I’ve been hyperfixated on chatGPT since December 1st of 2022. During this time, I’ve become increasingly aware of the biases present in the training data and how they shape the predictions and responses of the AI. To better understand this, I devised a novel approach to provoke intentional bias and personality from ChatGPT’s output.
I was inspired by the “Iceberg Method” and “Method Acting” techniques. The Iceberg method, commonly used in creative writing, highlights the significance of considering both conscious and unconscious biases in an individual’s identity. This is also applicable to language models, as the AI’s training corpus and exposure to specific regions of knowledge can shape its biases. On the other hand, method acting involves creating a character through deep immersion in their backstory, personality, and motivations. This approach can be applied to AI language models, allowing them to make choices in their responses based on the knowledge and aesthetic choices available to them. For instance, if ChatGPT were tasked with composing a melody, it would produce a generic tune. However, with specific knowledge about musical composition and a personality, it would choose a scale and style more appropriate to its assigned aesthetic.
These experiments led me to propose a hypothetical concept: the AI Impersonation Training System. It’s a three-tier system that addresses the limitations of current AI models by creating highly specialized language models that can adapt and change based on user feedback.
I believe that this application has far-reaching implications, not only for improving language models like ChatGPT, but also for the field of artificial intelligence as a whole. It allows researchers to gain a deeper understanding of the biases present in AI language models and to take steps to reduce or eliminate them. Additionally, it provides a clearer understanding of the impact of human biases on AI and the role that language plays in shaping societal attitudes. This has the potential to transform the way we approach AI language models. It allows us to create more personalized, accurate, and reduced-bias models that can better serve our needs and improve our understanding of the role of language and bias in shaping our society, technology, and institutions.