Every day, your brain performs millions of “what if” calculations without you even noticing. What if I had taken the other route to work? What if I hadn’t said that in the meeting? What if the ball bounces differently than expected? This capacity for counterfactual reasoning, imagining alternative realities that never actually occurred, represents one of the most sophisticated computational achievements of biological intelligence.

A groundbreaking new study published in Nature Human Behaviour by Ramadan, Tang, Watters, and Jazayeri has shed new light on why humans rely on these mentally expensive “what if” simulations, revealing computational constraints that force our brains into remarkably clever problem-solving strategies. Their findings illuminate human cognition and change how we understand intelligence itself.

The Computational Mystery: Why Do We Think in “What Ifs”?

From a purely computational standpoint, counterfactual reasoning seems inefficient. When facing complex decisions, optimal algorithms should simply compute the joint probability of all possible outcomes and pick the best option. So why do humans constantly engage in the seemingly wasteful exercise of imagining alternatives?

The answer, as Ramadan and colleagues discovered, lies in the fundamental constraints that shape how our brains process information. Using an ingenious H-maze task where participants had to track an invisible ball through branching pathways, they uncovered three critical computational bottlenecks that force human cognition into hierarchical and counterfactual strategies:

1. Parallel Processing Bottleneck: Our brains cannot track all possible trajectories simultaneously. We must break complex problems into sequential, hierarchical steps.

2. Counterfactual Processing Noise: When we engage in “what if” thinking, our working memory introduces noise that degrades the fidelity of these mental simulations.

3. Rational Resource Allocation: Humans adaptively adjust their reliance on counterfactuals based on how much these mental simulations cost them.

Very Clever Use of Recurrent Neural Networks in Modeling Features of the Human Mind

The research reveals profound insights about intelligence itself. When Ramadan et al. created artificial neural networks and subjected them to the same computational constraints humans face, something remarkable happened: only the networks constrained by all three bottlenecks reproduced human-like behavior.

This finding demonstrates the power of using recurrent neural networks to model human cognition. By constraining artificial networks with the same limitations that shape human thinking, Ramadan et al. created systems that behave remarkably like people. The key insight is that RNNs can capture mental processes like hierarchical and counterfactual reasoning when they face the same computational bottlenecks humans do.

Neural Architecture of Counterfactual Reasoning

The neural implementation of counterfactual reasoning tells a more complex story beyond frontal control. Van Hoeck and colleagues’ landmark fMRI study revealed that counterfactual thinking engages a distributed network that hijacks the brain’s episodic memory system.

When participants imagined “upward counterfactuals” (better outcomes for negative past events), their brains activated the same core memory network used for remembering the past and imagining the future: hippocampus, posterior cingulate, inferior parietal lobule, lateral temporal cortices, and medial prefrontal cortex.

What makes counterfactual reasoning computationally expensive becomes clear in this neural architecture. Counterfactual thinking recruited these memory regions more extensively than episodic past or future thinking, and additionally engaged bilateral inferior parietal lobe and posterior medial frontal cortex.

The extra brain activity reflects just how demanding this kind of mental juggling really is: counterfactual reasoning requires simultaneously maintaining factual and contrafactual representations while actively inhibiting the dominant factual reality.

The brain has evolved specialized circuitry for tracking “what might have been.” Boorman and colleagues discovered that lateral frontopolar cortex, dorsomedial frontal cortex, and posteromedial cortex form a dedicated network for encoding counterfactual choice values: tracking not just what happened, but whether alternative options might be worth choosing in the future.

This network operates in parallel to the ventromedial prefrontal system that tracks the value of chosen options, suggesting that the brain maintains separate computational channels for factual and counterfactual value processing.

Perhaps most remarkably, recent work has shown that counterfactual information fundamentally transforms how the brain codes value itself. When counterfactual outcomes are available, medial prefrontal and cingulate cortex shift from absolute to relative value coding.

