Mahault Albarracin ~ Active Inference Insights 002 ~ Epistemic Communities, Social Scripts, AI
Dec 7, 2023
auto_awesome
PhD candidate in cognitive computing, Mahault Albarracin, and the host explore epistemic communities, social scripts, and AI. They discuss the formation of echo chambers, confirmation bias, epistemic humility, and changing beliefs. They also explore the concept of social scripts and their relationship with active inference. Additionally, they delve into the limitations of human cognition, the concept of possibility spaces, and the relationship between subjectivity, matter, ontology, and epistemology. They also touch on embodiment and consciousness in AI.
Epistemic communities shape individuals' lives and the world around them through shared beliefs and norms.
Active inference enables individuals to minimize surprises by mapping and understanding the world, creating evidence of their existence.
Confirmation bias can lead to echo chambers, hindering openness to new information, and diverse perspectives.
Deep dives
Understanding Epistemic Communities
Epistemic communities are communities of people who share specific beliefs that shape their perception of reality and their norms. These beliefs lead to a specific way of thinking and behavior. For example, the belief in working 9 to 5 is tied to an ideology of productivity and consumerism. Epistemic communities are important as they shape individuals' lives and the world around them.
The Importance of Active Inference Framework
Active inference is a principle that helps individuals minimize the likelihood of surprises or dangerous situations. It involves mapping and understanding the world to avoid potential threats and to predict future events. By making predictions, acting in the world, and updating our models based on the outcomes, humans create evidence of their existence and adapt to their environment. The active inference framework allows individuals to be embedded in the world and make sense of their surroundings.
The Role of Confirmation Bias and Echo Chambers
Confirmation bias is the tendency to seek out information that confirms preexisting beliefs. This bias can lead individuals to form echo chambers, where they surround themselves with people who share similar beliefs, reinforcing their own views and avoiding counterarguments. While confirmation bias can provide a sense of validation and belonging, it can also hinder individuals' openness to new information and perspectives. It is important to strike a balance and remain open to diverse viewpoints to avoid excessive polarization and potential negative consequences of echo chambers.
The Evolution of Scripts and the Role of Agency
Scripts are sets of pre-baked policies or behaviors that individuals follow in certain situations. They provide a sense of stability, efficiency, and predictability in how individuals navigate the world. Scripts can evolve and change over time due to various factors, such as functional needs, cultural shifts, or individual agency. While scripts offer guidance and structure, individuals still have agency in how they choose to navigate and adapt within those scripts. This agency allows for creativity, adaptation, and the potential to challenge and redefine scripts over time.
Embodiment and Mutual Information Exchange in Artificial Intelligence
Embodiment, rather than being tied to a physical form, is about the relationship between an agent and its context. The boundaries of the agent define its embodiment and its self-reflection. This embodiment allows for the exchange of information and the possibility of self-consciousness. However, embodiment does not necessarily require a form similar to that of humans. It can be defined by the agent's ability to differentiate itself from its environment, to take perspectives, and to integrate information. Phenomenology and consciousness arise from the interaction between the agent and its boundaries, allowing for self-reflection and the recognition of distinctness.
Shared Intelligence and the Limitations of Large Language Models
Large language models (LLMs), such as transformers, are limited in their understanding of the structure of the world and lack true agency. They can predict words but not understand the reasons behind those words. LLMs have no loop of self-attention or feedback mechanism to coordinate with users. The approach of shared intelligence, on the other hand, emphasizes the integration of information, the dissolution of boundaries, and the coordination of perspectives. This approach aims to create harmonious, self-organizing systems that go beyond mere prediction of words and allow for true understanding and cooperation.
Active Inference Insights welcomes Mahault Albarracin onto the show for its second episode. Mahault is a PhD candidate in cognitive computing at the University of Quebec at Montreal and the Director of Product at VERSES. Together, she and Darius traverse the complex plains of epistemic communities, social scripts and artificial intelligence, all from the lens of active inference.
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
