Skip to content
Movemental Partner

Section 1: Research Support & Further Reading

Portrait of Josh Shepherd

Josh Shepherd

Founder & Architect · Movemental

title: "Section 1: Research Support & Further Reading" slug: section-1-research-and-references shape: guide author: Josh Shepherd audience: [any] status: eeat-candidate eeat_score_band: 85-99 purpose: "(1) What 2026 and recent research would support Section 1 if gathered agentically; (2) a footnoted reference list for further reading."

Section 1: Research Support & Further Reading

Section 1 scope: Where we are now (re: AI)—kairos, threat landscape, credibility crisis, trust collapse, AI as problem/solution, reframe (anthropological/adaptive), where people are, audience shift, speed of adoption, why movement leaders were right to ignore SEO, new reality (discoverability + network verification).

Part 1: Research That Would Support This Section (Agentic Search Angles)

Use these as search angles for papers, reports, videos, and core work. Each maps to Section 1’s claims or tone.

Kairos & threat landscape

  • Search: “AI threats 2023” + “Center for Humane Technology” / “The AI Dilemma” + transcript or slide deck
  • Why: The March 2023 “AI Dilemma” talk (Tristan Harris & Aza Raskin) is the reference for the “early-LLM threat list” (racing dynamics, safety gaps, existential risk framing). Good for a footnote and for naming “the list we’re still deciphering.”
  • Also: “synthetic media threats” + “deepfake” + “AI fraud” 2025–2026; “generative AI enabled fraud” projections ($40B by 2027 type figures).

Credibility crisis & “what’s real”

  • Search: “AI credibility crisis” 2025 2026; “trust collapse” + AI; “trust in AI” survey 2025
  • Key papers/reports:
    • “The Collapse of Trust in AI Assistants” (Zenodo 2025)—volatility of model outputs (e.g. 61% materially different answers across runs).
    • “Technology Trust Crisis 2026” / “$40 Billion Confidence Collapse” style reports—trust metrics and economic impact.
    • Pew Research and similar: “How Americans View AI” 2025—concern vs excitement (e.g. 50% more concerned than excited).
  • Search: “percentage of content AI-generated” 2025; “AI-generated articles” share of web
  • Key stats: Ahrefs 900k-page study (e.g. 74% of new webpages with AI content, April 2025); Graphite (e.g. >50% of new articles AI-generated); “more articles created by AI than humans” (2024/2025). Update the 40–60% and “can’t tell what’s real” with latest figures.

Human ability to distinguish human vs AI text

  • Search: “human detect AI-generated text” 2024 2025; “Turing test” + GPT-4; “AI detection” human accuracy
  • Key work:
    • Study that general participants cannot reliably distinguish GPT-4 from human (e.g. ~54% judged human, near chance).
    • PMC/NCBI “Ability of AI detection tools and humans to accurately identify…” (2025).
    • “People who frequently use ChatGPT… accurate and robust detectors” (arxiv 2501.15654)—shows experts vs general population gap.
  • Use: Support “most people can’t tell what’s real” and “effective exposure without experience.”

Volume, polish, presence (signals breakdown)

  • Search: “credibility signals” + AI; “AI replicate expertise signals”; “volume content” + AI generation rate
  • Use: Less a single paper than a synthesis—cite content-prevalence stats (above) plus one or two pieces on “AI can produce in a week what writers produce in a year” or “polish from day one.” Trade/industry reports (e.g. content marketing + AI) are useful.

Trust collapse & movement leaders

  • Search: “trust collapse” + “online” or “content”; “skepticism” + “AI content”
  • Use: Pew/survey data on concern, desire for control, and “erosion of human abilities” (creative thinking, relationships). Section 1’s “especially movement leaders” is argument, not a study—support with general trust/credibility data plus one sentence on “expertise is hard-won / credibility is relational.”

AI as both problem and solution

  • Search: “AI problem and solution” or “dual nature” AI; “using the tool that created the problem”
  • Use: Mostly conceptual; optional short footnote to a 2024/2025 think-piece or report that names the same tension. Not essential.

Anthropological & adaptive leadership reframe

  • Search: “Heifetz” + “adaptive leadership” + “technical vs adaptive”; “Leadership Without Easy Answers”
  • Canonical: Ronald A. Heifetz, Leadership Without Easy Answers (Harvard University Press, 1994).
  • Search: “AI anthropological” or “AI human flourishing” 2024 2025
  • Use: Heifetz for the technical/adaptive distinction; optional one recent paper or report framing AI as “human” or “societal” rather than purely technical.

