Every claim, linked to its source.

Public opinion & usage

• Pew Research Center — "Key findings about how Americans view AI" (March 2026). The 64%/jobs figure, the 10%/more-excited figure, and the broader picture of the public-expert divide on AI. Read the findings →
• Pew Research Center — "34% of US adults have used ChatGPT" (June 2025). The adoption and demographic breakdown cited in the FAQ. Survey of 5,123 US adults, Feb–March 2025. Read the piece →
• Pew Research Center — "Views of risks, opportunities, and regulation of AI" (April 2025). The public-expert divide on AI's economic impact. Read the study →
• Gallup — Gen Z AI adoption & skepticism (2026). The shift in Gen Z sentiment: "excited" 36%→22%, "angry" 22%→31%. Conducted for Walton Family Foundation and GSV Ventures. Read the poll →
• Stanford HAI — AI Index Report 2026 (Public Opinion chapter). The canonical annual snapshot of public sentiment on AI. Read the chapter →

Energy, cost & environment

• International Energy Agency — "Energy and AI" report (2025). The 2% of global electricity figure, the comparison to EVs/A/C/industrial demand, and AI's share within data centers (5–15% today, possibly 35–50% by 2030). Read the report →
• Epoch AI — "How much energy does ChatGPT use?" Research-grade analysis of per-query energy. The source for the 0.3 Wh–19 Wh range depending on model. Read the analysis →
• Carbon Brief — "Five charts that put data-centre energy use and emissions into context." Visual comparison of AI's energy share against other sectors. Read the piece →
• "The Rising Costs of Training Frontier AI Models" — Cottier et al., arXiv 2024. Documented training costs: GPT-4 ~$78M, Gemini Ultra ~$191M, Llama 3.1 405B ~$170M. Read the paper →
• Lawrence Berkeley National Laboratory — 2024 United States Data Center Energy Usage Report. The authoritative Congressional report on US data center energy and water consumption. Source for the 17 billion gallons direct / 211 billion gallons indirect figures for 2023. Read the report →
• "Making AI Less Thirsty" — Shaolei Ren et al., UC Riverside, 2023. The leading independent research on AI's water footprint. Covers both direct cooling water and indirect water consumed through electricity generation. Read the paper →
• Environmental and Energy Study Institute — Data Centers & Water Consumption. Policy-oriented synthesis of the sector-by-sector comparisons (agriculture, residential, industrial). Useful for proportion-framing. Read the brief →

Deepfakes & misinformation

• Sumsub — Identity Fraud Report 2025–2026. The source for deepfake volume growth (UK government projection of 8M in 2025, up from 500,000 in 2023) and the 700% jump in deepfake-powered fraud. Released November 2025. Access the report →
• World Economic Forum — "How cognitive manipulation and AI will shape disinformation in 2026." Context on the systemic impact of deepfakes and synthetic media on elections and public discourse. Read the article →
• FBI Internet Crime Complaint Center (IC3) — Public Service Announcements on AI-enabled fraud (2024). Series of advisories on voice-cloning scams, deepfake video conferences, and AI-polished phishing. Authoritative law-enforcement source for what's actually being reported. PSA 2024 →
• Hong Kong Police / SCMP — "Multinational firm Hong Kong loses HK$200 million after deepfake video meeting" (Feb 2024). The canonical case study: a finance employee wires ~US$25M after a video conference where every "colleague" was a deepfake. Cited throughout our Spot the Fake page. Read the case →
• Kamali, Black, Lin, Groh et al. — "How to Distinguish AI-Generated Images from Authentic Photographs" (arXiv, 2024). Northwestern's five-category taxonomy of AI-image artifacts (anatomical, lighting, texture, functional, sociocultural). The framework behind the "spot the tells" exhibits across this site. Read the paper →
• Hany Farid — UC Berkeley. Image-forensics researcher and pioneer of digital-image authentication; lighting and shadow analysis is the foundation of AI-image detection. Faculty page lists ongoing publications. Faculty page →

