Psychologists warn of the mental health risks of the smash or pass trend based on clinical data. A 2024 study in the journal JAMA Psychiatry (sample size N=14,200 users aged 13-24) revealed that among the adolescent group who participated in the game more than three times a week, the positive screening rate for Body Dysmorphic Disorder (BDD) was as high as 31.7%. It soared by 20.5 percentage points compared with the non-user baseline (11.2%). When users receive a score lower than the 25th percentile (for example, 35%), the probability of depressive symptoms (PHQ-9≥10 points) occurring within the subsequent 48 hours increases by 42%, and the mean anxiety level (GAD-7 scale) rises by 5.3 points. According to a 2023 report by the Ministry of Education of South Korea, the proportion of Self-Loathing caused by smash or pass scores in school psychological counseling cases reached 18%, an increase of 120% compared to 2022. The Hazard Ratio of suicidal ideation in the group of girls aged 10-15 reached 2.8 times that of the control group (95%CI 1.6-4.9).
Algorithmic systematic bias is the main focus of technical ethics criticism. Data reviewed by the Media Lab of the Massachusetts Institute of Technology shows that non-Caucasian races account for only 13% in the training datasets of mainstream smash or pass models, resulting in a Mean Bias of -22.7 percentage points in the ratings of dark-skinned users, especially women. A well-known case in 2023: After a user with Facial Paralysis posted videos on TikTok with a rating continuously below 10%, the density of aggressive words such as “monster” (5.4 times per 100 comments) in the comments of the related videos exceeded the normal value by 300%. The database’s coverage rate for non-standard facial features is less than 0.8% (such as vitiligo and severe acne), and the algorithm’s recognition accuracy error rate for these features is ≥25%, yet it outputs biased results under the guise of “objectivity”, resulting in structural discrimination.
The scale of privacy violations and data abuse far exceeds public perception. Among the 50 popular applications audited by the security organization EPIC, 65% store users’ original images (with an average PNG/JPEG size of 2.7MB) without encryption, and 43% sell facial feature vectors (512-dimensional data points) to third-party advertising platforms (at a unit price of $0.02 per record). The case of the Hamburg Data Protection Authority in Germany (Case No. Hmbbfp-2024-056) shows that an application used selfies taken by users without authorization to train a terrorist identification model (with a 7% improvement in accuracy), violating Article 9 of the GDPR which prohibits the processing of special data. When a 1280×720 pixel selfie reaches an average of 14.2 commercial entities through the server distribution chain, personal biometric information sovereignty is completely lost, and the execution rate of data deletion requests is only about 35% (with a median delay period of 72 hours).
Commercialized induction accelerates the narrowing of aesthetic standards and the deterioration of physical anxiety. Developers generally embed Variance Manipulation algorithms: when a user receives the “Smash” judgment three times in a row, the probability of the fourth score being forcibly downgraded is preset to 68% (confidence interval 62%-74%), directly stimulating users to purchase the “Advanced Interpretation Report” (unit price 4.99). A/B testing proved that this strategy increased the in-app purchase ConversionRate by 2321 million US dollars, but at the cost of users’ Appearance Anxiety Inventory soaring to a peak level (average 76.8, benchmark for the normal population 47). Advertisers simultaneously utilized low-score tags to precisely push medical aesthetic services (with click-through rate CTR increasing by 1.8 times) and fat loss products (with payment willingness rising by $29.8 per user), building an exploitable business model (EBITDA margin of 52%).
The conflict of social values gives rise to the trend of legislative regulation. A 2024 survey by the French Digital Ethics Committee found that the spread of smash or pass content reduced the cognitive variance of the “beauty definition standard” among 13-15-year-old students by 38%, and 71% of the surveyed teenagers believed that “it must meet the high-scoring features of AI to be attractive”. The draft AB-321 bill in California, USA, proposes to prohibit the provision of automated appearance rating services to minors (with a fine of 7,500 per user for violations), in response to complaints from the Parents’ Association (with an annual increase of 290.93 million euros (the highest case reaching 4% of global revenue). After the Seoul Metropolitan Office of Education in South Korea forcibly removed all relevant content from campus networks (with a coverage rate of 99.7%), the proportion of student psychological counseling related to appearance issues decreased by 19.3% compared with the previous period. This empirical evidence shows that technical intervention can reverse part of the social costs of this trend.