Think of it this way: losing $10 feels terrible if you could have won $50, but feels great if you could have lost $100. The same neural outcome is processed as positive in a loss context (absence of punishment) but negative in a gain context (absence of reward).

This neural flexibility mirrors the adaptive computational strategies revealed in behavioral studies: the brain dynamically reconfigures its representational schemes based on available information and processing constraints.

These findings illuminate why counterfactual reasoning is both computationally expensive and evolutionarily preserved. The enhanced neural demands reflect genuine computational costs: maintaining multiple alternative representations, binding novel scenario elements, and managing conflict between factual and counterfactual worlds. Yet this system enables the kind of flexible, context-sensitive reasoning that allows humans to learn from paths not taken and adapt behavior based on imagined alternatives.

The Bounded Rationality Renaissance

These discoveries are part of a broader renaissance in understanding bounded rationality, the idea that intelligent behavior emerges not from perfect optimization, but from smart adaptations to computational limitations.

Herbert Simon’s revolutionary concept of bounded rationality challenged the assumptions of perfect rationality in classical economic theory, proposing instead that individuals “satisfice” (seeking good enough solutions rather than optimal ones) due to limitations in computation, time, information, and cognitive resources.

Simon’s work recognized that “perfectly rational decisions are often not feasible in practice because of the intractability of natural decision problems and the finite computational resources available for making them.” This insight has profound implications for both understanding human cognition and designing artificial intelligence systems.

The Bigger Picture

The Ramadan study reveals something profound: the cognitive strategies we think of as distinct (hierarchical reasoning, counterfactual thinking, simple optimization) actually lie along a continuum. Human intelligence dynamically shifts between these approaches based on available mental resources and task demands.

This has implications beyond neuroscience. If counterfactual reasoning emerges from computational constraints rather than being hardwired, it suggests these “what if” processes might be fundamental to any sufficiently complex intelligence, biological or artificial.

Clinical Frontiers: When Counterfactuals Break Down

From a clinical perspective, this research offers new windows into psychiatric and neurological conditions. Counterfactual reasoning depends on integrative networks for affective processing, mental simulation, and cognitive control. These are systems that are systematically altered in psychiatric illness and neurological disease.

Consider a patient with OCD who gets trapped in endless loops of “what if I didn’t check the door?” or someone with depression whose counterfactual thinking spirals into “if only I were different, everything would be better.” Understanding the computational basis of these patterns could lead to more targeted therapeutic approaches.

Patients with schizophrenia show specific deficits in counterfactual reasoning when complex non-factual elements are needed to understand social environments. By mapping how these computational processes break down, we’re gaining new tools for both diagnosis and treatment.

The Bottom Line: Constraints as Features

The story of counterfactual reasoning is a story about the power of constraints. What initially appears to be a computational limitation (our inability to process all information in parallel) turns out to be the very foundation of human cognitive flexibility.

The human brain’s “what if” engine represents an elegant solution that emerges from the interplay between computational constraints and adaptive intelligence. As we stand on the brink of artificial general intelligence, perhaps the secret lies not in building systems that can process everything at once, but systems that can gracefully adapt to the fundamental constraints that shape all intelligence.

The future of AI may not lie in eliminating human limitations, but in understanding why those limitations exist and what remarkable capabilities they make possible.

This convergence of neuroscience, cognitive science, and AI represents a fundamental shift in how we understand intelligence. Rather than seeing computational constraints as problems to solve, we’re beginning to recognize them as the very features that make flexible, adaptive intelligence possible. The brain’s “what if” engine may be a blueprint for the next generation of truly intelligent machines.

The next time you wonder what might have been, remember: that question may be the very core of what makes you human.

Bibliography

Boorman, E. D., Behrens, T. E., & Rushworth, M. F. (2011). Counterfactual choice and learning in a neural network centered on human lateral frontopolar cortex. PLoS Biology, 9(6), e1001093.