Non-technical leaders / interaction via language

  • Search: “AI interaction” “natural language” “non-experts”; “LLM” “written verbal communication”
  • Use: Obvious enough that a single citation (e.g. UX or adoption study showing primary interaction is chat/prompt) suffices, or leave as assertion.

Where people are (overwhelmed, anxious, overconfident)

  • Search: “Pew” “AI” “concern” “excited” 2025; “Americans view AI”
  • Key: Pew Research Center, “How Americans View AI and Its Impact…” (Sept 2025); “Concern and excitement about AI” (Oct 2025). Gives 50% more concerned than excited, 10% more excited, demographics.
  • Search: “AI attitudes” “overwhelmed” “anxious” “by country” or “demographics” 2025
  • Use: Tables or one paragraph summarizing “overwhelmed, anxious, disengaged, enthusiastic, overconfident” with percentages.

Dunning–Kruger & overconfidence with AI

  • Search: “Dunning-Kruger” “AI” “overconfidence” 2025; “AI use overestimate performance”
  • Key:
    • Aalto University 2025: reversal of Dunning–Kruger with AI—higher AI literacy correlated with more overconfidence when using ChatGPT.
    • arxiv “Confidence is Not Competence” (2510.24772).
    • “AI use makes us overestimate our cognitive performance” (Tech Xplore summary of 2025 study).
  • Use: Footnote for “Dunning–Kruger on steroids” and “effective exposure without wisdom.”

Audience shift & speed of adoption

  • Search: “ChatGPT fastest growing” “100 million” “two months”; “fastest adopted technology”
  • Key: UBS analyst note (e.g. Reuters, Arstechnica): ChatGPT as fastest-growing consumer app, 100M MAU in two months; comparison to TikTok (9 months), Instagram (2.5 years).
  • Search: “ChatGPT users” 2025 2026 (e.g. 900M+ by late 2025).
  • Use: “Fastest-adopted” and “speed is an issue”; optional “velocity of language” if you have a specific piece.

Gatekeepers, SEO, and new credibility (discoverability + network)

  • Search: “gatekeepers” “credibility” “digital” “publishing”; “network verification” “credibility”
  • Use: More conceptual/historical. Optional: media or publishing report on “decline of gatekeepers” or “creator economy” + trust. “Why movement leaders were right to ignore SEO” is argument; “discoverability + network verification” can be supported by one report on trust in networks or recommendations.

Synthetic media, deepfakes, fraud

  • Search: “deepfake statistics” 2025; “generative AI fraud” “billion”; “synthetic identity” “fraud” report
  • Key:
    • Entrust 2026 Identity Fraud Report (deepfakes, social engineering, injection attacks).
    • McAfee “State of the Scamiverse” 2026 (hours spent questioning what’s real, scam volume).
    • Deepfake growth (e.g. 8M deepfake files 2025 vs 500K 2023), “one in five biometric fraud attempts,” human detection rate for deepfake video (~24.5%).
  • Use: Threat list and “we don’t know what’s real” (especially for video/voice).

Part 2: Further Reading — Footnoted Reference List

Curated so Section 1 can be footnoted and readers can go deeper. Ordered by theme to match Section 1’s flow. Format: Author/Org. Title. Source, date. [1–2 sentence note.]

A. Kairos, threat landscape, and “the list we’re still deciphering”

  1. Center for Humane Technology. The A.I. Dilemma (presentation). March 9, 2023. San Francisco. Delivered by Tristan Harris and Aza Raskin; transcript and video widely available. [Canonical reference for the early-LLM threat list: racing dynamics, safety gaps, existential risk. Use for “that slide’s list is still the list.”]

    • Video: The A.I. Dilemma – Tristan Harris & Aza Raskin, YouTube (e.g. humanetech.com, March 2023).
    • Transcript: e.g. ytscribe.com or Center for Humane Technology transcripts.

  2. Entrust. 2026 Identity Fraud Report. November 2025 (Business Wire summary). [Deepfakes, social engineering, injection attacks; deepfakes in one in five biometric fraud attempts; voice cloning as top vector.]

  3. McAfee. State of the Scamiverse 2026: AI, Deepfakes & Scams – Research Data. 2026. [Americans spend ~114 hours/year questioning if messages are real; ~14 scam messages/day; 1 in 3 lost money to scams. Supports “can’t tell what’s real” and time cost.]

  4. Deloitte. Gen AI Trust, Standards, and Deepfake Disruption. Technology, Media, and Telecom Predictions 2025. [Deepfake as “cybersecurity-scale challenge”; consequences for trust and institutions.]

