Learning and performance with GenAI | Franck Nassiri | TEDxEDHECBusinessSchool

Key Concepts:

• Industrial Revolutions (Steam Power, Electricity, Digital Technologies)
• Artificial Intelligence (AI) as the "new electricity"
• Generative AI (GenAI) and Large Language Models (LLMs)
• General Purpose vs. Narrow Purpose GenAI
• AI Hallucinations
• Academic Integrity and Authentic Learning
• Over-reliance on AI
• Augmented Humans
• Expertise Levels (Humans and AI)
• AI-Resistant Assessment Systems

1. Industrial Revolutions and the Rise of AI

• The speaker begins by reflecting on their career shift from semiconductor engineering to teaching strategy, highlighting the increasing importance of technology, particularly GPUs, in modern economies.
• The video discusses the past three industrial revolutions:
  • 1st: Steam power (18th century)
  • 2nd: Electricity (19th century)
  • 3rd: Digital technologies, personal computers, and the internet
• Each revolution brought progress and challenges, including initial fears and anxieties (e.g., electricity as an "invisible killer").
• Andrew Ng's statement: "AI is the new electricity," emphasizing its invisible, pervasive, and essential nature.
• AI's impact on productivity: augmenting knowledge workers and automating repetitive tasks.
• AI as a source of innovation, driving industries like intelligent autonomous systems and personalized education.

2. Generative AI and Large Language Models

• Generative AI applications are powered by Large Language Models (LLMs) trained on massive datasets.
• General Purpose GenAI (e.g., ChatGPT): Trained on publicly available data, capable of discussing various topics but not specialized.
• Narrow Purpose GenAI: Trained on specific datasets for expertise in a particular field (e.g., higher education), resulting in higher accuracy and reliability.
• GenAI applications learn patterns and structures of languages rather than "understanding" them.
• Example: GenAI doesn't "know" what a hamburger is but learns patterns associated with the term.
• When prompted, AI generates answers word by word, predicting each word based on the prompt and learned patterns.

3. Limitations and Weaknesses of Generative AI

• Accuracy of ChatGPT varies from 7% to 86% depending on the field and user guidance.
• Human Biases: Data used to train AI can contain biases.
• Misinformation: Data can contain misinformation.
• AI Hallucinations: AI can generate incorrect or nonsensical information with confidence.
  • Example: AI incorrectly stating that Albert Angen discovered oxygen in 1921.

4. Generative AI in Higher Education

• Researchers and faculty use GenAI for content generation, structuring, writing, coding, creating assignments, and prompting discussions.
• Students have embraced GenAI tools.
• Challenges:
  • Academic Integrity: Difficulties in preventing misuse and protecting academic integrity.
  • Authentic Learning: Difficulties in ensuring genuine learning experiences.
• Some institutions reported higher rates of academic dishonesty.
• Trend: Institutions developing AI-resistant assessment systems.

5. Impact on Learning and Skill Development

• Research focuses on using AI to augment, not replace, human capabilities.
• Concern: Offloading cognitive effort to AI can hinder skill development.
• Using GenAI as a shortcut (copy-pasting answers) prevents the development of critical thinking and problem-solving skills.
• More effective approach: Analyzing and breaking down AI-generated outputs to understand them.
• Shift in emphasis: From writing solutions from scratch to critically analyzing AI-generated outputs.

6. Consequences of Over-Reliance on AI

• Using GenAI as a shortcut can lead to students being underprepared for the workforce, which requires independent thinkers.
• Personal ethics and integrity are compromised when convenience is prioritized over genuine effort.
• Erosion of trust in academic and societal contexts.

7. Boston Consulting Group (BCG) Study: Pairing Humans and AI

• Two key factors for successful pairing:
  • Expertise level of the GenAI tool.
  • Expertise level of the human user.
• Key Takeaways:
  • Skillful Humans + High-Expertise AI: Augmented humans outperform humans working alone and AI working solo.
  • Skillful Humans + Moderate-Expertise AI: Over-reliance on AI can lead to worse performance than working alone.
  • Moderate-Expertise Humans + Moderate-Expertise AI: Skill development is possible with expert supervision.

8. Specialized GenAI Applications in Education

• Trend: Institutions developing narrow purpose GenAI applications integrated into learning platforms.
• Benefits:
  • Training AI models with specific academic datasets for higher accuracy.
  • Offering more conceptualized and authentic learning experiences.
  • Minimizing misinformation, human biases, and AI hallucinations.
  • Guiding students step-by-step through solutions for better learning.

9. Conclusion

• Caring for the future is a shared responsibility of educators, institutions, and humanity.
• The goal is to develop future leaders effectively.

Key Quotes:

• Andrew Ng: "AI is the new electricity."