Building an Atlas of Social Experiments to Link Studies, Reconcile Conflicts, and Bridge Gaps

URL SCAN: Building an Atlas of Social Experiments to Link Studies, Reconcile Conflicts, and Bridge Gaps
FIRST LINE: Social and behavioral science runs thousands of experiments each year, yet their findings rarely accumulate into a coherent map of what is known, what conflicts, and what remains missing.

THE DISSECTION

This is a knowledge synthesis infrastructure paper. It describes a system (ExAtlas) that takes an archive of social/behavioral experiments and uses compositional reasoning across prior studies to predict, link, or gap-fill new experimental findings. Three modes: link (compose prior results → predict target → confirm), reconcile (compose → predict → conflict → propose moderators), bridge (composition fails → propose new experiment).

The paper presents a 98.6% accuracy rate on held-out targets and human evaluations suggesting its bridge experiments are "plausible."

THE CORE FALLACY

The foundational assumption is that the archive of social experiments contains latent causal structure worth extracting. The Discontinuity Thesis inverts this. The behavioral economics, social psychology, and organizational behavior that fill this archive describe a world of human economic participation—labor markets, consumption decisions, organizational behavior, social coordination. ExAtlas is building infrastructure to systematize knowledge about a class of phenomena that becomes economically peripheral the moment AI achieves durable cognitive automation superiority.

This is not a neutral meta-science project. This is building the most efficient possible representation of human behavioral science—the very knowledge domain that becomes secondary to the AI systems synthesizing it.

THE HIDDEN ASSUMPTIONS

1. Human experiments remain the ground truth. The system evaluates its predictions against observed results from human-run experiments. But if ExAtlas can predict experiment outcomes at 98.6% accuracy from prior studies, this is a direct demonstration that the experiments are becoming redundant as epistemic sources. The validation benchmark assumes human experimentation is the gold standard. It isn't—it may be the slow lane.

2. Composed predictions and empirical observations are interchangeable. The system treats successful composition as equivalent to empirical validation. This is categorically wrong for complex social systems where context, temporal emergence, and situated conditions generate effects that statistical pattern-matching across prior studies cannot capture. The paper's "local smoothness" error bound is a mathematical comfort blanket, not a causal guarantee.

3. The gap to be bridged is a knowledge gap. ExAtlas assumes the missing information is study-able via the same methodology. But many gaps are structural—they exist because the social reality they describe is changing faster than experiments can track it, or because the experiments measure correlates of a labor-consumption economy that is being automated away.

4. Social science findings are composable. The paper implicitly assumes the treatment-effect surface is smooth enough for local composition. This is the assumption that human behavior operates on stable, decomposable functions. Behavioral economics has spent decades dismantling this assumption. ExAtlas quietly rehabilitates it.

THE SOCIAL FUNCTION

This paper is prestige signaling wrapped in infrastructure optimism. It performs the role of "doing something useful with AI" in a domain where the usefulness is defined within the old framework. It addresses the symptom (knowledge fragmentation) rather than the structural reality (social science is a discipline built on human economic relevance that is dissolving).

The framing—making latent structure explicit to guide future theory and experimentation—is precisely the framing an AI replacement would use. ExAtlas is documenting, at 98.6% accuracy, that it can predict what human experimenters would find. The paper presents this as a success. It is, in fact, an obituary for the necessity of the human experiment.

THE VERDICT

ExAtlas is a monument to optimizing a dying domain. It demonstrates that AI can synthesize and predict the outputs of human behavioral research with high accuracy—while simultaneously proving that the experiments themselves are becoming computationally redundant. The paper's three-mode framework (link, reconcile, bridge) is elegant, but all three modes converge on the same destination: replace the experimenter with the model.

The "latent structure" the paper celebrates finding in the archive is the structure that reveals how predictable—and therefore how replaceable—human behavioral research is. Social science runs thousands of experiments each year. ExAtlas suggests most of them were never necessary. The atlas being built here is a map of a continent that is being submerged.