light management policy.