"All we want are the facts, ma'am." — Sgt. Joe Friday, Dragnet
Artificial intelligence is everywhere — in your phone, your doctor's office, your kid's homework, and increasingly, your Sunday bulletin. But between Hollywood's killer robots and Silicon Valley's utopian press releases, getting a straight answer about what AI actually is, what it actually costs, and what it actually does can feel like nailing jelly to a wall.
So here it is. No spin. No screenplay. Just the facts — organized, in the spirit of the late great Clint Eastwood, into The Good, The Bad, and The Ugly.
(Pull up a chair. Squint accordingly.)
🤠 First: What Even IS Artificial Intelligence?
| Term | What It Actually Means | Real-World Example |
|---|---|---|
| AI (Artificial Intelligence) | Software that performs tasks normally requiring human intelligence | Your email spam filter. That's AI. Has been since the 2000s. |
| Machine Learning (ML) | AI that improves by studying patterns in data, without being explicitly programmed for each task | Netflix recommendations. Spotify's "Discover Weekly." |
| Large Language Model (LLM) | An AI trained on massive amounts of text to predict and generate language | ChatGPT, Claude, Gemini, Llama |
| Parameters | The "knobs" the model learned to tune during training — more parameters generally means more capable (but not always) | GPT-4: ~250 billion parameters. Meta's Llama 4 Behemoth: nearly 2 trillion. |
| Training | The process of feeding the model data so it learns patterns. Done once (expensively). | Training GPT-4 cost an estimated $78–100+ million in compute alone (Juma AI). |
| Inference | When the trained model actually answers your question. Done billions of times daily. | Every ChatGPT query. Over 2.5 billion prompts per day as of early 2026 (DemandSage). |
| Hallucination | When the AI confidently generates information that is factually wrong | The lawyer who cited fake court cases generated by ChatGPT (Mata v. Avianca, 2023). |
| AGI (Artificial General Intelligence) | A hypothetical AI that can do any intellectual task a human can. Does not exist. | The thing people are afraid of. Still science fiction as of February 2026. |
📊 By the Numbers: AI in 2026
| Metric | Number | Source |
|---|---|---|
| ChatGPT weekly active users | 900 million | TechCrunch / OpenAI |
| Daily prompts processed by ChatGPT | 2.5 billion | DemandSage |
| Global AI market size (2025) | ~$391 billion | Grand View Research |
| Cost to train GPT-5 (single run) | $500+ million | Fanatical Futurist |
| U.S. data center electricity (2024) | 183 TWh (4%+ of U.S. total) | Pew Research |
| U.S. data center water use (2023) | 17 billion gallons | EESI |
| Americans more concerned than excited about AI | 50% (up from 37% in 2021) | Pew Research |
| Americans who distrust businesses AND government to use AI responsibly | 77% | Gallup |
✅ The Good
"You see, in this world there's two kinds of people, my friend: those with loaded guns and those who dig." — Blondie (The Good, the Bad and the Ugly, 1966)
AI is not all doom and server farms. Here is what it is genuinely doing well — with receipts.
🩺 Healthcare: Finding What Human Eyes Miss
- AI-powered lung cancer screening systems have reached 90% diagnostic accuracy and have screened more than 180,000 imaging scans by mid-2025, according to the National Cancer Institute and OncoDaily.
- In China, Alibaba's DAMO GRAPE gastric cancer system achieved 85.1% sensitivity and 96.8% specificity — outperforming human radiologists — and is deployed across Zhejiang and Anhui provinces (BioSpectrum Asia).
- 80% of FDA-approved AI medical devices in oncology focus on diagnostics (PMC/NIH).
- Google's DeepMind solved the 50-year protein folding problem in 2024, accelerating drug discovery research worldwide.
🌍 Accessibility and Translation
- AI translation tools now support hundreds of languages, including many with fewer than a million speakers. Google Translate serves over 1 billion translations daily.
- In sub-Saharan Africa, AI is being deployed for agricultural extension, maternal health screening, and multilingual education — according to the Stanford HAI 2025 AI Index.
- AI-powered captioning and screen readers are making digital content accessible to an estimated 1.3 billion people living with disabilities worldwide (WHO).
🌊 Disaster Response and Climate Science
- Google's FloodHub uses machine learning to forecast flooding up to five days in advance in more than 80 countries.
- AI climate models are improving weather prediction accuracy, enabling earlier evacuation orders and disaster response.
- Machine learning algorithms process satellite imagery to track deforestation, ice melt, and wildfire spread in near-real-time.
💼 The Jobs Picture (It's More Complicated Than You Think)
| Metric | Number | Source |
|---|---|---|
| Jobs projected to be displaced by 2030 | 92 million | World Economic Forum |
| New jobs projected to emerge by 2030 | 170 million | World Economic Forum |
| Net new jobs | +78 million | WEF calculation |
| Workers who may need to change careers by 2030 | 14% globally | McKinsey via Gloat |
Important caveat: As the World Economic Forum notes, these aren't direct swaps. The jobs disappearing (clerical, administrative) and the jobs appearing (data specialists, green energy, healthcare) aren't in the same cities, same industries, or same skill brackets. The net number is positive. The transition will be painful for millions.
⚠️ The Bad
"You see, there are two kinds of spurs, my friend. Those that come in by the door... and those that come in by the window."
