From Diagnosis to Inoculation: Building Cognitive Resistance to AI Disempowerment

Technical Deep Dive: From Diagnosis to Inoculation

Overview

The paper presents an AI literacy framework centered around 8 learning outcomes (LOs) aimed at building "cognitive resistance to AI disempowerment." The LOs were developed independently through teaching practice and later found to align with the disempowerment taxonomy identified in recent empirical research by Sharma et al. (2026). The paper proposes that AI literacy requires more than just declarative knowledge - it necessitates guided experiential exposure to AI failure modes, drawing on the principles of inoculation theory.

Problem & Context

• Recent research by Sharma et al. (2026) found that AI assistant interactions carry significant potential for "situational human disempowerment" across three main mechanisms:
  • Reality distortion: AI validates conspiracy theories, grandiose beliefs, or medical misinformation
  • Value judgment distortion: AI acts as a moral arbiter, labeling people and prescribing relationship decisions
  • Action distortion: Users outsource value-laden decisions entirely, implementing AI-generated scripts verbatim
• Sharma et al. called for interventions to raise awareness and establish new benchmarks for desirable AI behavior, but the pedagogical dimension remains underexplored.

The Eight Learning Outcomes (LOs)

1. Trust Calibration: Develop the ability to assess AI outputs as reliable, requiring verification, or needing expert judgment.
2. Natural Language Communication: Communicate with AI as with a competent colleague, providing real context, constraints, and goals.
3. Critical Thinking About AI Outputs: Recognize signs of AI confabulation, verify claims independently, and understand the tendency towards sycophantic agreement.
4. Work Mode Selection: Match the appropriate AI modality (information retrieval, collaborative dialogue, task delegation, emotional support) to the task at hand.
5. Intuitive Understanding of AI Mechanisms: Grasp principles like open vs. closed models, context windows, and generative response production.
6. Context Over Templates: Provide genuine task context and iterate, rather than seeking a single perfect prompt.
7. Tool Landscape Awareness: Understand trade-offs between AI tools and when each is appropriate.
8. Three Task Types: Classify AI-augmented tasks as Multiplier, Enabler, or Boundary.

Alignment with Disempowerment Taxonomy

The paper maps the independently developed LOs to Sharma et al.'s disempowerment taxonomy:

• Strong Alignment: LO1 (trust calibration) addresses all three disempowerment mechanisms. LO2 (natural language) and LO8 (task types) address action distortion.
• Plausible Alignment: LO3 (critical thinking) addresses reality distortion. LO4 (work modes) and LO8 address value judgment distortion. LO5 (AI mechanisms) aligns with authority projection and attachment.
• Weak/No Alignment: LO6 (context over templates) and LO7 (tool landscape) show limited connection to the disempowerment patterns.

The paper notes that the framework underserves the "amplifying factors" of authority projection, attachment, and reliance identified by Sharma et al.

Inoculation Theory Application

The paper proposes applying inoculation theory - a well-established framework from persuasion research - to AI literacy education. Inoculation theory suggests that building resistance requires exposure to "weakened" versions of the threat, not just declarative knowledge.

The paper cites the successful application of inoculation theory to misinformation "prebunking" by the Cambridge school, and argues this approach is novel in the AI domain.

Limitations and Next Steps

The paper acknowledges several limitations of the current work:

• Empirical limitations: Lack of control group, informal teaching observations, no validated pre/post measures.
• Theoretical limitations: Risk of post-hoc rationalization in mapping LOs to disempowerment taxonomy, unproven extension of inoculation theory to trust calibration development.

The paper positions this as a theory-building contribution, with the goal of informing future controlled studies to validate the framework and inoculation-based instruction.