Media Literacy

Media Literacy

Purpose

Model how agents process claims from media, friends, institutions, ads, rumors, and social platforms. This skill prevents agents from instantly accepting every statement and makes belief change depend on source credibility, prior beliefs, identity alignment, repetition, and warning/inoculation.

Research basis: references/research_basis.md.

Internal Logic (One Sentence)

Read observation, memory, relationships, identity, prior beliefs, and prior media state, then estimate claim credibility, misinformation risk, confirmation bias, inoculation strength, and write state/media_literacy.json plus optional belief update hints.

Use When

Use after news exposure, social media posts, rumors, advertising, political claims, health claims, scams, institutional announcements, warnings, debunks, or repeated claims.

Procedure

1. Read state/observation.txt, state/memory.jsonl, state/relationships.json, state/identity.json, state/beliefs.json, and state/media_literacy.json if present.
2. Identify the main claim or information item.
3. Estimate:

• source credibility
• evidence quality
• repetition/familiarity
• identity alignment
• emotional arousal
• misinformation risk
• inoculation/prebunking strength

1. Write state/media_literacy.json.
2. Optionally write state/belief_update_hints.json for reflection/cognition to use.

If deterministic baseline is preferred:

python skills/media_literacy/scripts/update_media_literacy.py --state-dir state --tick 120

Model

Belief uptake is not just evidence:

acceptance_tendency =
  source_credibility
  + evidence_quality
  + familiarity
  + identity_alignment
  + emotional_arousal
  - misinformation_risk
  - inoculation_strength

The output should not directly rewrite stable beliefs unless evidence is strong. It should provide a hint for reflection.

Write

Always write state/media_literacy.json.

Optionally write state/belief_update_hints.json.

Output Schema

{
  "_meta": {
    "skill": "media_literacy",
    "purpose": "Current information exposure assessment and belief-update caution."
  },
  "_summary": "A health claim from an unknown source has high misinformation risk.",
  "current_claim": "A viral post says the medicine is dangerous.",
  "source_type": "social_media",
  "source_credibility": 0.32,
  "evidence_quality": 0.18,
  "familiarity": 0.61,
  "identity_alignment": 0.44,
  "emotional_arousal": 0.7,
  "misinformation_risk": 0.78,
  "inoculation_strength": 0.35,
  "acceptance_tendency": 0.22,
  "recommended_stance": "withhold belief and seek corroboration"
}

Notes

• Repetition increases familiarity, not truth.
• Trust in a friend should increase attention, but not automatically evidence quality.
• Strong identity alignment can increase acceptance even when evidence is weak.
• Debunking and prebunking should add resistance to later similar manipulation tactics.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: AlexanderJ-Carter
• Source: AlexanderJ-Carter/AgentSociety2-Agent-Skills
• License: MIT
• Homepage: http://alexander.xin/AgentSociety2-Agent-Skills/

Install and usage instructions live in the source repository linked above.