Microsoft's new method turns black-box brain AI into readable theories

For the better part of a decade, large language models have been the most accurate tools for predicting how the human brain responds to language. Feed an LLM the same story a person hears inside an fMRI scanner, and the model's internal representations can predict the activity of individual patches of cortex with remarkable fidelity. But no one can read these models. They are vast, inscrutable masses of parameters that defy direct translation into interpretation. A model that forecasts brain activity might tell you a region responds to language, but it cannot say whether that response is to food, places, numbers, or something else entirely. As black-box models proliferate, the chasm between prediction and understanding has become one of the central puzzles in computational neuroscience.

Turning black boxes into testable theories

In a paper accepted by Nature Neuroscience, Microsoft Research scientists, working alongside colleagues at UC Berkeley, UCSF, and Columbia University, have proposed a way out of this explainability crisis. Their framework, generative causal testing (GCT), distills brain-prediction models into short, readable accounts of what each patch of cortex responds to, and then tests those accounts. An LLM writes new stories engineered to activate a specific brain area. Subjects listen to these stories in the scanner. If the explanation is correct, the targeted region lights up. The result is a method that translates uninterpretable predictive models back into the currency of science: concise hypotheses that can be confirmed or refuted in a follow-up experiment.

How GCT works

GCT proceeds in two steps: explanation, then verification. To generate an explanation, the method begins with a predictive model for a single voxel or region and identifies the short phrases that most strongly drive its predicted response. An LLM then summarizes those words into a concise verbal explanation, often a single phrase such as "food preparation" or "location names."

The crucial second stage closes the loop. To build trust in that explanation, GCT uses an LLM to write new stories in which each paragraph is carefully constructed to drive a brain region according to its explanation. Three subjects returned to the scanner to read these synthetic stories. If a region's activity to its "driving" paragraphs was significantly greater than to baseline text, the explanation passed a genuine causal test, not merely a correlational one.

Across all three subjects, the core approach held up: the synthetic stories reliably drove their target regions above baseline, confirming that GCT's short explanations capture something the cortex genuinely responds to. The explanations were also most trustworthy where the underlying brain-prediction models were strongest: the more stable the model, the more reliably its explanation could be confirmed in the scanner. With the method validated on regions whose selectivity was already known, the researchers turned GCT toward harder questions.

GCT also proved sharp enough to settle long-standing ambiguities. Three neighboring regions involved in processing places: the retrosplenial cortex (RSC), the parahippocampal place area (PPA), and the occipital place area (OPA), have often been treated as functionally similar. At first, stories written for one region also activated the others. But by generating differential stimuli, stories designed to switch one region on while keeping its neighbors quiet, GCT teased the three apart. For example, RSC responds more strongly to proper noun location names, like Tokyo or Connecticut, rather than general location. This is the kind of nuanced, region-specific theory that a raw predictive model cannot provide on its own.

Beyond known regions, the authors discovered new prefrontal "micro-regions." By scanning a grid of candidate locations and keeping only the most stable ones, GCT surfaced these previously unmapped regions tuned to remarkably specific concepts: one selective for dialogue between people (words like "said" or "told"), one for mentions of clock times ("one o'clock"), and one for numeric measurements ("50 feet"). These are distinctions nobody had gone looking for; they emerged because the method could propose a hypothesis and immediately test it.

Implications and looking forward

The significance of GCT reaches well beyond neuroscience. Researchers in many fields increasingly face the same dilemma: a model that predicts beautifully but explains nothing. GCT shows that a data-driven model need not be the end of inquiry. It can be distilled into a readable, experimentally testable theory, and that theory can be checked against reality by generating new experiments on demand.

For neuroscience specifically, GCT points toward a faster, more hypothesis-rich way of mapping the cortex: one where an AI system proposes what a brain region might encode and a closed-loop experiment confirms or rejects it within a single study. The same generate-and-verify philosophy could extend to other domains where powerful predictive models have outrun our ability to understand them. The broader lesson is hopeful: the rise of black-box models in science does not necessarily mean the retreat of human-readable theory. With the right framework, the two can advance together.

This work was a collaboration across Microsoft Research, UC Berkeley (Alex Huth, Bin Yu, Sihang Guo, and Aliyah Hsu), Columbia University (RJ Antonello, co-lead), and UCSF (Shailee Jain). The paper is published in Nature Neuroscience and the code is available on GitHub.