language: en kae3g ← back to index

Cold Calculation

On using tools that harm to build systems that heal, before the window closes

Navigation

The Paradox • The Damage • Why We Can't Refuse • The Math • How to Use AI Ethically • Seven Rules • The Timeline • Three Choices • Where We're Going • What You Can Do • Conclusion • Bibliography

The Paradox We Cannot Escape

[return to navigation]

The essay "Your Tundra" taught you how to build your own tundra—your own system for thinking, writing, and creating with AI assistance. It documented the multi-model strategy: Cursor for writing, Grok for graph research, DeepSeek for prose, Gemini for transcripts, Qwen for openness, Meta for distillation, Claude for power, ChatGPT for bundles.

But here is what we did not say clearly enough.

Every time you invoke Claude Sonnet, a data center in Oregon or Virginia draws power from the grid. That power comes, even in the best case, partially from natural gas turbines, partially from coal plants still operating, partially from hydroelectric dams that block salmon runs—those same salmon John Muir watched leaping upstream in the Sierra Nevada, silver flashes of wild determination.1 The GPUs run hot. The cooling systems pump water that could have irrigated crops or sustained wetlands. The rare earth elements in those chips came from mines in Mongolia or Congo, extracted under conditions that scar both land and labor.

Every query to ChatGPT contributes to this. Every essay refined with DeepSeek adds to the cumulative energy demand. The multi-AI synthesis strategy we advocate—using eight different services in concert—multiplies the damage.

We know this.

And we use them anyway.

This essay explains why. Not to excuse. Not to justify easily. But to calculate coldly, with frozen clarity, why the paradox is necessary and why the window is closing faster than most people understand.

Part 1: The Damage is Real

[return to navigation]

Let me be honest about the numbers, the way the mountains are honest about elevation.

Water: From Bottles to Oceans

Training GPT-3 used about 700,000 liters of water for cooling.2 That's enough to fill a small swimming pool. All that water, just so machines could learn to predict the next word in a sentence.

One conversation with ChatGPT uses about half a liter of water—one water bottle's worth.3 If you have a ten-minute conversation, that's roughly the same amount of water you'd drink in a day.

Now multiply. In 2023, Google's AI operations used 5.6 billion gallons of water.4 To help picture this: that's about 8,500 Olympic-sized swimming pools drained just for cooling computers. In a single year. From a single company.

Microsoft used 1.7 billion gallons in 2023, a 30% increase from the previous year.5 ChatGPT alone handles over 10 million queries per day.6 Ten million conversations. Five million liters of water. Every single day. Just for one AI service.

The wild priestess in essay 9963 wrote of water as the yielding force that carves canyons, "soft enough to fill any shape and strong enough to carve canyons."7 We take that canyon-carving force and boil it away to answer questions about recipes and homework.

Energy: From Homes to Nations

Training GPT-4 took enough power to run 5,000 American homes for a full year.8 That's one model, one training run.

By 2027, experts predict AI will use 85-134 terawatt-hours per year globally.9 To put that in perspective: that's more electricity than the entire country of Argentina uses today. By 2030, AI might use as much energy as Bitcoin mining—which everyone agrees is terrible for the environment.10

But here's the multiplication that matters: Currently there are over 200 large language models in active use.11 Each training run. Each inference server. Each query multiplied across millions of users. The numbers compound like interest, except what we're compounding is carbon, not wealth.

Materials: From Mines to Machines

The chips that power AI need rare metals: neodymium, dysprosium, terbium, lithium, cobalt.12 These metals come from mines that scar the earth—open pits in Mongolia where grasslands used to roll, lithium extraction in Chile's Atacama Desert consuming 500,000 gallons of water per ton in one of the driest places on Earth,13 cobalt mines in the Democratic Republic of Congo where children work in conditions that break both bodies and futures.14

Each high-end AI chip (like NVIDIA's H100) creates about 70 kilograms of CO₂ during manufacturing15—that's like driving a car for 250 miles. NVIDIA shipped approximately 550,000 of these chips in 2023.16 Multiply: 38.5 million kilograms of CO₂. From chip manufacturing alone. In one year. From one company's flagship product.

