Making Disagreement Explicit: Lessons from the Final Stage of an Adversarial Collaboration

Our project brought together fans of representations, who believe that an important form of explanation in cognitive neuroscience is representational, and sceptics, who believe that explanation must first and foremost be mechanistic—and that, on closer inspection (Kohár 2023), mechanistic and representational explanation do not seem to mix easily.

Over the last two years, we have extensively discussed the roles of mechanistic, computational, and representational explanation in cognitive neuroscience. We identified three case studies and spent considerable time discussing them with scientists and among ourselves to fully understand their content and implications. We held three larger workshops, two of them combined with fellow meetings that brought us together with experts on the core topics of our project. And we met online almost every other week to plan, discuss, and understand. Along the way, individual members of the project started drafting articles, giving talks, and working on their PhD theses.

As our adversarial collaboration approaches its final stage, we want to know: Was the adversarial spirit of our collaboration fruitful? Have we changed our minds over these two years? What have we actually learned—and where do we still disagree?

Answering these questions turned out to be less straightforward than we had expected. Disagreements are not always explicit. Even after working together on the same questions for an extended period, it remained unclear exactly where our views diverge, which arguments are responsible for the divergence, and what would need to be established to resolve it. We therefore developed a procedure for making these remaining disagreements explicit.

Step 1: Taking stock individually

Each participant prepared a slide addressing two basic questions: What have I learned? and Which questions remain open for me? More specifically, we asked ourselves: What is my current view on the compatibility challenge? Has it changed? What were the most important takeaways from the case studies? Are there any new open questions?

This allowed us to document individual conclusions as well as uncertainties that had survived—or emerged from—the collaboration.

Step 2: Identifying shared open questions

In an online meeting, we presented and discussed our slides. We then collected questions that remained open for several participants and that seemed particularly important to the central issues of the project.

Step 3: Discussing potential disagreements adversarially

In subsequent meetings, we began discussing these questions one by one. We focused especially on questions that were central to the project and on which we suspected that the two sides of our collaboration held opposing views. Rather than merely exchanging positions, the aim was to determine as precisely as possible whether we disagreed, where the disagreement lay, and why.

An important part of this procedure was to record the online discussion and use AI to transform the recording into a dialogue-style reconstruction of the exchange between the two sides of our adversarial collaboration. The aim was to remain as close as possible to the original discussion while restructuring it enough to make the core lines of argument more visible. This created a record not only of what our disagreement ultimately turns out to be, but also of how we discovered it.

We are currently still discussing the first of our open questions. But this first discussion has already proved surprisingly informative. [See the dialogue on our first question: How is representational content explanatory, and based on what notion of explanatory relevance?]

The reconstructed dialogue makes it possible to trace how an initially unclear difference in views gradually becomes explicit: which questions reveal the disagreement, which arguments lead us to recognize it, at which points our reasoning diverges, and which further questions would need to be answered in order to resolve the disagreement.

Step 4: Identifying the questions behind the disagreement

This leads to a final step. Once a disagreement has been reconstructed in dialogue form, we can use that reconstruction to identify the more fundamental questions at its core.

Where exactly does the argument branch? Which premises does one side accept and the other reject? Are we using key concepts differently? Are there empirical questions that would need to be settled, or is the disagreement ultimately conceptual? And, most importantly: What would we need to find out or agree on in order to resolve the disagreement?

The result of the procedure is therefore not simply a list of things we agree and disagree about. It is a progressively more precise map of the disagreement: from individual open questions, through adversarial discussion, to the arguments underlying opposing positions and, finally, to the more fundamental questions on which those arguments depend.

In this sense, the procedure does more than document the results of an adversarial collaboration. It documents the process by which disagreement itself becomes an identifiable—and potentially tractable—result.