When designing laboratory experiments, researchers often treat each trial as an isolated event. An animal observes a stimulus, makes a choice, receives a reward, and the trial ends. However, biological brains do not simply reset to a baseline state between trials. Instead, our choices are deeply tethered to our immediate past. This phenomenon—where previous choices and outcomes systematically skew subsequent decisions—is known as choice history bias.
Rather than being a form of random noise or a simple behavioral flaw, recent computational and neural evidence suggests that choice history bias is a structured, deterministic feature of how the brain integrates information over time.
The Mechanics of Bias: Starting Point vs. Drift Rate
To understand why past choices color current ones, we must look at how the brain accumulates information. In standard computational frameworks like the Drift Diffusion Model (DDM), a decision is modeled as the accumulation of noisy sensory evidence over time. Evidence is added up until it crosses one of two boundaries, triggering a choice.
A choice history bias can alter this accumulation process in two fundamentally distinct ways:
- Starting Point Bias: Before any new evidence is presented, the initial baseline of the decision variable is shifted closer to one of the boundaries.
- Drift Bias: The baseline remains centered, but the rate at which incoming evidence is accumulated (the drift rate) is modulated, meaning the brain weights incoming evidence differently depending on its prior state.
Distinguishing between these two mechanisms is essential for understanding whether history acts as a temporary "cold start" strategy or a persistent constraint on information integration, a question framed by Tor Tarantola et al. in their study on how Prior preferences beneficially influence social and non-social learning.
Starting Point Bias Drift Bias
Boundary A Boundary A
| /•. | /•.
|• / •. | / •.
| \ / • Boundary | / • Boundary
| \/ |=====/
+-------------------> +------------------->
Initial baseline shifted Accumulation slope altered
The distinction carries distinct behavioral and physiological signatures. For instance, Jan Willem de Gee et al. investigated how neuromodulators affect this process in Catecholamines reduce choice history biases in perceptual decision making. They discovered that boosting catecholamines (such as noradrenaline and dopamine) causally reduces choice history bias by modulating the drift bias rather than the starting point. This adjustment ensures that incoming sensory data is processed more objectively, minimizing the impact of prior internal states.
Conversely, history biases can also act directly on the starting point. Diksha Gupta et al. demonstrated in Trial-history biases in evidence accumulation can give rise to apparent lapses in decision-making that the starting point of the decision variable can be adjusted trial-by-trial using an exponential filter of previous choices and outcomes. This starting-point shift leaves a clear temporal signature: history-driven errors occur with faster reaction times than correct trials because the accumulator starts closer to the incorrect boundary and trips it prematurely.
Reinterpreting "Lapses" as Structured History
When an expert human or animal subject makes a glaring mistake on an incredibly easy trial, researchers traditionally categorize it as a "lapse"—a moment of random inattention or motor failure. However, dynamic behavioral modeling suggests these lapses are often not random noise at all.
Nicholas A. Roy et al. developed an efficient dynamic regression model in Efficient inference for time-varying behavior during learning to track trial-by-trial strategies. They found that apparent lapses are frequently the predictable consequence of subjects over-weighting recent trial history.
As Diksha Gupta et al. note, if the brain operates under the internal assumption that the environment is highly dynamic and constantly changing, it will update its starting point so aggressively that the bias can completely override clear, easy sensory evidence. What looks like a careless error is actually the deterministic byproduct of a history-dependent strategy running in the background.
Temporal Weighting and the Decay of Evidence
Why does the brain carry this historical baggage? The answer lies in how neural circuits naturally manage time. Decisions require integrating information across multiple timescales—both within a single trial and across consecutive trials.
Within a single trial, evidence integration is rarely perfectly uniform. Waitsang Keung et al. showed in A divisive model of evidence accumulation explains uneven weighting of evidence over time that human decision-making often deviates from ideal integration due to divisive normalization, a canonical neural computation that scales responses but can cause uneven temporal weighting. This can manifest as a primacy effect, where early information dominates the final choice. Lucas Pinto et al. observed similar primacy effects in mice navigating virtual reality in An Accumulation-of-Evidence Task Using Visual Pulses for Mice Navigating in Virtual Reality, suggesting that early inputs bias the cognitive framework through which subsequent data is interpreted. To quantify these dynamics, Alexandre Hyafil et al. utilized psychophysical kernels in Temporal integration is a robust feature of perceptual decisions to differentiate between stable integration and leaky integration, where the influence of older evidence decays over time.
When this temporal discounting or decay is extended across trials, choice history bias emerges naturally. Camille E. Rullán Buxó et al. demonstrated in A sampling-based circuit for optimal decision making that when a neural circuit applies a decay factor to past samples to prioritize recent information, previous experiences exert a residual influence on current probabilistic estimates. This creates a natural recency bias. Shensheng Wang confirmed this pattern in Modeling value integration during decision making under uncertainty with the Florida and Georgia gambling task, showing that sequential information gathering frequently leads to a recency bias where the most recently observed outcomes exert a disproportionately large influence on the final decision due to memory decay.
The Distributed Neural Architecture of History
Where does this historical memory live? Rather than being confined to a single "decision hub," choice history is represented globally across the brain.
Charles Findling et al. discovered brain-wide representations of prior information in Brain-wide representations of prior information in mouse decision-making, showing that subjective priors are shaped more significantly by an agent's recent behavioral choices than by the actual statistical frequency of external events. This internal model of past actions is maintained across a complex, loopy network spanning from sensory areas to motor output centers.
Specific regions act as critical nodes within this network:
- Lateral Orbitofrontal Cortex (lOFC): David Hocker et al. identified distinct functional clusters of neurons in Subpopulations of neurons in lOFC encode previous and current rewards at time of choice that uniquely re-activate memories of previous rewards right at the moment of choice, bridging historical reward data with current decisions.
- Cerebellum: Nico A. Flierman et al. found in Encoding of cerebellar dentate neuron activity during visual attention in rhesus macaques that dentate nucleus neurons maintain a representation of past trial outcomes, contributing to the adaptive calibration of behavioral strategies based on reinforcement history.
- Sensory Cortex: Even in early sensory areas like the MT cortex, Ramon Nogueira et al. showed in Trial by trial decoding of decisions in monkey MT cortex from small neuronal populations that population-level decoding reveals carry-over effects, where the initial confidence level at the start of a trial reflects the outcome of the preceding trial.
This pervasive reliance on past actions over pure rewards is further highlighted by automated cognitive model discovery. Using symbolic AI tools, Pablo Samuel Castro et al. revealed in Discovering Symbolic Cognitive Models from Human and Animal Behavior that choice history often plays a more dominant role in shaping human and animal decisions than reward history alone, pointing to internal perseveration as a core cognitive constraint.
While choice history bias is clearly a fundamental feature of mammalian cognition, the exact triggers that cause a brain to switch from objective, evidence-based decision-making to history-dependent perseveration remain an open question, as noted by Dario Campagner et al. in Prediction of Choice from Competing Mechanosensory and Choice-Memory Cues during Active Tactile Decision Making. Whether driven by sensory noise, attentional fluctuations, or internal neuromodulatory states, understanding these transitions remains central to deciphering how the brain balances its past with its present.
Key Foundational Work
Sources of noise during accumulation of evidence in unrestrained and voluntarily head-restrained rats (2015) — contributed 1 concept elements to this direction.
Trial by trial decoding of decisions in monkey MT cortex from small neuronal populations (2013) — contributed 1 concept elements to this direction.