Surprising Sounds Influence Risky Decision Making

The study investigates whether unexpected, task-irrelevant sensory events systematically bias decisions under risk. Sensory prediction errors are salient and common in real-world environments but often irrelevant to goals. Prior work suggests unpredictable sensory inputs can impair decision performance via distraction or noise, yet sensory surprise might also trigger stereotyped responses such as uncertainty seeking (risk taking) and option switching, behaviors that are adaptive in learning contexts. Here, the authors test whether auditory surprises, devoid of informational value about outcomes, alter risk preferences and choice perseveration in a risky choice task where option values vary unpredictably across trials. This design isolates incidental sensory effects from value learning. The central hypothesis is that sensory prediction errors elicit stereotyped, value-independent biases toward risk taking and away from perseveration.

Prior research shows unpredictable, irrelevant sensory information can impair task performance through distraction and changes in salience or loudness. Other work links surprise to exploratory behaviors in learning tasks and proposes dopaminergic mechanisms for value-independent decision biases. Pharmacologically increasing dopamine has been associated with increased value-independent risk taking and reduced perseveration. Neurophysiological studies indicate that sensory prediction errors (including identity violations) engage dopaminergic midbrain and distributed cortical signals. The authors position their work within this literature by testing for stereotyped, value-independent effects of auditory surprise on risk and perseveration, and contrasting such effects with accounts based on attentional lapses, increased choice stochasticity, or learning-related adaptations.

Ethics and participants: Approved by the University IRB (protocol #200028524). Data were collected online (Gorilla Experiment Builder) via Prolific between 02/24/2021 and 09/05/2023. Inclusion: fluent English, age ≥18. Headphone use was enforced via an online auditory screening task. Across seven experiments, n = 1600 (695 female, 291 male, 1 other/prefer not to say; mean age 29.0 ± 1.7 years). Flat monetary compensation (~$8.36/h). After the task, participants completed PHQ-9 and GAD-7.

Task and stimuli: On each trial, participants chose between a risky and a safe option. Risky options had equal probabilities of win and loss outcomes; outcomes (if risky chosen) were revealed after a 1.3 s delay. Immediately before option presentation, participants heard a six-tone auditory sequence (100 ms tones separated by 400 ms; pure tones A=1000 Hz, B=1500 Hz, C=500 Hz). In most experiments (1–4, 7), 75% of trials used a common sequence and 25% a rare sequence; in Experiments 5–6, two sequences were presented equally often (50/50), either alternating deterministically (Exp. 5) or randomly shuffled (Exp. 6). Sequences were task-irrelevant and carried no information about outcomes.

Experimental manipulations: Experiments 1–2 used rare sequences ending with a deviant tone (e.g., AAAAA B/C), with Experiment 1 fixing the risky option’s side and Experiment 2 alternating left-right positions every 10 trials. Experiments 3–4 swapped rarity/deviance structure so rare sequences ended on standard tones, testing whether perseverance and risk-taking effects dissociate. Experiments 5–6 balanced sequence frequencies to eliminate predictability/rarity. Experiment 7 was preregistered and assessed effects while measuring participants’ beliefs about tone relevance.

Design matrix: 48 unique trial types: 16 Gain (certain gain vs risky higher gain or zero), 16 Mixed (certain outcome vs risky with gain and loss), 16 Loss (certain loss vs risky larger loss or zero). Gain/loss magnitudes used multipliers on base amounts (e.g., gain certain amounts +25, +50; loss certain amounts −25, −50; multipliers 1.64, 1.76, 1.88, 2.0, 2.5, 3.0, 4.0; Mixed gains +40, +80 with multipliers 0.2, 0.4, 0.55, 0.67, 0.8, 1.0, 1.33, 2.0). Experiment 1: 192 trials; Experiments 2–7: 120 trials, with pseudo-randomization ensuring target sequence frequencies and coverage of trial types.

Computational modeling: A prospect-theory-based choice model was fit, with a softmax mapping of subjective utilities to choices. To capture surprise effects, models included difference parameters on Rare vs Common trials for value-independent risky bias (δriskBias) and for perseveration (θpreserv), forming the Risky Bias Perseveration Difference model. Alternative models placed differences on utility curvature, risk parameters, loss aversion, lapse rate (Lapse Rate Difference model), or choice stochasticity/temperature (Stochasticity Difference model). Model comparison and parameter recovery were performed; nonparametric statistics (Wilcoxon tests) and Bayes Factors were reported. Reaction times and side alternation effects were also analyzed.