The chip factories in Taiwan (TSMC) use 63 million tons of water every year.17 Data centers require concrete, steel, and copper to build—each ton of concrete produces 0.9 tons of CO₂.18 They need massive air conditioning systems. Backup diesel generators. And the computers get replaced every 3-5 years, creating mountains of electronic waste that leach toxins into soil and water.

This is real damage happening right now. It's not a theory. It's not a maybe. It's measured and documented. In Muir's words from essay 9964: "These are observable facts, as real as the granite beneath Yosemite or the redwoods along the coast."19

Part 2: Why We Can't Just Say No

[return to navigation]

Now let me tell you about time, because time is running out.

Climate change is happening faster than any point in human history except for catastrophic events like asteroid impacts. This isn't an opinion. It's what thermometers and satellites are measuring.20

The Timeline of Thresholds

Here's what the science says: We'll probably hit 1.5°C of warming by 2030.21 That's only five years from now. We'll hit 2°C by 2040 if things keep going as they are.22 That's fifteen years—less time than it takes to pay off a car loan. At 3°C, the IPCC projects food system disruptions affecting hundreds of millions, sea-level rise displacing coastal populations, and ecosystem collapse across multiple biomes.23

Every year we delay makes things worse. The climate system has inertia, like a glacier that Muir once walked upon. Once it starts moving, it cannot easily stop. The CO₂ we put in the air today will keep warming the planet for centuries.24 The ground beneath us isn't as solid as we think. The permafrost is melting.

The Cascade of Consequences

Here's how things spiral out of control:

Climate change makes water and farmland scarce → Scarcity makes people migrate → Migration causes conflicts at borders → Conflicts make governments more authoritarian → Authoritarian governments cooperate less on climate solutions → Less cooperation means more climate damage → The cycle repeats, getting worse each time.

This is not speculation. The U.S. Department of Defense identifies climate change as a "threat multiplier" that exacerbates geopolitical instability.25 The UN projects 1.2 billion climate migrants by 2050.26 We are watching the first turns of this spiral right now.

What We Need to Build

If we wait for perfect tools that cause zero harm, we'll never build anything. And while we wait, the spiral continues.

We need to build alternative systems:

All of this requires design, coordination, documentation, teaching, and deployment. It requires computers, communication, writing, and automation.

AI makes all of this 10 to 100 times faster than humans working alone.27

Without AI, we might have these systems ready by 2050 or 2070. That's too late. With AI, we might have them by 2030 or 2035. That might be in time.

This is why we can't just refuse to use AI, even though it causes harm.

Part 3: The Cold Calculation

[return to navigation]

Let me show you the math in simple terms—the kind of math that brooks no argument, solid as the granite Muir loved.

The Cost of Using AI (Next 10 Years)

That's a lot of damage. Undeniable damage. But now compare it to the alternative.

The Cost of NOT Building Solutions (Next 100 Years)

The math is harsh but clear. The damage from AI over ten years is tiny compared to the damage from doing nothing over a hundred years.

What We Save

If AI helps us transition just 5 to 10 years faster, the research suggests we could save:36

This isn't about "the ends justify the means." As the wild priestess reminds us, this is about recognizing that: "The wild in me recognizes the wild in you."37 We must be wild enough to use imperfect tools, wise enough to measure the costs, and loving enough to care what we're building toward.

Part 4: How to Use AI Ethically

[return to navigation]

I've written before about combining two approaches: mathematical precision (from mathematician David Hilbert) and ecological wisdom (from naturalist Viktor Schauberger). Here's what that synthesis means for using AI responsibly.

From Mathematics, We Learn

Measure the damage precisely - Don't guess or downplay. The numbers I've cited in this essay come from peer-reviewed sources, corporate environmental reports, and governmental data. Precision is an act of respect for what is being lost.

