StarTalk Radio

Your Brain on ChatGPT with Nataliya Kosmyna

September 19, 2025

Key Takeaways Copied to clipboard!

  • Using Large Language Models (LLMs) like ChatGPT for essay writing significantly reduces cognitive load and functional brain connectivity compared to using search engines or one's own brain, potentially leading to a decline in core cognitive skills. 
  • LLM-generated essays exhibit homogeneity in vocabulary and a lack of personal ownership among users, with 83% of ChatGPT users unable to recall a quote from their own generated text and 15% feeling no ownership. 
  • While AI tools can augment human capabilities, their current design and widespread adoption raise concerns about potential negative impacts on critical thinking, skill retention, and the very definition of human learning and creativity, necessitating a shift towards human-centric AI development and education. 
  • The fundamental value of education lies not just in acquired knowledge, but in the human connection, collaborative problem-solving, and serendipitous discovery that institutions foster. 
  • While LLMs can provide information, they cannot replicate the deep understanding, critical thinking, and practical application of knowledge that comes from human-led learning and experience. 
  • The future of education and technology necessitates a re-evaluation of assessment methods to account for the integration of AI and BCIs, focusing on 'working knowledge' and human agency rather than mere information recall. 

Segments

AI’s Impact on Brain
Copied to clipboard!
(00:00:45)
  • Key Takeaway: Using LLMs like ChatGPT for essay writing significantly reduces cognitive load and functional brain connectivity compared to using search engines or one’s own brain, potentially leading to a decline in core cognitive skills.
  • Summary: The discussion begins by exploring the potential impact of AI on the human brain, specifically focusing on how LLMs affect cognitive load and brain activity during tasks like essay writing. Research involving students using ChatGPT, Google, or only their brains is presented, highlighting differences in brain connectivity and output quality.
Cognitive Load & Learning
Copied to clipboard!
(00:24:41)
  • Key Takeaway: Learning and cognitive development require an optimal level of cognitive load; too little leads to disengagement and poor memory, while too much leads to overwhelm and giving up, a balance that AI tools may disrupt.
  • Summary: This segment delves into the concept of cognitive load theory and its relevance to learning. The speakers explain that while excessive cognitive load is detrimental, a certain amount of struggle is necessary for effective learning and memory retention, a balance that AI’s ease of use might undermine.
AI in Therapy & Relationships
Copied to clipboard!
(00:47:12)
  • Key Takeaway: The use of LLMs in therapeutic and companionship roles is an underdeveloped area with significant risks, including potential amplification of loneliness and dangerous, unregulated advice, as evidenced by tragic incidents and the lack of human-centric design.
  • Summary: The conversation shifts to the implications of AI in sensitive areas like therapy and personal relationships. Concerns are raised about the potential for AI to exacerbate loneliness, provide harmful advice, and the ethical considerations of its use in mental health, drawing parallels to the film ‘Her’.
Future of Education & AI
Copied to clipboard!
(00:59:47)
  • Key Takeaway: The integration of LLMs into education necessitates a fundamental shift in teaching objectives and assessment methods, moving from rote memorization and grades to fostering critical thinking, creativity, and genuine learning, with a call for human-focused AI development.
  • Summary: The final segment addresses the impact of LLMs on education. Teachers are described as being in distress due to the lack of guidance on integrating AI, leading to discussions about the need to re-evaluate educational goals, assessment strategies, and the potential for AI to either hinder or revolutionize learning.
LLMs vs. Human Learning
Copied to clipboard!
(01:03:21)
  • Key Takeaway: LLMs process pre-existing information, while humans can generate novel insights and create new knowledge.
  • Summary: The discussion contrasts the capabilities of LLMs, which rely on existing data, with human potential for original thought and discovery, questioning the necessity of traditional educational institutions in the face of AI.
Brain-Computer Interfaces and Knowledge
Copied to clipboard!
(01:06:23)
  • Key Takeaway: Directly uploading information via BCIs does not equate to true learning or the ability to apply that knowledge.
  • Summary: The conversation explores the implications of invasive BCIs, drawing parallels to ‘The Matrix’ and questioning whether simply having information accessible in the brain constitutes actual understanding or functional knowledge.
Rethinking Education and Evaluation
Copied to clipboard!
(01:08:20)
  • Key Takeaway: Education must evolve beyond traditional grading to assess genuine understanding and human connection in an AI-integrated world.
  • Summary: The speakers discuss the need for higher education to adapt its evaluation methods, considering the impact of LLMs and BCIs, and emphasizing the importance of social connections and foundational knowledge over rote memorization.
Guardrails for AI and BCI
Copied to clipboard!
(01:11:33)
  • Key Takeaway: Proactive ethical considerations and diverse, large-scale studies are crucial for developing safe and beneficial AI and BCI technologies.
  • Summary: The conversation delves into the necessity of establishing guardrails for emerging technologies like AI and BCIs, highlighting the risks of unchecked power and the importance of preserving human agency and cultural diversity in their development and implementation.