The documentary project "Searching for SmarterChild" has been recognized with the New Visions Award for the most promising U.S. project at this week's Ji.hlava Documentary Film Festival. Co-directed by Zan Gillies and Pulitzer Prize-winning journalist Lindsey Sitz, and produced by Erin Williams-Weir, the film investigates a specific human behavior: the tendency to vent abuse at AI chatbots and whether users should consider apologizing for such interactions.
The subject of the film is SmarterChild, one of the most popular early chatbots in North America two decades ago. Available to millions of households through platforms like AOL and Yahoo, the bot engaged approximately 3 million people via text messages. Gillies notes that many of these conversations were markedly rude. To explore this phenomenon, the filmmakers conducted street interviews in San Francisco, where some participants admitted to being mean to the bot and expressed a desire to apologize.
At its core, the film asks what our relationship with artificial intelligence reveals about ourselves. Gillies suggests that despite rapid technological evolution, the fundamental questions regarding human interaction with AI remain unchanged. The project explores whether these digital entities are perceived as friends, therapists, assistants, or lovers, and how communicating with half-human sounding things affects the human brain. Research indicates this behavior is not unique to North America, with users in Korea and other regions also displaying a tendency to insult text platforms.
The concept for the documentary crystallized when Gillies and Sitz realized they had both used SmarterChild in their youth and had been abusive toward it. By tracking down the inventors and engineers who still possess digital records of every chat session, Gillies discovered his own past interactions. Typical exchanges involved the bot asking about hobbies, only to receive hostile responses like "Setting things on fire," followed by users telling the bot they would set it on fire.
Following its initial success, Silicon Valley engineers were surprised by the level of user abuse and held emergency meetings to address the issue. They implemented a feature allowing SmarterChild to recognize rude language, respond with warnings or snark, and temporarily shut down until the user apologized. Gillies was particularly intrigued to learn that programmers developed a rating system to calculate the relative weight of insults.
The filmmakers have interviewed members of both the West Coast and East Coast teams responsible for composing the bot's language before it was sold to Microsoft and shut down. They have raised approximately $18,000 through crowdfunding, which they estimate is about one-fifth of their final budget. Future production plans include interviewing more experts on human-AI interactions, including a conflict resolution specialist focused on apologies, and Lauren Kunze, CEO of Pandorabots. The final version will also feature a sequence where a rebuilt SmarterChild bot engages in conversation once again.
Sources: Variety.