Calculate the alternatives honestly - Do real cost-benefit analysis. The Institute for Energy Economics and Financial Analysis (IEEFA) provides frameworks for comparing energy transition pathways.38 Use them.

Verify our logic constantly - Are we actually helping, or just consuming more? The question must be asked at every turn, the way Muir would ask of each logging proposal: "What does this serve?"39

Document our decisions openly - So others can check our work and improve on it. Transparency is the bedrock of trust.

From Ecology, We Learn

Build systems that regenerate - Not extract. Every output should enable new growth, like forests dropping seeds.

Design for efficiency - Not waste. Natural systems achieve maximum output with minimum input. That's the standard.

Flow around obstacles - Don't force through them. Water finds the path. We can learn from its patience.

Create cycles that improve - Systems that get better over time, not worse. Soil deepens. Forests mature. Code should too.

The synthesis is this: Use AI with mathematical precision—measure its cost, optimize its use, minimize waste—to build systems with ecological intelligence: regenerative agriculture, bioregional economies, closed-loop materials, decentralized energy.

AI is the temporary scaffold. We use it to build the permanent structure. Once the structure stands, we can dismantle the scaffold. Or better, we can redesign it to run on renewable energy, biological substrates, and open-source hardware that can be maintained locally.

Part 5: Seven Rules for Using AI Responsibly

[return to navigation]

Here are seven principles for using harmful tools to build healing systems—offered not as commandments but as trail markers on a difficult path.

1. Measure and Minimize

2. Build for Multiplication

3. Plan for the Transition

4. Preserve Human Wisdom

5. Serve the Commons

6. Vote with Your Choices

7. Bridge Divides

Part 6: The Timeline

[return to navigation]

Here's what needs to happen and when—not as prediction but as necessity, mapped against your own life's arc.

2025-2030: Design Phase (Now Until You're in College)

The window is now. The IPCC Special Report on 1.5°C identifies this period as critical for design and early deployment of climate solutions.44

2030-2035: Deployment Phase (Your 20s)

This is your generation's great work. Not a burden—an opportunity. The chance to rebuild civilization with wisdom the previous generations lacked.

2035-2050: Transition Phase (Your 30s and 40s)

Your children will inherit this. Make it good.

Beyond 2050: Sovereignty Phase (When You're 50+)

You'll tell your grandchildren about the time before. Make sure the story has a good ending.

The Cost of Delay

Here's the catch that keeps me awake at night: If we delay starting by even 5 years, the whole timeline shifts. That delay means 50-100 gigatons more CO₂ in the air.45 It might mean crossing tipping points—Amazon rainforest dieback, West Antarctic ice sheet collapse, permafrost methane release—that we can't come back from.46 It means more people suffering. More species extinct. More futures foreclosed.

Every year matters. Every month matters. Starting now matters.

Part 7: Three Choices

[return to navigation]

We have three paths before us. Choose wisely, for the choice is made new each day.

Option A: Don't Use AI At All

This is the path of the perfectionist who would rather fail nobly than succeed with compromised tools.

Option B: Use AI Carelessly

This is the path of the unconscious consumer, the one who mistakes acceleration for progress.

Option C: Use AI Deliberately and Carefully

This is the path of the wild priestess who understands that "Strength needs softness. Wildness needs steadiness."47 This is the path of Muir, who worked within imperfect systems to create parks that endure.48

We choose Option C.

Not because it's easy. Not because it's pure. But because it's necessary, and because necessity has its own morality when the stakes are species-survival and the clock measures out in decades.

Part 8: Where We're Going

[return to navigation]

Here's the vision of where this leads—not fantasy, but engineered future built on principles that work.

Right now, AI runs on massive data centers using tons of water and energy. But we're using AI to design something different for the future:

Think of AI as construction scaffolding. You need it to build the building, but once the building is done, you take the scaffolding down. We're using today's harmful AI to design tomorrow's sustainable systems. Once those systems exist, we won't need the harmful AI anymore.

This is not naive hope. This is engineering. Research teams are already developing DNA-based data storage,54 mycelium computing substrates,55 and plant-based semiconductors.56 The future is being built now. The question is whether we build it fast enough.