Labor & work changes

• McKinsey Global Institute — "Generative AI and the Future of Work in America." Base research on AI-driven task automation, workforce displacement patterns, and reemployment timelines. Read the research →
• World Economic Forum — "Future of Jobs Report 2025." Base figures on job creation vs. displacement projections (170M new / 92M displaced between 2025 and 2030). Read the report →

Brains, cognition & AI use

• MIT Media Lab — "Your Brain on ChatGPT" (June 2025). EEG study on cognitive engagement during LLM-assisted writing. 54 subjects, three groups (LLM / search / brain-only). LLM users showed weakest brain connectivity and lowest self-reported ownership of their work. Note: preprint at release; small sample size. Read the paper →
• Sparrow, Liu & Wegner — "Google Effects on Memory," Science, 2011. The foundational paper on how search engines change what we remember. DOI: 10.1126/science.1207745. Access the paper →
• Maguire et al. — "Navigation-related structural change in the hippocampi of taxi drivers," PNAS, 2000. The classic London cab driver study on how intensive spatial memory use changes brain structure. Access the paper →
• Lee et al. — "The Impact of Generative AI on Critical Thinking" (Microsoft Research / Carnegie Mellon, ACM CHI 2025). Peer-reviewed survey of 319 knowledge workers. Self-reported cognitive effort dropped when using GenAI; higher AI confidence correlated with reduced critical evaluation. The strongest peer-reviewed AI-specific evidence to date. Read the paper →

Teens, AI, and mental health

• APA Health Advisory — AI and Adolescent Well-Being (2026). The American Psychological Association's formal guidance for parents and clinicians. Covers emotional-dependency signs and when to seek professional help. Read the advisory →
• Common Sense Media — "Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions" (2025). The most-cited research on how teens actually use AI companions and what they say they get from them. Read the report →
• The Jed Foundation — Response to APA advisory. JED is a youth mental-health nonprofit. Practical framing for parents. Read →
• Child Mind Institute — "AI Chatbots and Teens". Parent-friendly explainer with age-specific guidance. Read →
• 988 Suicide & Crisis Lifeline (US). Call or text 988. Free, confidential, 24/7. Chat at 988lifeline.org/chat.

How AI actually works

• Brown et al. — "Language Models are Few-Shot Learners" (arXiv, 2020). The foundational GPT-3 paper — establishes how modern LLMs are built and what they're doing under the hood. Access the paper →
• Bender, Gebru, McMillan-Major & Shmitchell — "On the Dangers of Stochastic Parrots" (ACM FAccT 2021). The canonical argument for why LLMs are sophisticated pattern-matchers, not minds. Access the paper →
• Vectara Hallucination Leaderboard. A continuously-updated benchmark of how often various AI models make up facts. Live, public, open-source. View the leaderboard →
• Stanford HAI — AI Index Report 2026. The annual comprehensive snapshot of the state of AI, including benchmarks, hallucination data, public opinion, and economic effects. Read the report →

Educators & explainers worth following

These aren't the primary sources behind specific claims on this site — those are above. These are accessible, ongoing public-education resources you can subscribe to if you want to keep learning past what we cover here.

• Prof. Casey Fiesler — Information Science, CU Boulder. Short-form videos on how language models work, AI ethics, and tech policy. A good next step if you want a slightly more research-flavored treatment of the topics we cover here. YouTube playlists →
• Prof. Mark Riedl — Georgia Tech. Long-running blog posts that explain LLM internals (embeddings, self-attention, training stages) for non-CS readers. Often-cited entry point for the technical-but-accessible register. Medium →
• AI4K12 Initiative. The "Five Big Ideas in AI" framework used by K-12 educators in the US. Useful if you want to teach this material in a classroom. ai4k12.org →
• Prof. Hany Farid — UC Berkeley. Image-forensics researcher; lighting/shadow analysis for AI-image detection. Underlying methodology for several of our deepfake-detection exhibits. faculty page →

Find an error or a better source? Email robert@completeideas.com. I'll fix it.