In this episode of Local Disturbances we delve into the complexities of AI development, language modelling, and the exploration of different historical and cultural perspectives. The team at UKAI is pushing the boundaries by attempting to train AI bots on specific historical periods, aiming to bridge the gap between literature, culture, and AI.
Our focus lies on the transition from Medieval to Renaissance thought in 16th century Western Europe, a pivotal period that birthed the modern novel. By training language models on this narrow cosmology, we hoped to offer audiences a unique window into historical ways of knowing the world.
However, our technical choices have posed challenges. While symbolic AI allows for the introduction of high-level rules and representations, we are working with deep learning approaches. Despite our efforts, the resulting AI-generated text, while intriguing, lacked evidence of tapping into the deeper layers of ancient collective thought.
The discussion touches upon how language and perception are intertwined, drawing parallels between linguistic shifts and changes in environmental awareness. Just as languages evolve to reflect our surroundings, so too do our conceptual metaphors evolve to explain our experiences.
The conversation delves into the notion of chronotypes, or perceived architectures of time and space, as described by Mikhail Bakhtin. Different historical periods embody distinct chronotypes, shaping the narratives and stories of their time.
The episode also explores the concept of the road as a chronotope, highlighting how modern transportation has shifted our perception of space and time. Through literature and language modelling, the team aims to capture and explore these diverse chronotypes, offering insights into alternative ways of understanding reality.
Yet, the challenge remains in reconciling these diverse perspectives with the underlying assumptions of modernity embedded within AI systems. Despite progress, the subsoil of ancient collective thought remains elusive, challenging the very foundations of AI development.
As the episode concludes, it prompts us to question the limitations of synthetic language and the need to explore alternative approaches to modelling reality. In a world dominated by large language models, the quest for new conceptual metaphors and diverse perspectives becomes ever more crucial.
Produced by Kasra Goodarznezhad
Sound by Koohyar Habibi
Words by Jerrold McGrath