An Abbreviated Survey of AI (In Fiction) (rznicolet.com)

A brief, informal survey of film portrayals of AI used Wikipedia’s list of movies and plot summaries to ask two questions: how often do on-screen AIs “go wrong,” and does that vary by the AI’s coded gender? The author identified roughly 40% of AI characters malfunctioning or turning dangerous (using a generous definition). Gender coding (based on actor/voice) found male-coded AIs went bad 43% of the time, ungendered/neutral 42%, and female-coded 33%. About half of the AIs were male-coded, with the rest split between feminine and neutral. Crucially, AI-centric films were far riskier: 53% of those AIs went wrong versus 18% in stories where AI was a background element. In AI-focused works, neutral AIs were worst (58%); in non-AI-focused films, feminine AIs misbehaved least (10%, though sample sizes are small). Beyond the numbers, the themes driving failures are telling for practitioners: “incompetent programming,” hacking/damage, and rebellion against authority dominate plot justifications. That matters because fictional tropes shape public expectations and policy pressure—reinforcing fears about weaponized or ungoverned systems even as current ML (e.g., LLMs) remain narrow tools. Practical takeaway: fictional narratives emphasize failure modes developers repeatedly warn about (poor safeguards, adversarial compromise, reward-misalignment), so better engineering, governance, and communication could shift the cultural script away from Skynet and toward Data/WALL‑E outcomes.