Sociotechnical explanations can contribute to individual failure investigations and even normative critiques of an entire system, in the event that clear descriptions of causes and conditions of failure can be specified. But in many failed systems, what has happened is complicated. At present, sociotechnical systems approaches in organizational studies, engineering, and philosophy do not meet standards of clarity that are required of explanation. Further, sociotechnical system approaches so far have not been used to predict instances of failure, though it would be highly desirable if we could develop such predictions. Complications with explanation and the dim prospects for prediction lead us in the direction of new computational technologies, but these are not without risks. Progress in understanding sociotechnical systems means, in effect, that explanations and predictions would become useful for “unenhanced” human understanding, not just for computational models of sociotechnical behavior. One byproduct of this better understanding could be guidance on when to hold certain parts of AI sociotechnical systems accountable—namely, when the algorithms by which they operate could be held accountable for damages that they cause.