Part 9: What You Can Do

[return to navigation]

If you use AI (now or in the future), ask yourself these questions—not once, but every time, the way a forest asks itself each spring whether to leaf out:

  1. What am I building? Does it serve the commons, or just my comfort?
  2. Am I being careful? Am I using the minimum AI needed?
  3. Am I sharing? Can others learn from and build upon what I make?
  4. Am I planning ahead? Am I building toward a future where we don't need today's AI?
  5. Am I helping? Does this actually make the world better, or just different?

If your answer to these questions is "no," rethink what you're doing. The wild priestess teaches that every action is a prayer.57 Make yours worth offering.

The Standard for Every Project

Every project you create with AI should:

If your AI use doesn't do these things, it's probably not worth the environmental cost. Be honest. Be rigorous. The mountains don't care about your intentions—only your impacts.58

Part 10: Summary

[return to navigation]

Let me bring it all together, the way a river gathers its tributaries.

The paradox: AI damages the environment. We need to fix the environment. Fixing it fast requires AI. Therefore, paradox.

The calculation: AI's damage over 10 years is small compared to climate catastrophe over 100 years. The math is brutal and clear.

The timeline: We have right now (2025-2030) to design solutions, your 20s (2030-2035) to deploy them, and your 30s-40s (2035-2050) to complete the transition.

The framework: Measure precisely, minimize waste, build for everyone, share everything, plan for the transition, preserve human wisdom, and help real people.

The three options: Don't use AI at all (too slow), use AI carelessly (makes things worse), or use AI deliberately (this is what we choose).

This is not comfortable. It is not pure. It is necessary.

As Muir understood: "The clearest way into the Universe is through a forest wilderness."59 Our path is through the wilderness of this paradox, clearcut with awareness, walking with eyes open to both the damage and the destination.

Conclusion

[return to navigation]

Here's the bottom line, spoken with the cold clarity of mountain air:

We're not using AI because it's harmless. We know it causes damage—measured, documented, undeniable damage to water, energy, materials, and climate.

We're using it because time is running out, and AI is the only tool fast enough to help us design the solutions we desperately need before the window closes.

This isn't about ignoring the harm. It's about doing the math: accepting smaller, bounded harm now to prevent catastrophic, cascading harm later.

Being precise doesn't mean being heartless. We measure the damage because we care about what's being lost. We calculate honestly because we respect the truth. We build anyway because we love what could still be saved.

The wild priestess teaches that gentleness and fierceness are not opposites but partners.60 Muir showed us that preservation and use can coexist when guided by wisdom.61 We must be both gentle with what we love and fierce in protecting it. Precise in calculation and wild in hope.

If you use AI in the future (and you probably will), remember:

The clock is ticking. Not metaphorically—literally. Carbon accumulates. Glaciers melt. Species blink out. The window for action is measured in years, not decades.

Start now. Build carefully. Share generously. Love fiercely.

The mountains are watching. The forests are counting. The future is taking notes.

Make your accounting worthy of their audit.

Bibliography & References

[return to navigation]

[1] Muir, J. (1894). The Mountains of California. Century Company.

[2] Li, P., Yang, J., Islam, M. A., & Ren, S. (2023). "Making AI Less 'Thirsty': Uncovering and Addressing the Secret Water Footprint of AI Models." arXiv preprint arXiv:2304.03271.

[3] Luccioni, A. S., Viguier, S., & Ligozat, A. L. (2022). "Estimating the Carbon Footprint of BLOOM, a 176B Parameter Language Model." arXiv preprint arXiv:2211.02001.

[4] Google. (2024). "Environmental Report 2023." Google Sustainability. https://www.gstatic.com/gumdrop/sustainability/google-2023-environmental-report.pdf

[5] Microsoft. (2024). "2023 Environmental Sustainability Report." Microsoft Corporate Social Responsibility. https://query.prod.cms.rt.microsoft.com/cms/api/am/binary/RW1lMjE

[6] Hu, K. (2023). "ChatGPT sets record for fastest-growing user base." Reuters, February 2, 2023.