B. Credibility crisis, trust collapse, and “what’s real”

  1. Pew Research Center. How Americans View AI and Its Impact on Human Abilities, Society. September 2025. [50% more concerned than excited; 10% more excited; concern about erosion of creative thinking and meaningful relationships.]

  2. Pew Research Center. Concern and Excitement About AI. October 2025. [Demographics and trends; control and oversight attitudes.]

  3. Zenodo / research. The Collapse of Trust in AI Assistants: A Practical Examination for Decision Makers. 2025 (e.g. Zenodo record 17837188). [Volatility of outputs: 61% materially different answers across identical runs; 48% shifted reasoning; 27% contradicted themselves. Supports “trust collapse” and “can’t rely on stability.”]

  4. Axis Intelligence (or equivalent). Technology Trust Crisis 2026: The $40 Billion Confidence Collapse. 2025/2026. [Trust metrics; projected generative AI–enabled fraud losses to $40B by 2027; breach costs.]

  5. ScienceDirect. The Impact of Learning About AI Advancements on Trust. 2025 (e.g. S0160791X25001484). [How information about AI advances affects trust; useful for “trust is fragile.”]

C. How much content is AI, and can people tell?

  1. Ahrefs. What Percentage of New Content Is AI-Generated? Blog/study, April 2025 (900,000 pages). [74.2% of new webpages with AI-generated content; 71.7% mixed human/AI, 25.8% pure human, 2.5% pure AI. Update “40–60%” with this or similar.]

  2. Graphite / coverage. More Articles Are Now Created by AI Than Humans; AI Content In Search & LLMs. 2024–2025 (e.g. graphite.io/five-percent). [>50% of new articles AI-generated; AI content share in search and LLM citations. Use for “flood” and “half of what you find.”]

  3. ACM / conference. People Cannot Distinguish GPT-4 From a Human in a Turing Test. 2024/2025 (e.g. ACM DOI 3715275.3732108). [General participants ~54% judge GPT-4 as human; near chance. Supports “most people can’t tell what’s real.”]

  4. PMC / NCBI. Ability of AI Detection Tools and Humans to Accurately Identify Different Forms of AI-Generated Written Content. 2025 (e.g. PMC12752165). [Human and tool accuracy; supports “can’t tell” or “detection is hard.”]

  5. arXiv. People Who Frequently Use ChatGPT for Writing Tasks Are Accurate and Robust Detectors of AI-Generated Text. 2025 (2501.15654). [Frequent users much better than general population; supports “effective exposure without experience” vs “actual use.”]

D. Deepfakes, synthetic media, fraud scale

  1. Deepfake / fraud statistics. Deepfake Statistics 2025: AI Fraud Data & Trends (e.g. deepstrike.io or equivalent). [~8M deepfake files 2025 vs 500K 2023; 3,000% spike in deepfake fraud attempts 2023; human detection rate for high-quality deepfake video ~24.5%.]

  2. Scamwatch HQ / identity fraud. Identity Fraud Report 2025–2026: Key Insights and Analysis. [Synthetic identity, first-party vs third-party fraud; growth in sophisticated fraud.]

E. Dunning–Kruger, overconfidence, and AI

  1. Tech Xplore / Aalto University. AI Use Makes Us Overestimate Our Cognitive Performance, Study Reveals. October 2025. [Summary of reversal of Dunning–Kruger with AI; higher AI literacy → more overconfidence when using ChatGPT.]

  2. arXiv. Confidence is Not Competence (2510.24772). 2025. [Confidence–competence gap with LLMs; mechanistic or empirical. Use for “overconfidence” footnote.]

  3. OpenReview / ICLR 2026. LLMs and Overconfidence (e.g. learning from failure to reduce overconfidence). 2025/2026. [If available: LLM overconfidence and failure feedback. Use for “we haven’t begun to learn.”]

F. Speed of adoption and “fastest technology”

  1. UBS (via Reuters / Arstechnica). ChatGPT Sets Record for Fastest-Growing User Base. February 2023. [100 million MAU in two months; “fastest ramp in a consumer internet app”; TikTok 9 months, Instagram 2.5 years.]

  2. Backlinko / Exploding Topics. ChatGPT Statistics 2026: How Many People Use ChatGPT? 2025/2026. [User counts e.g. 900M+ weekly active; growth timeline. Use for “fastest-adopted” and “speed is an issue.”]