Pischedda, D., Palminteri, S., & Coricelli, G. (2020). The effect of counterfactual information on outcome value coding in medial prefrontal and cingulate cortex: From an absolute to a relative neural code. Journal of Neuroscience, 40(16), 3268-3277.

Ramadan, M., Tang, C., Watters, N., & Jazayeri, M. (2025). Computational basis of hierarchical and counterfactual information processing. Nature Human Behaviour. doi:10.1038/s41562-025-02232-3.

Simon, H. A. (1955). A behavioral model of rational choice. Quarterly Journal of Economics, 69(1), 99-118.

Van Hoeck, N., Ma, N., Ampe, L., Baetens, K., Vandekerckhove, M., & Van Overwalle, F. (2013). Counterfactual thinking: An fMRI study on changing the past for a better future. Social Cognitive and Affective Neuroscience, 8(5), 556-564.

Van Hoeck, N., Watson, P. D., & Barbey, A. K. (2015). Cognitive neuroscience of human counterfactual reasoning. Frontiers in Human Neuroscience, 9, 420.

Zador, A., Escola, S., Richards, B., et al. (2023). Catalyzing next-generation Artificial Intelligence through NeuroAI. Nature Communications, 14, 1597.

The study by Mackenzie et al. (2025, bioRxiv) represents a particularly clever approach to understanding human thalamic function. Rather than relying on correlational neuroimaging, the researchers capitalized on an unintended consequence of focused ultrasound thalamotomy for essential tremor. When post-surgical vasogenic edema extended beyond the intended motor target into cognitive thalamic regions, it created a rare opportunity to assess the causal contribution of different thalamic nuclei to decision-making behavior.

What makes this approach so powerful is the precision it affords. Patients served as their own controls, tested before and after surgery on a sophisticated decision-making paradigm that probes the exploration-exploitation trade-off under uncertainty.

Computational Dissection of Behavioral Changes

Using the restless four-armed bandit task, which requires continuous adaptation to changing reward contingencies, the researchers could probe multiple aspects of decision-making simultaneously. The task’s Gaussian random walk structure creates ongoing uncertainty, forcing participants to balance between exploiting currently favored options and exploring alternatives that might yield better outcomes.

The key innovation came from their computational modeling approach. Rather than simply observing that patients made more “stay” choices post-surgery, the authors fitted multiple reinforcement learning model variants to decompose the underlying decision processes. This revealed that the behavioral shift was best captured by a Bayesian learning model with increased reward sensitivity (β) but eliminated exploration bonus—suggesting that patients weren’t simply perseverating, but had fundamentally altered confidence in their value estimates.

Most strikingly, when using their winning model to classify choice types, the researchers found a dramatic reduction in directed exploration—the strategic sampling of uncertain options to gain information (Mackenzie et al., 2025). This wasn’t random exploration or simple indecision, but the specific loss of information-seeking behavior that would normally help resolve uncertainty about option values.

Anatomical Precision and Circuit Specificity

The neuroimaging analysis provided crucial anatomical specificity. The degree of behavioral change correlated specifically with edema extension into the lateral (parvocellular) mediodorsal nucleus—not other thalamic regions including the intended surgical target (VIM). This specificity is important given the known functional subdivisions within MD:

• Lateral MD → DLPFC/Frontal Pole: Dense reciprocal connectivity with areas involved in cognitive control and abstract rule formation
• Medial MD → OFC/vmPFC: Connections with valuation and reward-processing regions

The functional connectivity analysis further supported this anatomical specificity. Using normative connectome data, individual patients’ behavioral changes could be predicted from the connectivity profile between their lesioned MD voxels and prefrontal cortex—but only for MD, not other thalamic nuclei.