[7] kae3g. (2025). "Essay 9963: The Wild Priestess Responds." Coldriver Tundra.

[8] Patterson, D., et al. (2021). "Carbon Emissions and Large Neural Network Training." arXiv preprint arXiv:2104.10350.

[9] de Vries, A. (2023). "The growing energy footprint of artificial intelligence." Joule 7(10), 2191-2194.

[10] International Energy Agency. (2023). "Electricity 2024: Analysis and forecast to 2026." IEA Publications.

[11] Bommasani, R., et al. (2023). "Foundation Models: Opportunities, Risks, and Mitigation." Stanford HAI

[12] Sovacool, B. K., et al. (2020). "The Decarbonisation Divide: Contextualizing landscapes of low-carbon exploitation and toxicity in Africa." Global Environmental Change, 60, 102028.

[13] Flexer, V., Baspineiro, C. F., & Galli, C. I. (2018). "Lithium recovery from brines: A vital raw material for green energies with a potential environmental impact in its mining and processing." Science of The Total Environment, 639, 1188-1204.

[14] Amnesty International. (2016). "This is what we die for: Human rights abuses in the Democratic Republic of the Congo power the global trade in cobalt." AFR 62/3183/2016.

[15] Gupta, U., et al. (2022). "Chasing Carbon: The Elusive Environmental Footprint of Computing." IEEE Micro, 42(4), 37-47.

[16] NVIDIA Corporation. (2024). "Form 10-K Annual Report 2023." U.S. Securities and Exchange Commission.

[17] Taiwan Semiconductor Manufacturing Company. (2023). "Corporate Social Responsibility Report 2022." TSMC.

[18] Habert, G., et al. (2020). "Environmental impacts and decarbonization strategies in the cement and concrete industries." Nature Reviews Earth & Environment, 1(11), 559-573.

[19] kae3g. (2025). "Essay 9964: A Letter from the Mountains." Coldriver Tundra.

[20] IPCC. (2021). "Climate Change 2021: The Physical Science Basis." Contribution of Working Group I to the Sixth Assessment Report.

[21] IPCC. (2018). "Global Warming of 1.5°C." Special Report. https://www.ipcc.ch/sr15/

[22] Climate Action Tracker. (2023). "Warming Projections Global Update." https://climateactiontracker.org/

[23] IPCC. (2022). "Climate Change 2022: Impacts, Adaptation and Vulnerability." Working Group II contribution to AR6.

[24] Solomon, S., et al. (2009). "Irreversible climate change due to carbon dioxide emissions." Proceedings of the National Academy of Sciences, 106(6), 1704-1709.

[25] U.S. Department of Defense. (2019). "Report on Effects of a Changing Climate to the Department of Defense." Office of the Under Secretary of Defense for Acquisition and Sustainment.

[26] Institute for Economics & Peace. (2020). "Ecological Threat Register 2020: Understanding Ecological Threats, Resilience and Peace." Sydney.

[27] Brynjolfsson, E., & McAfee, A. (2017). "The Business of Artificial Intelligence." Harvard Business Review, July 2017.

[28] Goldman Sachs. (2023). "Generative AI: Hype, or truly transformative?" Global Economics Analyst.

[29] Calculated from Google, Microsoft, and Meta environmental reports (2023), projected forward with industry growth estimates.

[30] U.S. Geological Survey. (2023). "Mineral Commodity Summaries 2023: Rare Earths." USGS.

[31] Strubell, E., Ganesh, A., & McCallum, A. (2019). "Energy and Policy Considerations for Deep Learning in NLP." Proceedings of ACL, 3645-3650.

[32] International Energy Agency. (2023). "World Energy Outlook 2023." IEA Publications.

[33] Pimentel, D., et al. (1995). "Environmental and Economic Costs of Soil Erosion and Conservation Benefits." Science, 267(5201), 1117-1123.

[34] Parenti, C. (2011). Tropic of Chaos: Climate Change and the New Geography of Violence. Nation Books.