  3. Pew Research Center. ChatGPT Usage by Americans (or similar). 2023–2025. [E.g. 34% of U.S. adults had used ChatGPT by June 2025 vs 18% July 2023; age breakdown. Supports “where people are” and adoption curve.]

G. Adaptive leadership (reframe)

  1. Heifetz, Ronald A. Leadership Without Easy Answers. Cambridge, MA: Belknap Press of Harvard University Press, 1994. ISBN 9780674518582. [Canonical: technical vs adaptive challenges; adaptive work requires learning, loss, and change in people and culture. Essential footnote for “therefore it’s an adaptive leadership task.”]

  2. Harvard / NCS or similar. Distinguishing Adaptive From Technical Work (toolkit or summary). [One-page summary of technical vs adaptive; useful for a short in-text gloss.]

H. Where people are (landscape and attitudes)

  1. Pew Research Center. Republicans, Democrats Now Equally Concerned About AI in Daily Life, But Views on Regulation Differ. November 2025. [Convergence of concern across parties; 50% vs 51% more concerned than excited.]

  2. Pew Research Center. How Americans See AI Impacting Human Skills; Its Role in Science, Matchmaking, Religion and More. September 2025. [Impact on human abilities; creative thinking, relationships; negative vs positive expectations.]

I. Design, citations, and trust in AI systems

  1. arXiv. GenAI Search and Trust (or similar). 2025 (e.g. arxiv 2504.06435). [Reference links/citations increase trust even when hallucinated; highlighting uncertainty can reduce trust. Use if discussing how people evaluate AI output.]

Part 3: Agentic Search Checklist (High Level)

If an agent is doing the gathering:

  1. Update stats: Replace any “40–60%” or “68%” with latest Ahrefs/Graphite/Pew/ACM-style figures (and note detection limits).
  2. Trust and credibility: Pull 2025–2026 trust/credibility and “collapse” language from Zenodo, Axis, Pew, and one academic trust paper.
  3. Threat list: Confirm Center for Humane Technology “AI Dilemma” title, date, and where to link; add one 2025/2026 fraud or deepfake report.
  4. Dunning–Kruger + AI: Get Aalto 2025 and arxiv 2510.24772 (and Tech Xplore summary) for overconfidence.
  5. Adoption speed: Get UBS/Reuters and one 2025/2026 user-count source.
  6. Heifetz: Use Leadership Without Easy Answers (1994) as the single adaptive-leadership citation unless you add a second (e.g. Heifetz & Linsky).
  7. Pew: Use 2–3 Pew titles (Sept/Oct/Nov 2025) for “where people are” and concern/excitement.
  8. Human vs AI detection: At least one Turing-style study and one “human accuracy” or “detector” paper (ACM, PMC, or arxiv).

Part 4: Suggested Footnote Map to Section 1 Statements

Rough mapping from “Section 1 full chain” to this list (by statement number):

Statement (short) Suggested refs from list
1. Kairos moment 1 (CHT), 5–6 (Pew), 7 (Zenodo)
2. Core threats still the list 1, 2, 3, 15, 16
3. Large share AI, can’t tell 10, 11, 12, 13
4. Volume, polish, presence broken 10, 11
5. Trust collapse 5, 6, 7, 8, 9
6. Movement leaders especially 5, 6 (Pew concern); rest is argument
7. AI problem and solution Conceptual; optional 7 or 8
8. Ignoring not an option, uncritical dangerous 5, 6, 7
9. Anthropological, adaptive 23, 24
10. Interact via language, non-technical can lead 23; optional UX study
11. Where people are varied 5, 6, 22, 25, 26
12. Dunning–Kruger, overconfidence 17, 18, 19
13. Audience shift March → now 22, 25 (Pew over time)
14. Fastest adopted, speed, VUCA 20, 21, 22
15. Right to ignore SEO Argument; optional gatekeeper/creator economy
16. Gatekeepers losing power, discoverability + network Optional; 11 (Graphite) for “new content” landscape
17. Adapt not abandon Conceptual

End of document.
Use Part 1 for agentic search design, Part 2 for footnotes and further reading, Part 3 for an execution checklist, and Part 4 to attach references to specific Section 1 claims.

Josh Shepherd

About the Author

Josh Shepherd

Josh Shepherd is the founder and architect of Movemental, an entrepreneur and technologist building relational credibility networks, AI-enabled publishing infrastructure, and economic sustainability systems for global movement leaders.

About Josh

The Living Archive

Occasional notes from the build. What worked, what broke, what I would do again.

An occasional letter from Josh Shepherd.