Mechanistic Insights: From Confidence Calibration to Circuit Function

The computational framework reveals something more nuanced than simple “inflexibility.” The post-lesion behavioral profile suggests a specific breakdown in uncertainty representation—what we might call miscalibrated confidence. When MD-PFC communication is compromised, the system appears to default to high confidence in existing value representations, reducing sensitivity to contradictory information.

This aligns with emerging theoretical frameworks positioning the MD thalamus as a critical node in hierarchical inference, helping to coordinate distributed computations for flexible and efficient learning (Scott et al., 2024). The loss of directed exploration particularly supports this view, as this behavior specifically emerges when agents are uncertain about their value estimates and seek information to reduce that uncertainty.

Bridging Animal Models and Human Neuroscience

The convergence with our rodent findings is striking:

Animal Studies (MD inactivation/optogenetics):

• Reduced flexibility in volatile environments
• Animals fail to revise beliefs when contingencies change
• Inflated certainty in action values
• Deficit specific to belief updating, not initial learning

Human Study (accidental MD lesions):

• Reduced switching in uncertain environments
• Patients fail to explore when exploration would be beneficial
• Increased confidence in value estimates
• Preserved basic learning ability

This cross-species convergence suggests we’ve identified a fundamental computational principle rather than a species-specific curiosity.

Therapeutic Implications: Beyond Motor Applications

The findings suggest intriguing therapeutic possibilities, particularly for disorders characterized by altered belief updating and confidence calibration. The demonstration that MD disruption leads to overconfident exploitation with reduced information-seeking offers a compelling framework for understanding psychiatric conditions where belief revision goes awry.

Consider schizophrenia, where patients often exhibit pathological certainty in delusional beliefs despite contradictory evidence. The current findings suggest a potential mechanism: if MD-PFC circuits that normally regulate confidence in beliefs become dysregulated, patients might lose the capacity for adaptive doubt that would otherwise prompt belief revision. The specific loss of directed exploration observed here—the strategic sampling of information to resolve uncertainty—parallels the clinical observation that individuals with psychosis often fail to seek disconfirming evidence for their beliefs.

This connects to broader hypotheses about uncertainty processing in cognitive control. Rather than viewing delusions simply as “false beliefs,” they might reflect a fundamental breakdown in the brain’s ability to appropriately weight confidence in its own predictions. When the system becomes overconfident in existing representations (as seen post-thalamotomy), it loses the motivation to gather information that might challenge those representations—a hallmark of delusional thinking.

Broader Significance: Rethinking Thalamic Function

This work contributes to a fundamental reconceptualization of thalamic function – from simple relay station to active computational processor. The thalamus isn’t just routing information; it’s dynamically modulating the confidence and precision of cortical computations based on behavioral context.

Personal Reflection: When Theory Meets Unexpected Validation

For our lab, this study is quite awesome – when years of circuit-level investigation receive independent validation from an entirely different methodology. The fact that this confirmation came through a clinical study aimed at treating human suffering makes it even more meaningful.

It’s the rare convergence where theory and evidence transform each other: the theory gains human causal validation, while the evidence gains mechanistic understanding. Together, they point toward a future where we might not just understand the circuits of adaptive decision-making, but actively repair them when they break.

The patients in this study, seeking relief from debilitating tremor, graciously contributed to our understanding of one of the brain’s most fundamental computations: how to balance confidence with curiosity. Their experience shows us what happens when certainty becomes a cage – when we lose the capacity to doubt ourselves when doubt would serve us best.

The work discussed builds on extensive research into thalamocortical circuits and decision-making, offering new insights into the neural mechanisms underlying adaptive behavior and potential therapeutic applications for disorders of motivation and cognitive flexibility.

References:

• Mackenzie, G., et al. (2025). Focused ultrasound neuromodulation of mediodorsal thalamus disrupts decision flexibility during reward learning. bioRxiv.
• Scott, D.N., Mukherjee, A., Nassar, M.R., & Halassa, M.M. (2024). Thalamocortical architectures for flexible cognition and efficient learning. Trends in Cognitive Sciences, 28(7), 639-652.

