An fMRI study of scientists with a Ph.D. in physics confronted with naive ideas in science

The study addresses how persistent naive ideas are in science, even among advanced experts. Naive ideas often conflict with scientific concepts, are reinforced by everyday experience, and can be hard to suppress. Conceptual change is thought to involve inhibiting prepotent naive responses to allow scientific reasoning. Prior behavioral work shows slower, less accurate performance when naive ideas conflict with correct judgments. Neuroimaging in students and undergraduates links overcoming naive ideas to activation in inhibitory control regions (inferior and middle frontal gyri and anterior cingulate cortex). This study investigates whether these naive ideas persist in scientists with Ph.D.s in physics and tests the hypothesis that judging statements that trigger naive ideas (incongruent) will engage IFG, MFG, and ACC more than matched statements where naive and scientific ideas align (congruent). A secondary hypothesis is that such inhibitory control engagement would be more pronounced for biology (basic expertise) than for physics (advanced expertise) in these participants.

Prior research indicates that naive ideas are resilient across ages and can coexist with scientific conceptions. Behavioral evidence shows increased response times and errors for tasks that elicit naive ideas, supporting the role of executive control and inhibition in conceptual change. Experts still exhibit intuitive biases (e.g., teleological reasoning, impetus beliefs). Neuroimaging comparisons between experts and novices have consistently implicated inhibitory control regions—bilateral IFG, MFG, and ACC—in successful overcoming of naive ideas in physics-related tasks. These findings suggest that learning does not erase naive ideas but requires their suppression during scientific reasoning. However, previous neuroimaging work has focused mainly on undergraduates; persistence in fully trained scientists remained untested prior to this study.

Participants: 25 right-handed adults with a Ph.D. in physics (23 male, 2 female; age 28–60, mean ~45). Roles included university professors (n=17), postdocs (n=3), senior high school teachers (n=5), and one developer-researcher. All gave informed consent; ethics approval by Quebec's Neuroimaging Network (CMER RNQ 13–14–023). Task and stimuli: A scientific statements verification task comprising 128 statements: 64 physics and 64 biology. Within each domain, statements were organized as matched pairs of congruent (naive and scientific ideas align) and incongruent (naive idea conflicts with the correct scientific judgment). Examples: congruent—"a campfire contains thermal energy"; incongruent—"an ice cube contains thermal energy." Statements were validated by professional physicists and biologists and targeted persistent naive ideas (e.g., mechanics, thermodynamics, natural selection, photosynthesis). Across congruent vs incongruent conditions, statements were matched for response type (64 true, 64 false), word count (8.8±3.7 vs 8.9±3.8 words; t(126) = -0.05, p=0.96), syllable count (13.0±5.2 vs 13.0±5.1; t(126) = 0.02, p=0.99), and readability (Flesch ~73). Readability across domains was similar (physics 75.7; biology 70.2). Procedure and design: Block design with four fMRI runs; each run had eight blocks (four statements each) presented in random order; two additional mixed filler blocks per run (not analyzed). Each trial lasted up to 10 s or until response; inter-block fixation 15 s. Instructions emphasized speed and accuracy. Pre-scan practice (16 trials) was conducted on desktop and in a mock scanner. Responses were made via a right-hand button box: index finger = scientifically correct; middle finger = scientifically incorrect. Stimulus presentation: E-Prime 2.0 projected to a screen; mirror above head coil. MRI acquisition: 3T Siemens Prisma, 32-channel head coil. Functional EPI: TR 2000 ms, TE 30 ms, FA 90°, FOV 192 mm, matrix 64×64, voxel 3×3 mm, 33 slices (3 mm, 25% gap), interleaved; 120–200 volumes per run; first two volumes discarded. Structural T1 MPRAGE: TR 2300 ms, TI 900 ms, TE 2.26 ms, FA 9°, FOV 256 mm, matrix 256×256, 176 slices, 1×1 mm. Preprocessing: SPM8. Realignment (motion correction), coregistration (T1 to EPI), normalization to MNI using single generative model segmentation, smoothing with 8 mm FWHM Gaussian kernel. High-pass filter 128 s. Statistical analysis: First-level GLM modeling block conditions convolved with HRF. Second-level whole-brain ANOVA (within-subject 2×2): EXPERTISE (advanced: physics; basic: biology) × CONGRUENCY (congruent vs incongruent). Main effect and interaction reported with cluster-wise FWE-corrected p<0.05 across the whole brain; primary voxelwise threshold p_uncorr<0.005; expected voxels per cluster as specified. Post hoc t-contrasts (incongruent>congruent and vice versa) at p_uncorr<0.005. Region labels from AAL atlas. Behavioral measures: Accuracy (%) and response times (ms) recorded; paired t-tests assessed congruency effects within domain and domain differences.

