Concerns over the use of AI in films | BBC News

Key Concepts

• Artificial Intelligence (AI) in Filmmaking: The application of AI technologies to various stages of film production, from scriptwriting and visual effects to editing and performance.
• AI Tools Mentioned: Runway (AI video generation platform), Google's V3 (likely referring to a text-to-video or image generation model).
• Prompting: The method of interacting with AI models by providing text-based instructions.
• Bias in AI Training: The phenomenon where AI models reflect and perpetuate biases present in the data they are trained on, leading to unintended or discriminatory outputs.
• Hybrid Model: A future scenario where AI tools are integrated with traditional filmmaking techniques rather than replacing them entirely.
• Human Creativity and Skill: The enduring value of human artistic input, directorial vision, and technical expertise in filmmaking.

AI's Potential and Perils in Filmmaking

The film industry is exploring the significant potential of Artificial Intelligence (AI) to revolutionize production processes, promising benefits such as reduced development costs and accelerated production paces. However, this technological advancement is met with apprehension from industry professionals who fear AI could threaten jobs, leading to past industrial actions. AI correspondent Mark Chislac reports on this evolving landscape.

Testing AI Capabilities with a Director

To assess the practical implications of AI in filmmaking, Benjamin Baroot, director of the psychological horror film "Daddy's Head," was enlisted to test current AI technologies. The experiment involved using a single camera and AI tools like Runway and Google's V3.

Challenges and Limitations of AI Tools

While AI offers "a world of potential for movie makers," the practical application revealed significant drawbacks. Baroot encountered issues with AI's ability to execute large-scale scenes, often producing "strange mistakes."

• Unintended Ethnic Alterations: A recurring problem was the AI's persistent attempt to make the subject (Mark Chislac) appear white, despite original intentions.
• Difficulty with Specific Elements: An attempt to create a dragon in the background failed, though a volcano was eventually generated.
• Inefficiency and Randomness: To achieve just "10 working shots," Baroot had to generate "150 shots." This inefficiency stems from the lack of granular controls beyond prompting, leading to excessive interpretation and randomness. Baroot described the prompting process as "so annoying" and "random."

Control and Reasoning Deficiencies

A major challenge identified by Baroot was the accessibility of AI tools to a consumer level without corresponding control or reasoning mechanisms. The AI's output is described as "so random" and "really hard to control."

• Unpredictable Recreations: The AI can "get rid of you and makes you again," but the lighting and performance are not consistently controllable. This raises questions about the authenticity of the recreated performance, with Baroot questioning, "Was that Mark?" and "Was that his performance?"
• Impact on Actor Performances: The prospect of actors' performances being "destroyed and then remade" by AI is considered "nuts" and "absolutely crazy."

Bias in AI Training Data

The AI's consistent attempts to alter the ethnicity of the subject highlighted a critical issue: bias in AI training data. This demonstrates how AI models can perpetuate societal biases.

Comparison with Traditional Filmmaking

Comparing the AI experience with working on a traditional film set, Baroot emphasized the stark difference in control and feedback.

• Traditional Set Control: On a set, directors rely on their crew to meticulously adjust lights, costumes, and other details, engaging in constant "tweaking." This process involves detailed discussions and explanations from crew members about their choices and why they work.
• AI Lack of Control and Feedback: With AI, there is "no level of control" and "no one to really explain the rules." Directors are left feeling "lost" and without feedback, unlike the collaborative and explanatory nature of human crews.

Future Outlook: A Hybrid Model

Baroot envisions a future where AI technology is "slowly integrated into a hybrid model" alongside traditional filmmaking. The core elements of successful filmmaking – a "brilliant script, brilliant actors, and brilliantly directed" – are expected to remain paramount and unchanged.

Conclusion

The current state of AI in filmmaking is described as "a work in progress." For now, "human creativity and skill remain the gold standard."