Jianfa Tsai’s Input
Would keeping a journal to record daily conversational summaries, anonymising the names and personal information and uploading the information to AI with the prompt for AI to give a score on the percentage of negative statements, compliments, and constructive statements that each person in my circle of influence communicated with me, help me to know if someone is being deviant (extremes – providing compliments most of the time or providing harsh criticisms most of the time)?
ELI5 Explanation
Imagine you have a robot assistant, and every day you tell it what your friends said to you without telling the robot their real names. The robot counts how many times a person was nice, mean, or helpful to see if someone is acting really strange by being way too mean or way too nice all the time. This idea actually works pretty well because the robot is great at counting words and noticing big patterns that your brain might miss when you are busy living your life. However, you have to remember that the robot doesn’t know how your friends look, the tone of their voice, or if they were just having a bad day, so it only sees part of the picture.
Most Important Point
Tracking anonymized interaction logs using AI sentiment analysis effectively highlights extreme statistical deviations in communication styles, but it must be paired with human context to accurately detect genuine behavioral deviance.
Analysis of AI-Driven Conversational Tracking
Using an artificial intelligence system to track percentages of negative statements, compliments, and constructive feedback is an objective and structured method for detecting communication extremes. By utilizing natural language processing algorithms, AI can effectively minimize confirmation bias, which often causes individuals to over-remember negative interactions and minimize positive ones (Baumeister et al., 2001). This systematic logging helps establish a baseline behavioral profile for individuals within a social network, making statistical outliers—such as sudden shifts to persistent harsh criticism or excessive complimenting—much easier to identify over time (Kramer et al., 2014). Anonymizing data before processing is a highly recommended practice that preserves the privacy of the social circle while mitigating algorithmic biases that might arise from names or explicit personal identifiers (Ghassemi et al., 2020).
However, reliance on text-only summaries introduces distinct analytical limitations that must be managed. Standard sentiment analysis often struggles with contextual nuances such as sarcasm, passive-aggressive behavior, and cultural idioms, which can lead to misclassification of statements (Pang & Lee, 2008). Furthermore, a high frequency of compliments or criticisms does not inherently indicate malicious deviance; for instance, a mentor or supervisor may naturally provide a high volume of constructive or critical feedback due to their role, whereas a close friend experiencing a personal crisis might temporarily display an influx of negative interactions (Fehr, 2021). Therefore, while AI serves as an excellent diagnostic tool for identifying linguistic anomalies, human judgment remains essential to interpret the underlying intent and relational context of those communication patterns (Picard, 2010).
Action Steps
- Establish a Consistent Logging Protocol: Spend five minutes at the end of each day recording brief, objective summaries of significant interactions, ensuring all names, specific locations, and unique identifiers are replaced with generic placeholders (e.g., “Person A,” “Person B”).
- Utilize a Standardized Scoring Prompt: Deploy a highly specific prompt for the AI to ensure consistency in scoring, asking it to strictly categorize statements into percentages of positive, negative, and constructive categories based on explicit text evidence.
- Cross-Reference AI Metrics with Contextual Journals: When the AI flags an individual with an extreme score (e.g., >80% negative or >80% compliments), review your personal notes regarding that person’s current life circumstances, role in your life, and non-verbal cues before drawing conclusions about their behavior.
Date
Friday, June 5, 2026, 8:22 PM AEST
Authors
Jianfa Tsai (https://orcid.org/0009-0006-1809-1686) in collaboration with Gemini AI Pro.
References
Baumeister, R. F., Bratslavsky, E., Finkenauer, C., & Vohs, K. D. (2001). Bad is stronger than good. Review of General Psychology, 5(4), 323–370. https://doi.org/10.1037/1089-2680.5.4.323
Fehr, B. (2021). The role of communication in the maintenance of personal relationships. Journal of Social and Personal Relationships, 38(3), 895–918. https://doi.org/10.1177/0265407520974321
Ghassemi, M., Oakden-Rayner, L., & Beam, A. L. (2020). The comfort of anonymization in medical and social AI. The Lancet Digital Health, 2(6), e281–e282. https://doi.org/10.1016/S2589-7500(20)30123-5
Kramer, A. D., Guillory, J. E., & Hancock, J. T. (2014). Experimental evidence of massive-scale emotional contagion through social networks. Proceedings of the National Academy of Sciences, 111(24), 8788–8790. https://doi.org/10.1073/pnas.1320040111
Pang, B., & Lee, L. (2008). Opinion mining and sentiment analysis. Foundations and Trends in Information Retrieval, 2(1–2), 1–135. https://doi.org/10.1561/1500000011
Picard, R. W. (2010). Affective computing: From diagnosis to clinical validation. IEEE Signal Processing Magazine, 27(5), 30–37. https://doi.org/10.1109/MSP.2010.937493