Behavioral: - Overall accuracy higher in physics than biology: physics 88.4±11.4% vs biology 83.0±9.0%; t(24)=4.360, p<0.001, d=0.5, 95% CI [-8.0, -2.9]. - Congruent > incongruent accuracy in both domains: • Physics: 96.3±4.9% vs 80.6±10.6%; t(24)=8.522, p<0.001, d=1.7, 95% CI [11.8, 19.4]. • Biology: 89.6±4.9% vs 76.4±7.1%; t(24)=8.888, p<0.001, d=1.5, 95% CI [10.2, 16.3]. - Congruent faster than incongruent in both domains: • Physics: 3542±754 ms vs 4181±811 ms; t(24)=10.988, p<0.001, d=2.1, 95% CI [-759.1, -519.0]; Δ +639 ms (+18%). • Biology: 3182±699 ms vs 3455±749 ms; t(24)=4.830, p<0.001, d=0.9, 95% CI [-389.1, -156.2]; Δ +273 ms (+9%). Neuroimaging—Main effect of congruency (incongruent>congruent): - Greater activation in left inferior frontal gyrus (orbital/ventrolateral PFC), bilateral superior frontal gyrus (medial), and bilateral anterior cingulate cortex. T-contrasts also showed bilateral middle frontal gyrus and bilateral cerebellum (crus I/II) more active for incongruent. - Congruent>incongruent: bilateral middle occipital gyrus. Interaction EXPERTISE × CONGRUENCY: - Significant cluster in bilateral dorsal anterior cingulate cortex (dACC), FWE cluster p<0.05. Within-domain contrasts: - Physics, incongruent>congruent: left supplementary motor area, left middle occipital gyrus, left superior frontal gyrus, bilateral middle frontal gyrus, left inferior frontal gyrus. Congruent>incongruent: right middle occipital gyrus. - Biology, incongruent>congruent: left inferior frontal gyrus, bilateral anterior cingulate cortex, bilateral caudate, right insula, bilateral superior frontal gyrus, right midcingulate gyrus. Congruent>incongruent: bilateral middle occipital gyrus, left precentral gyrus. Interpretation: Incongruent (naive-idea-interfering) trials robustly engaged frontal executive/inhibitory control networks (IFG, MFG, ACC), while congruent trials recruited posterior regions associated with more automatic/perceptual processing (middle occipital gyrus). ACC involvement was stronger in biology, consistent with greater uncertainty/conflict monitoring under basic expertise.

Findings show that even scientists with Ph.D.s in physics experience interference from persistent naive ideas: when naive and scientific ideas conflict, judgments are slower and less accurate. Corresponding BOLD increases in IFG and MFG support the need for inhibitory control to suppress prepotent naive responses and select the scientifically correct answer. Congruent statements, which do not require such suppression, elicited more posterior, perceptual processing patterns, consistent with more automatic processing. The EXPERTISE × CONGRUENCY interaction localized to dACC indicates domain-dependent conflict monitoring: incongruent biology statements (basic expertise) elicited stronger ACC activity, plausibly reflecting increased conflict/uncertainty and recruitment of control. In physics (advanced expertise), inhibition-related IFG/MFG engagement remained evident without a strong ACC effect, suggesting that while inhibition is still required, confidence is higher and conflict monitoring may be reduced. Overall, results support a coexistence model of conceptual understanding in which naive and scientific ideas cohabit; expert performance relies on dynamic suppression of misleading intuitions when context necessitates it.

The study demonstrates that some naive scientific ideas persist even among advanced experts. When faced with statements that trigger naive ideas, Ph.D. physicists show performance costs and increased activation in frontal inhibitory control networks (IFG, MFG), with additional ACC engagement particularly when judging outside their core domain (biology). Congruent statements preferentially recruit posterior regions consistent with more automatic processing. These results argue against complete replacement of naive ideas through training and instead support a model of expertise characterized by vigilant, flexible control over competing representations. Future research should pursue longitudinal designs tracking novices to experts, compare multiple scientific fields with better-matched stimuli across domains, integrate uncertainty measures, examine individual differences (including sex as a biological variable), and employ complementary analytic strategies and thresholds to refine spatial specificity.

- Reverse inference: inferring cognitive processes (inhibition, conflict monitoring) from activation patterns is inherently limited; multiple processes can map to similar networks. - Cluster-extent thresholding with a liberal primary voxelwise threshold (p_uncorr<0.005) may reduce spatial specificity and increase false positives; large clusters can span multiple regions. - Small sample (N=25) and sex imbalance (23 male, 2 female) limit generalizability and preclude analysis of sex differences. - Cross-sectional design; no causal or developmental inference about conceptual change over time. - Domain comparison caveats: statements across physics and biology were not strictly matched for word/syllable counts at the domain level and may differ in content and epistemological features; time since formal study differed between domains for participants. - Behavioral task uses verbal statements; findings may not generalize to other modalities or problem types.