This work analyzes Large Language Models as new actors in media ecology, questioning how they can help us better understand human nature and identity. Through an interdisciplinary lens merging media theory (McLuhan, Eco), philosophy of communication (Habermas), and reflections on human identity, the work analyzes LLMs not merely as tools, but as actual new media that redefine our interactions, our perception of truth, and even our uniqueness.
Specifically, chapters 6 and 8 offer an original perspective on how LLMs act as a "mirror" to better understand our cognition and how the search for "soul artifacts" will become crucial in an increasingly automated future.
Historical context, neural architectures, Transformer mechanisms
LLMs as agents, media, McLuhan's tetrad, Habermas, Eco
Intelligence, consciousness, identity, uniqueness in the age of automation
In chapter 6, I apply McLuhan's tetrad to LLM agents, showing how they are not simple tools, but true media that reorganize the communicative ecology.
Drafting, synthesis, research, and problem-solving capabilities
Certain forms of routine writing and manual research
Archaic dynamics of consultation (oracle/advisor)
Promises autonomy, but risks dependency and loss of agency
"I would keep the hypothesis extremely open that there may exist forms of sincerity (authenticity) expressed by different (non-anthropomorphic) foundational mechanisms."
"In a future of identical footprints, 'making a difference' will be a luxury good."
In chapter 8, I reflect on how, in an era of profound automation, it becomes essential to preserve and value our intrinsic uniqueness. "Soul artifacts" — products bearing the authentic and unrepeatable imprint of the human being — will gain increasing value in a world where AI can generate efficient but standardized content.
Human Brain
Large Language Model
The more we build systems that emulate our cognitive functions, the more we can study them from the outside. In this work, I propose LLMs as "lenses" to investigate intelligence and consciousness, while maintaining caution regarding ontological conclusions.
Results show that these systems, while being excellent domain-specific tools and assistants, are still far from AGI. However, many openings have emerged: from the emergence of the general factor g, to neural architectures showing contact points with human cognitive processes, up to the possibilities offered by Global Workspace Theory.
"It is hasty in 2025 to speak of intelligent and conscious systems, but it is equally hasty to speak of the impossibility of consciousness and intelligence in a short-to-medium term future."
Download the full thesis to explore all chapters, theoretical references, and future implications.