Research
Before industry I spent about three years in cognitive science labs at UCLA and UC Riverside. Most of the work asked one question: does a person know when they are right?
Try it
Six short rounds. Some of the dots move together to the left or the right. Say which way, then say how sure you are.
This is a random-dot kinematogram, a standard stimulus in vision science. I built its first open browser version in 2017. Nothing you enter leaves this page.
Can you perceive something correctly without knowing that you can?
Some studies said healthy people can, for emotional faces. I ran a bias-free test: people judged facial emotions and then bet on which of their own answers to trust. Whenever they could tell the emotion, they also knew which answers to trust, exactly as a Bayesian ideal observer predicts.
Read the full story
Some patients with damage to the visual cortex can discriminate things they say they do not see. This is called blindsight. Earlier studies suggested that healthy people might show a version of it for emotional faces, especially fearful ones.
On each trial people saw two faces of the same person, one neutral and one with a small amount of emotion. They named the emotion for both, and then bet on which of their two answers was more likely to be right. Asking "did you see it?" mixes what someone perceived with how willing they are to say yes. A bet between two of your own answers removes that, because both answers face the same criterion.
Compared participant by participant against a Bayesian ideal observer and against versions with added noise that simulate blindsight, the ideal observer fit best. We found no perception without awareness in this task, and the paper recommends a lab study with more trials as the next step.
Asking whether an AI system's confidence tracks its accuracy is the same question, and asking "which of these two are you surer of?" strips out response bias in both cases. That link is my own analogy, not a claim the paper makes.
Rajananda, Zhu and Peters, Neuroscience of Consciousness, 2020. Read the paper. Experiment code.
A standard vision experiment, rebuilt for the browser
In 2017 the random-dot kinematogram existed in MATLAB and Python but not in a browser, so it could not be used in online experiments. I built the first open-access version in JavaScript and measured how accurately browsers display it. Across the preprint and the journal it has about 13,000 views and 2,900 downloads as of November 2026, and vision labs around the world have used it since 2017.
Read the full story
A random-dot kinematogram is a cloud of moving dots in which some fraction move together. Vision scientists use it to measure motion perception. I made the browser version configurable and packaged it as a plugin for jsPsych, a common framework for online experiments.
The same code runs differently in different browsers, as the chart shows. So the plugin records the number of frames actually shown on every trial, and a researcher can exclude or repeat the trials that were displayed badly. You cannot control a participant's hardware, so you measure what happened and filter.
The lesson carries over to AI agents: log what actually happened on each run, not what was supposed to happen. That is my analogy, not the paper's.
Rajananda, Lau and Odegaard, Journal of Open Research Software, 2018. Read the paper. Code. Original demo. I was invited to present these tools at the Vision Sciences Society conference.
Other papers
- Natural statistics support a rational account of confidence biasesNature Communications, 2023. Contributing author. A model of confidence built from deep neural networks on naturalistic images. It explains confidence biases that look irrational as the result of being tuned to real sensory inputs.
- The relative psychometric functionPsychological Review, 2026. Contributing author. A general method, with an open-source toolbox, for relating one psychological measure to another, such as accuracy to confidence, across conditions.
- New measures of agency from an adaptive sensorimotor taskPLOS ONE, 2020. Second author, software and programming. A short online task that measures the sense of being the author of your own movements, reliable in healthy participants.
- Elective deliveries and the risk of autismAmerican Journal of Preventive Medicine, 2022. Third author, data curation. A study of about 1.5 million low-risk California births. I cleaned and linked the birth and medical records. It reports an association, not a cause.
- Altered processing of complex visual stimuli after concussionAmerican Journal of Neuroradiology, 2021. Software and programming. An fMRI study of why some people become sensitive to visual motion after a concussion.
- Blockchain for electronic vaccine certificates: more cons than pros?Frontiers in Big Data, 2022. Third author. A review I initiated during my master's. It concludes the technology is poorly suited to the job.
- Subjective inflation of color saturation in the peripheryPreprint, 2017. First author. Rapid flashing makes colors at the edge of vision look more saturated than they are, by a modest amount.
Open-source tools
- Random-dot kinematogram for the web, available as a jsPsych plugin.
- A PEST adaptive staircase procedure in JavaScript, the first open-access version.
- A replication of the Confidence Signal Detection model, in JavaScript and MATLAB.
Full publication list
On the first-author papers I led the experiment design with a postdoctoral mentor, built and ran the experiments, analyzed the data, and wrote the first drafts of the methods and results. Citation counts are on my ResearchGate profile.
Show all nine references
- Maniscalco, B., Graham Castaneda, O., Odegaard, B., Morales, J., Rajananda, S., Denison, R. N., & Peters, M. A. K. (2026). The relative psychometric function: A general analysis framework for relating psychological processes. Psychological Review.
- Webb, T. W., Miyoshi, K., So, T. Y., Rajananda, S., & Lau, H. (2023). Natural statistics support a rational account of confidence biases. Nature Communications.
- Toubiana, R., Macdonald, M., Rajananda, S., Lokvenec, T., Kingsley, T. C., & Romero-Brufau, S. (2022). Blockchain for electronic vaccine certificates: More cons than pros? Frontiers in Big Data.
- Liu, K.-Y., Teitler, J. O., Rajananda, S., Chegwin, V., Bearman, P. S., Hegyi, T., & Reichman, N. E. (2022). Elective deliveries and the risk of autism. American Journal of Preventive Medicine, 63(1), 68–76.
- Allen, J. W., Trofimova, A., Ahluwalia, V., Smith, J. L., Abidi, S. A., Peters, M. A. K., Rajananda, S., Hurtado, J. E., & Gore, R. K. (2021). Altered processing of complex visual stimuli in patients with postconcussive visual motion sensitivity. American Journal of Neuroradiology, 42(5), 930–937.
- Wang, S., Rajananda, S., Lau, H., & Knotts, J. D. (2020). New measures of agency from an adaptive sensorimotor task. PLOS ONE.
- Rajananda, S., Zhu, J., & Peters, M. A. K. (2020). Normal observers show no evidence for blindsight in facial emotion perception. Neuroscience of Consciousness, 6(1), niaa023.
- Rajananda, S., Lau, H., & Odegaard, B. (2018). A random-dot kinematogram for web-based vision research. Journal of Open Research Software, 6, 6.
- Rajananda, S., Peters, M. A. K., Lau, H., & Odegaard, B. (2017). Subjective inflation of color saturation in the periphery under temporal overload. bioRxiv (preprint).