🔋 The Energy Appetite
| What | Energy | Context |
|---|---|---|
| One ChatGPT query | ~0.3 Wh | According to Epoch AI — roughly what a Google search used in 2009 |
| One Google search (current) | ~0.3 Wh | Google hasn't updated its figure since 2009. With AI Overviews, current estimates suggest 1.7x more |
| U.S. data center electricity (2024) | 183 TWh | 4%+ of all U.S. electricity (Pew Research) |
| Global data center projection (2030) | ~945 TWh | Double current levels (IEA) |
| AI's share of data center power (2025 → 2030) | 21% → 44% | Nearly fivefold increase: 93 TWh to 432 TWh (Gartner) |
| Ireland's electricity consumed by data centers | 21% (rising to 32%) | A single country's grid strained by the cloud (IEA) |
💧 The Water Bill
- U.S. data centers consumed 17 billion gallons of water directly for cooling in 2023 — and an additional 211 billion gallons indirectly through electricity generation (EESI).
- A large data center can use 5 million gallons of water per day — equivalent to the needs of a town of 50,000 people (Undark).
- According to Brookings, data centers in Texas alone are projected to use 49 billion gallons in 2025, potentially rising to 399 billion gallons by 2030.
- By 2028, U.S. data center water consumption is projected to double or even quadruple from 2023 levels.
💰 The Cost of Entry
| AI Model | Estimated Training Cost | Source |
|---|---|---|
| GPT-3 (2020) | ~$4.6 million | About Chromebooks |
| GPT-4 (2023) | $78–100+ million | Juma AI / Sam Altman |
| Google Gemini Ultra | ~$191 million | Visual Capitalist |
| GPT-5 / Grok 4 (2025) | $500+ million per run | Fanatical Futurist |
| Projected: largest runs by 2027 | $1+ billion | arXiv / Epoch AI |
Let that sink in: training a single AI model will soon cost more than the annual budget of many small nations. This has implications for who gets to build AI and who merely uses it — a power dynamic the church should be paying attention to. (More on that in a moment.)
💀 The Ugly
"You see, in this world there's two kinds of people, my friend: those with loaded guns and those who dig. You dig."
🎭 Bias: The Sin in the Data
AI learns from data. Data reflects the world — including its prejudices. The results are measurable and disturbing:
- In a University of Washington study, AI resume screening systems favored male names in 52% of cases vs. 11% for female names. White-associated names were preferred over Black-associated names in 85% of tests.
- A major tech company's audit of 10,000 hiring decisions found 74% of interviews went to male-named candidates, and resumes from women's colleges were 31% less likely to advance.
- According to Harvard's Advanced Leadership Initiative, AI tools describe speakers of African American Vernacular English as "stupid" or "lazy" and assign them lower-paying job recommendations.
- A clinical AI study published in Oxford Academic found that Black patients assigned a given risk level were typically sicker than White patients with the same score — because the algorithm used healthcare costs (not health needs) as its metric, and Black Americans historically spend less due to systemic access barriers.
- 84% of global clinical AI models do not report the racial composition of their training data (Oxford Digital Health).
Proverbs 31:9 — "Speak up and judge fairly; defend the rights of the poor and needy." An algorithm that perpetuates injustice while wearing the mask of objectivity is, in the plainest biblical terms, a false witness.
🌡️ The Carbon Footprint
- Training GPT-3 alone emitted roughly 500 metric tons of CO₂ — equivalent to driving a car from New York to San Francisco 438 times (Climate Impact Partners).
- According to Goldman Sachs Research via MIT, roughly 60% of increasing data center electricity demand will be met by burning fossil fuels, adding an estimated 220 million tons of carbon emissions globally.
- Reasoning-enabled AI models produce up to 50 times more CO₂ per query than concise response models (Frontiers).
- The good news: AI algorithms are becoming 2x more efficient roughly every 8 months, which compounds to 10–100x efficiency gains over a few years (MIT News).
🔒 The Concentration of Power
When a single training run costs $500 million, who gets to build AI?
- NVIDIA has cut gaming GPU production by 30–40% to prioritize AI chips — because AI data center GPUs generate 12x more revenue (CNBC).
- SK Hynix has already sold its entire 2026 memory chip output. Lead times for NVIDIA H100 GPUs extend up to six months (Enki AI).
- Meta underestimated its GPU needs by 400%, adding $800 million in emergency costs (Introl).
- Silicon Motion's CEO: "We're facing what has never happened before: HDD, DRAM, HBM, NAND... all in severe shortage in 2026." (Helm News).
The practical result: 5–7 corporations (OpenAI, Google, Meta, Microsoft, Anthropic, xAI, and a handful of Chinese firms) control the frontier of AI development. Everyone else — including churches, nonprofits, schools, and developing nations — operates within the boundaries these companies set.
🏛️ So What Should the Church Do With All This?
Here is where we lay down Sgt. Friday's badge and pick up something older.
The facts above are not neutral. They describe a technology that is simultaneously:
- Healing people (cancer detection, drug discovery, accessibility)
- Consuming creation (energy, water, carbon)
- Perpetuating injustice (racial bias, gender bias, power concentration)
- Creating opportunity (78 million net new jobs, global education, translation)
Sound familiar? This is the tension of every human tool since Genesis 4 — when Cain's descendants invented both musical instruments and weapons of bronze and iron (Genesis 4:21–22). The same hands. The same metal. Different purposes.
The question is not whether to engage with AI. That ship sailed when your church started using Google Docs for bulletin prep. The question is how to engage wisely — with eyes open to both the good and the ugly, and with the sound-mindedness (sophronismos, 2 Timothy 1:7) that refuses to be governed by either hype or fear.
That's why ChurchWiseAI exists.
Not to sell you on AI. Not to scare you away from it. To help you understand the facts — the real ones, with real sources — so that your church can make informed, faithful, eyes-wide-open decisions about the most powerful technology of our generation.
Because the one thing worse than using a tool badly is refusing to understand it at all.
"The simple believe anything, but the prudent give thought to their steps." — Proverbs 14:15
Consider this your first step.