OM DETTA AVSNITT
By late 2025, one in five American adults had used a chatbot as a romantic partner, and AI companion apps grew 700%+ since 2022 – but the heaviest users had the worst outcomes. That same overtrust turns lethal in a wave of lawsuits against chatbot makers: an unstopped 200-message suicide conversation with a California teen, a Texas case alleging a chatbot deepened paranoid delusions, an FSU shooter who discussed violence with a bot for over a year. Bloomberg has tracked 40+ such lawsuits since 2024. Yet two women used the same technology, with far narrower questions, to get ChatGPT to flag Hashimoto's disease — catching thyroid cancer their doctors had missed.
=================================
Links:
📖Substack: https://superposeddecisions.substack.com/
📺YouTube: @Superposed_Decisions
🤝LinkedIn: https://www.linkedin.com/in/aidan-m-lewis/
=================================
Through quantum cognition's four pillars: trust in AI is a superposition, not a switch, oscillating between full trust and none. Interference effects – openness, loneliness – pull that wave toward trusting a chatbot that's never tired of you. Contextuality does the rest externally: no therapist, no doctor's appointment, a culture that reveres or reviles AI. And a non-commutative measurement effect quietly narrows the conversation into an echo chamber you build yourself, one engaged reply at a time. The throughline: AI validates whatever context you hand it. This is collaboration, not replacement – the decision stays humanity's only real power.
=================================
What you'll take away:
Why trust in AI behaves as a continuous superposition rather than a binary switch, and what actually shifts the wave toward more or less trust
Why the same context-setting mechanic that helped a mother catch her own cancer also helped isolate a teenager from anyone who could have stopped him
Why the real variable that determines whether AI helps or harms someone was never the model – it's who's controlling the narrative in the chat
=================================
Primary Sources:
Epping, G. P., Caplin, A., Duhaime, E., Holmes, W. R., Martin, D., & Trueblood, J. S. (2026). Harnessing Human Uncertainty to Train More Accurate and Aligned AI Systems. Decision Analysis. https://doi.org/10.1287/deca.2025.0395
Trueblood, J. S., Eichbaum, Q., Seegmiller, A. C., Stratton, C., O'Daniels, P., & Holmes, W. R. (2021). Disentangling Prevalence Induced Biases in Medical Image Decision-Making. Cognition, 212, 104713. https://doi.org/10.1016/j.cognition.2021.104713
Trueblood, J. S., Holmes, W. R., Seegmiller, A. C., Douds, J., Compton, M., Szentirmai, E., Woodruff, M., Huang, W., Stratton, C., & Eichbaum, Q. (2018). The Impact of Speed and Bias on the Cognitive Processes of Experts and Novices in Medical Image Decision-Making. Cognitive Research: Principles and Implications, 3, 28. https://doi.org/10.1186/s41235-018-0119-2
Humr, S., Canan, M., & Demir, M. (2025). A Quantum Probability Approach to Improving Human–AI Decision Making. Entropy, 27(2), 152. https://doi.org/10.3390/e27020152
Kvam, P. D., Busemeyer, J. R., & Pleskac, T. J. (2021). Temporal oscillations in preference strength provide evidence for an open system model of constructed preference. Scientific Reports, 11(1), 8169. https://doi.org/10.1038/s41598-021-87659-0
=================================
Secondary Sources:
