"Model-less" Modeling?

One of the themes that seems to be emerging in some circles of finance is the notion of "model-free" analysis and reasoning. Here is a comment that I recently posted in response to such thinking in a popular public forum:

To all, the case for modeling probabilities and inferential statistics is a bit harder to dismiss than some are presupposing in this forum. Now, I do concede that researchers have many challenges to deal with when validating data and models. Moreover, the theoretical relationships between dependent and independent variables (including proxy measures) are often misspecified and/or interpreted by researchers to the point where some research is outright misleading (especially in the social sciences). Nevertheless, inferential modeling of complex pheonomena cannot be simply dismissed as incredible via declaratory argument and assertions. There is a tremendous burden of proof that is assumed when one takes the path of falsifying probability theory and inferential statistics as a discipline. Thus, I fear that the notion of "model-less" reasoning is problematic as an approach to understanding what is happening in this place called "reality" around us. For the record, I am not as pessimistic about the validity and reliability of such methods as some may be here and in society. My advice to all is "be prepared" to defend your evidence if your goal is to falsify probability theory as a tool for understanding and evaluating the risks (and dangers) that seem to prevail in this universe. Thanks for the opportunity to comment...