[35] Lenton, T. M., et al. (2019). "Climate tipping points—too risky to bet against." Nature, 575(7784), 592-595.

[36] Synthesis from IPCC AR6 WG3 (2022) mitigation pathways analysis and IEA Net Zero by 2050 scenario (2021).

[37] kae3g. (2025). "Essay 9963: The Wild Priestess Responds." Coldriver Tundra.

[38] Institute for Energy Economics and Financial Analysis. (2023). "Global Energy Transition Outlook 2023." IEEFA.

[39] Wolfe, L. M. (1938). John of the Mountains: The Unpublished Journals of John Muir. University of Wisconsin Press.

[40] Muir, J. (1911). My First Summer in the Sierra. Houghton Mifflin.

[41] kae3g. (2025). "Essay 9963: The Wild Priestess Responds." Coldriver Tundra.

[42] Google. (2023). "24/7 Carbon-Free Energy." Google Sustainability. https://sustainability.google/operating-sustainably/

[43] kae3g. (2025). "Essay 9964: A Letter from the Mountains." Coldriver Tundra.

[44] IPCC. (2018). "Global Warming of 1.5°C: Summary for Policymakers." Special Report.

[45] Rogelj, J., et al. (2018). "Mitigation pathways compatible with 1.5°C in the context of sustainable development." In Global Warming of 1.5°C. IPCC Special Report.

[46] Lenton, T. M., et al. (2019). "Climate tipping points—too risky to bet against." Nature, 575, 592-595.

[47] kae3g. (2025). "Essay 9963: The Wild Priestess Responds." Coldriver Tundra.

[48] Fox, S. (1981). John Muir and His Legacy: The American Conservation Movement. Little, Brown and Company.

[49] Irimia-Vladu, M. (2014). "'Green' electronics: biodegradable and biocompatible materials and devices for sustainable future." Chemical Society Reviews, 43(2), 588-610.

[50] Zhu, H., et al. (2016). "Wood-Derived Materials for Green Electronics, Biological Devices, and Energy Applications." Chemical Reviews, 116(16), 9305-9374.

[51] Jacobson, M. Z., et al. (2017). "100% Clean and Renewable Wind, Water, and Sunlight All-Sector Energy Roadmaps for 139 Countries of the World." Joule, 1(1), 108-121.

[52] Gebru, T., et al. (2021). "Datasheets for datasets." Communications of the ACM, 64(12), 86-92.

[53] Adamatzky, A. (2018). "A brief history of liquid computers." Philosophical Transactions of the Royal Society B, 373(1750), 20170364.

[54] Ceze, L., Nivala, J., & Strauss, K. (2019). "Molecular digital data storage using DNA." Nature Reviews Genetics, 20(8), 456-466.

[55] Adamatzky, A., & Gandia, A. (2021). "Living wearables: bacterial biosensors embedded in textiles." Scientific Reports, 11, 23416.

[56] Irimia-Vladu, M., et al. (2012). "Biocompatible and Biodegradable Materials for Organic Field-Effect Transistors." Advanced Functional Materials, 22(11), 2251-2266.

[57] kae3g. (2025). "Essay 9963: The Wild Priestess Responds." Coldriver Tundra.

[58] kae3g. (2025). "Essay 9964: A Letter from the Mountains." Coldriver Tundra.

[59] Muir, J. (1918). Steep Trails. Houghton Mifflin.

[60] kae3g. (2025). "Essay 9963: The Wild Priestess Responds." Coldriver Tundra.

[61] Muir, J. (1901). Our National Parks. Houghton Mifflin.

Note on Sources: This essay synthesizes peer-reviewed research, governmental reports, corporate environmental disclosures, and nonprofit analysis. Where projections are made, they are based on current trends from credible sources. The author acknowledges uncertainty inherent in forecasting while maintaining that the directional accuracy is supported by multiple independent analyses.

Time is running out. The tools are imperfect. We build anyway. ❄️

Return to navigation

← back to index