This collaborative work is driven by a burning question we have: Why is the brain wired this way? Why does the thalamus, nestled deep in the forebrain and reciprocally connected to cortex, play such a critical role in cognition? Our empirical work over the past decade has dissected thalamocortical dynamics in behaving animals, and our computational work including critical collaborations have helped us formalize these findings into testable frameworks. Sage’s new paper is a natural extension of this synergy—and with empirical support from our lab (spearheaded by postdoc Arghya Mukherjee), it opens new doors for exploration.

Key Advances in the New Model

1. The MD Thalamus as a Dynamic Router
   The study presents the MD thalamus not just as a passive relay, but as an active switchboard that flexibly gates information to the PFC based on task demands. This aligns with our lab’s empirical observations that thalamic neurons selectively amplify sensory inputs or internal signals depending on behavioral context.
2. Task-Dependent Cortical Prioritization
   The model captures how the MD thalamus biases PFC representations—for example, emphasizing sensory cues during perceptual decisions versus memory traces during recall. This mirrors findings from our 2018 (Rikhye, Gilra & Halassa) and 2022 (Hummos et al.) models, where thalamic input helped partition PFC activity to avoid interference across tasks.
3. Bridging Theory and Experiment
   Crucially, the model’s predictions were tested with in vivo data from our lab, reinforcing its biological plausibility. This back-and-forth between modeling and physiology is a hallmark of our approach, exemplified in Wei-Long Zheng’s 2024 study, where a thalamocortical RNN outperformed standard models in rapid inference tasks.

How This Drives Our Empirical Work Forward

1. New Experiments to Test Gating Mechanisms
   The model proposes specific thalamocortical connectivity rules for information routing. We’re now designing experiments to probe these mechanisms using optogenetics, electrophysiology, and imaging—asking how MD neurons dynamically recruit PFC microcircuits during task switching.
2. Linking to Schizophrenia-Relevant Dysfunction
   Disrupted thalamocortical gating is implicated in schizophrenia. By refining Sage’s model with disease-relevant perturbations (e.g., thalamic silencing), we aim to pinpoint how maladaptive routing contributes to cognitive inflexibility.
3. The Next Generation of NeuroAI Models
   Just as Wei-Long’s hybrid RNN incorporated biological constraints (e.g., thalamic reticular inhibition), future iterations of Sage’s model could integrate our latest empirical data—creating a virtuous cycle between theory and experiment.

The Bigger Picture: A Decade of Thalamocortical ModelingCognitive flexibility

This paper is the latest in a line of collaborative efforts to formalize MD-PFC interactions:

• Rikhye, Gilra & Halassa (2018): Showed thalamus mitigates “catastrophic forgetting” in PFC.
• Hummos et al. (2022): Derived a cortico-thalamic learning rule that compresses task context.
• Zheng et al. (2024): Demonstrated thalamus enables rapid inference and multi-task performance.
• Zhang, X. et al. (2025): extends this to hierarchical reasoning and handling multiple forms of uncertainty.

Together, these studies underscore the thalamus’s role as a locus of cognitive flexibility—a theme Sage’s work now extends with elegant mechanistic detail.

Looking Ahead

As NeuroAI gains momentum (evidenced by the 2024 Physics Nobel for foundational neural network work), our lab remains committed to grounding computational advances in biological reality. Sage’s model not only validates our empirical findings but also charts a course for future work—one where theory and experiment co-evolve to unravel the thalamus’s secrets.

For those who missed it, revisit our blog on Wei-Long Zheng’s paper here, and stay tuned as we put these models to the test!

Reference:

Zhang, X., Mukherjee, A., Halassa, M. M., & Chen, Z. S. (2025). Mediodorsal thalamus regulates task uncertainty to enable cognitive flexibility. Nature communications, 16(1), 2640. https://doi.org/10.1038/s41467-025-58011-1