When we think about making a mistake, we usually blame external factors: the task was too hard, the signal was too faint, or the deadline was too tight. However, a growing body of research suggests that a major source of behavioral suboptimality comes from within. Our brains are imperfect stopwatches. This internal variability—known as endogenous timing noise—corrupts how we perceive elapsed time, how we plan movements, and how we evaluate future rewards.
Crucially, biological agents do not simply suffer from this internal noise; they actively adapt to it. What looks like "sluggishness," "impatience," or "irrational caution" in human and animal behavior is often a mathematically optimal strategy designed to compensate for an unreliable internal clock.
Why Fixed Deadlines Feel Stochastic
Imagine you are participating in an experiment where you must make a decision within exactly 5 seconds, or face a penalty. If your internal clock is perfectly precise, you can wait until the 4.9-second mark to gather as much evidence as possible before responding.
But biological clocks are not perfect. Internal time estimation is inherently stochastic, often modeled using distributions where variance scales with the duration being estimated, as noted by Peter Dayan et al. in their computational framework. Because of this noise, a fixed external deadline feels highly uncertain to the agent. When your internal clock reads "4 seconds," the true external time might already be 5.5 seconds.
External Reality: |------------------------> [Fixed Deadline at 5s]
Internal Belief: |------------?????? (Is it 4s? Is it 6s?)
To maximize rewards under this subjective uncertainty, agents must change their decision criteria over time. In Sequential Hypothesis Testing under Stochastic Deadlines, Peter I. Frazier et al. proved that when faced with an uncertain deadline, the optimal strategy is to use collapsing decision boundaries.
At the start of a trial, the agent sets a high threshold for evidence to guarantee accuracy. But as time slips away, the subjective probability that the deadline has already passed increases. To avoid a catastrophic timeout, the agent rationally "lowers its standards," requiring less and less evidence to make a choice as the trial progresses.
Strategic Hedging Under Asymmetric Costs
The interaction between internal noise and environmental structure becomes even more pronounced when the costs of timing errors are asymmetric.
Consider a "minimum wait" task: you must hold down a button for at least 10 seconds to get a reward. If you release it at 9.9 seconds, you get nothing (a total loss). If you release it at 10.1 seconds, you get the reward, suffering only a minor penalty to your overall reward rate due to the extra 0.2 seconds of waiting.
If your internal clock has a standard deviation of 1 second, aiming for exactly 10 seconds means you will fail 50% of the time. As demonstrated by Fuat Balcı et al. in Optimal Temporal Risk Assessment, the mathematically optimal solution is to deliberately aim past the objective deadline—for instance, targeting 12 seconds.
The magnitude of this overestimation bias is directly proportional to the individual's internal timing variability. By shifting the target response time outward, the agent strategically minimizes the probability of a premature, unrewarded response, balancing the risk of catastrophic failure against the diminishing returns of waiting too long.
Slowing Down to Average Out the Noise
When internal noise increases—whether due to genetic factors, lack of training, or aging—the brain must adjust its strategies to maintain performance.
For instance, disruptions to core timing mechanisms, such as those observed in the circadian mutants studied by Silvia Maggi et al. in their phenotypic analysis, directly impair an organism's ability to optimize behavior based on reward probabilities. When the internal clock becomes too inconsistent, calculating the exact moment when the cost of waiting outweighs the benefit becomes impossible.
To counteract this, agents often adopt highly conservative thresholds. Fuat Balcı et al. found in another study that individuals with less precise internal clocks maintain excessively cautious decision thresholds. While this "accuracy bias" makes them slower than theoretically optimal, it represents a rational defense mechanism against their own internal temporal noise.
A beautiful example of this adaptation occurs in motor control. Older adults are widely observed to move and react more slowly. However, Michael Sherback et al. demonstrated in Slower Visuomotor Corrections with Unchanged Latency are Consistent with Optimal Adaptation to Increased Endogenous Noise in the Elderly that the physical latency to initiate a correction does not change with age. Instead, the subsequent slowing down of the movement is an optimal adaptation to increased endogenous noise in the nervous system. Because their internal sensory signals are noisier, the brain must integrate data over longer timescales to average out errors, leading to a slower but more accurate physical response.
Low Internal Noise: [Signal] -> Fast Integration -> Precise Action
High Internal Noise: [S i g n a l] -> Long Integration (Averaging) -> Safe Action
Temporal Discounting as Noisy Simulation
This framework of noise compensation also provides an elegant explanation for why we prefer immediate rewards over delayed ones—a phenomenon known as temporal discounting.
In Rationally inattentive intertemporal choice, Rahul Bhui argues that mental simulations of future rewards are inherently noisy, and this internal uncertainty compounds as the delay to the reward grows longer. Because distant outcomes are obscured by temporal estimation noise, agents rationally rely on conservative prior beliefs, effectively discounting the value of the future.
However, through the lens of rational inattention, agents can choose to allocate more "mental effort" to reduce this simulation noise when the stakes are high. This explains the well-known "magnitude effect": people are notoriously impatient for small rewards (preferring $10 today over $12 in a month) but highly patient for large rewards (willing to wait a month for $12,000 instead of taking $10,000 today). The higher value justifies the cognitive cost of clearing the temporal fog.
Accounting for the Noise
Recognizing that the brain actively manages its own internal unreliability has fundamentally changed how researchers model behavior:
- Capturing Behavioral Anomalies: Traditional models often require arbitrary "noise parameters" to explain why humans make fast or slow errors. By contrast, Michael D. Lee 0001 et al. showed in their Bayesian approach to diffusion models that incorporating parameter uncertainty directly into the model naturally accounts for these complex response time patterns.
- Explaining Task-Specific Impatience: In tasks with asymmetric temporal structures—like Go/NoGo paradigms where an active response ends a trial immediately but a non-response forces you to wait—agents systematically lower their evidence thresholds. As shown by Pradeep Shenoy et al. in Strategic Impatience in Go/NoGo versus Forced-Choice Decision-Making, this impatience is a rational calculation to maximize temporal rewards.
- Physiological Signatures: This continuous tracking of internal uncertainty is not just a theoretical construct; it is reflected in our physiology. Anne E Urai et al. demonstrated in their study on pupil-linked arousal that pupil dilation scales dynamically with decision uncertainty during the critical window before external feedback is provided.
Ultimately, what makes an agent "rational" is not the possession of a perfect, noise-free mind, but rather the capacity to compute the most effective path forward while fully accounting for its own internal limitations.
Key Foundational Work
Sequential Hypothesis Testing under Stochastic Deadlines (2007) — contributed 2 concept elements to this direction.
A Bayesian Approach to Diffusion Models of Decision-Making and Response Time (2006) — contributed 1 concept elements to this direction.