• Sensory prediction errors increase risk taking: In Experiments 1–2 (n = 400), participants chose the risky option more often on Rare than Common trials. Across both experiments, the average increase was approximately +2.06 ± 1.09% (p = 0.005), and effects replicated with alternating option sides. The effect held across Gain, Mixed, and Loss trial types (e.g., Gains: +3.09 ± 1.12%, p < 0.001; Mixed: +2.17 ± 0.97%, p = 0.010; Loss: +3.17 ± 0.88%, p < 0.001). Differences in risky outcome win rates between Rare and Common trials did not account for the effect (p = 0.099; BF01 ≈ 4.68).
• Sensory prediction errors decrease choice perseveration: Participants were less likely to repeat their previous choice on Rare vs Common trials (combined n = 400, −3.69 ± 0.81%, p < 0.001). The effect was present across trial types (e.g., Gains: about −3.5%, Mixed: about −4.6%, Loss: about −2.7%).
• Modeling supports value-independent biases: The Risky Bias Difference model (δriskBias on Rare trials) best accounted for increased risk taking (Exp. 1: 0.125 ± 0.052, p = 0.004; Exp. 2: 0.195 ± 0.056, p < 0.001). Adding a perseveration difference parameter captured decreased staying (Exp. 1 θpreserv: −0.149 ± 0.041, p = 0.001; Exp. 2 θpreserv: −0.126 ± 0.048, p = 0.010). Lapse Rate and Stochasticity Difference models failed to reproduce the risk/perseveration patterns and were disfavored by model comparison.
• Dissociability of effects: Risky Bias and Perseveration difference parameters were only weakly correlated (Spearman ρ = −0.119, p = 0.017). In Experiments 3–4 (rare sequences ended on standard tones), decreased perseveration persisted but increased risk taking was not observed (risk-taking Rare–Common differences not different from zero), indicating dissociable processes influenced by auditory statistics.
• Predictability eliminates effects: In Experiments 5–6 (balanced 50/50 sequence frequencies, deterministic alternation or random shuffle; n = 400), both the risk-taking and perseveration effects were eliminated behaviorally and in model parameters (δriskBias not different from zero; perseveration differences absent), indicating that surprise/prediction violation is necessary.
• Preregistered replication: Experiment 7 (n = 400) replicated main findings in a preregistered design; model parameters showed δriskBias > 0 (0.109 ± 0.005, p = 0.004) and δexplor/θpreserv < 0 (−0.208 ± 0.004, p < 0.001), demonstrating robustness even considering participants’ reported beliefs about tone relevance.

Findings show that incidental auditory surprises bias risky decision making in two value-independent ways: increasing a risky choice bias and reducing choice perseveration. These effects are not explained by increased lapses, attentional enhancement, or altered choice stochasticity, as alternative models failed to reproduce the empirical patterns. Instead, a prospect-theory-based model augmented with value-independent risky bias and perseveration parameters best captured behavior. The dissociation across experiments suggests distinct mechanisms responsive to different auditory statistics: deviant-ending sequences promote risk bias, while rarity/prediction violation reduces perseveration even without deviance. This aligns with proposals linking surprise to dopaminergic signaling; prior pharmacological and neural evidence connects dopamine to value-independent risk preference changes and perseverance modulation. The authors interpret increased risk taking as akin to directed exploration and reduced perseveration as akin to random exploration, indicating that incidental uncertainty can shift arbitration between exploration strategies even when such shifts are not adaptive. The ubiquity of surprising sensory events implies potential real-world impacts on everyday choices, from consumer behavior to financial and gambling contexts.

Across seven experiments, task-irrelevant auditory surprises systematically increased risk taking and decreased choice perseveration in a risky choice paradigm, with effects captured by value-independent decision biases rather than changes in decision noise. Manipulating auditory statistics dissociated the two effects and demonstrated necessity of unpredictability: when sequences were balanced or fully predictable, both effects vanished. The results motivate neurobiological investigations of dissociable mechanisms, potentially dopaminergic, and suggest that incidental sensory events may be an unrecognized source of variability in everyday decision making. Future work should test neural mechanisms directly (e.g., pharmacology, patient studies, animal models), explore other sensory modalities, and examine generalization to real-world environments.

• No direct neural measurements or dopaminergic manipulations were performed; mechanistic interpretations remain inferential.
• Online data collection with flat monetary compensation (not performance-contingent) could influence engagement and risk preferences.
• Auditory surprise phenomenology may vary across individuals (e.g., musical expertise), but individual differences were not a primary focus.
• Some experiments involved complex auditory statistics; in Experiments 3–4 the lower overall rarity of deviant tones may reduce sensitivity for detecting risk-taking effects without larger samples.
• Effects were shown with pure tone sequences; generalization across sensory modalities and more naturalistic sounds requires further testing.