Our Tech has a Climate Problem: Here's we solve it

Episode Summary

Executive Summary: The episode examines the climate costs and benefits of modern technology, arguing that AI, electric vehicles, and satellites can help solve environmental problems only if their hidden footprints are measured, constrained, and governed. Across AI research, mining, space debris, and renewable forecasting, the message is clear: sustainability requires transparency, smarter design, reuse, and fewer unnecessary tech-intensive choices.

Main Topics: AI’s hidden energy and carbon footprint (Priority: 5/5): AI researcher Sasha Luccioni explains that large language models consume significant energy, but companies rarely disclose usage data. She argues for measurement tools and public standards so users and developers can make informed choices. Making AI accountable with measurement tools (Priority: 5/5): Luccioni describes Code Carbon and proposes an Energy Star-style rating system for AI models to help compare emissions and efficiency across tasks and deployments. Electric vehicles and the mining trade-off (Priority: 5/5): The show explores how EVs reduce transport emissions but require minerals like lithium, cobalt, nickel, and graphite, creating environmental damage, Indigenous land conflicts, and labor abuses. Redesigning transportation to use fewer minerals (Priority: 4/5): Experts argue that the best solution is reducing car dependence, building smaller and repairable vehicles, expanding public transit, and creating recycling loops for batteries. Satellites as climate monitors and space debris as a new pollution crisis (Priority: 4/5): Satellites and AI help track emissions, methane leaks, storms, and droughts, but space is becoming crowded with debris, threatening essential services and demanding new rules and cleanup systems. AI for renewable energy forecasting (Priority: 4/5): DeepMind’s Sims Witherspoon shows that AI can improve wind power forecasting and help grids rely more on renewables, while warning that AI is not a silver bullet and must be used responsibly.

Key Arguments: AI’s environmental impact is significant but hard to quantify because companies do not consistently disclose energy and carbon data. Without transparent metrics, consumers and developers cannot choose lower-carbon AI systems or hold companies accountable. EVs are necessary for decarbonizing transport, but the clean-energy transition still carries major mining impacts on ecosystems and Indigenous communities. The most effective way to reduce battery mineral demand is not just better mining, but fewer cars, smaller cars, repairable vehicles, and stronger public transit. Recycling and closed-loop battery systems can reduce the need for virgin mineral extraction over time. Satellites are essential for climate monitoring, but orbital debris is creating a sustainability crisis that could threaten the infrastructure we rely on. A circular-economy approach should be extended to space through reusable rockets, recyclable satellites, and responsible end-of-life disposal. AI can materially help climate action, especially in forecasting wind and grid demand, but it must be deployed carefully and with clean power where possible.

Data Points: ChatGPT query energy estimate: ~5 watts per hour for a dead iPhone charge reference; exact query use not disclosed - Used to illustrate how hard it is to measure the energy of a single AI query Large language model emissions estimate: As much carbon as five cars over their lifetimes - Early estimate cited for training a model like ChatGPT ChatGPT-3 training energy estimate: As much energy as 130 American homes use in one year - Recent estimate cited during discussion of AI model training Bloom training energy: Equivalent to 30 homes for one year - Sasha Luccioni’s open large language model study Bloom training carbon: 25 tons of CO2 - Environmental impact of training Bloom GPT-3 relative emissions: 20 times more carbon than similar models like Bloom - Comparison used to show scale differences among models Model size growth: 2,000 times larger over the last five years - Luccioni on the expansion of large language models Larger vs smaller model emissions: 14 times more carbon for the same task - Recent work comparing efficient models to larger language models Lithium demand in the U.S.: Increase by 4,000% over 15 years - Government estimate tied to EV demand and Thacker Pass mining Road transport emissions share: 10% of global emissions - Elsa Dominish on why EVs matter for decarbonization Cobalt mining share in DRC: 60–70% - Dominish describing concentration of global cobalt extraction Informal mine worker pay: As little as $2.50 a day - Conditions in cobalt mining and subcontracted labor Nickel future demand: 40 times more nickel by 2040 - Concern about whether current extraction pathways are sustainable Space objects tracked: Over 50,000 - Moriba Jah on current tracked debris and objects in orbit Working satellites: Over 5,000 - Current operational satellites in orbit Elon Musk-owned satellites: Over half of working satellites - As stated in the segment on orbital congestion ISS avoidance maneuvers: About a dozen per year - Illustrates the frequency of collision-avoidance actions Google wind deployment test size: 700 megawatts - DeepMind’s real-world AI forecasting pilot Forecast improvement: 20% better than Google's existing systems - AI system for wind supply forecasting UK grid forecast improvement: Two times more accurate - Open Climate Fix demand-side forecasting deployment

Pivotal Quotes: "Once you start gathering the information, you have less plausible deniability." — Sasha Luccioni: On why companies avoid disclosing AI energy and emissions data "We can't flush out all the water from out of here and rip up everything that is out here and call it green energy. That's greenwashing." — Gary McKinney: Protesting the lithium mine at Thacker Pass "The most important thing we can do is to reduce the amount of minerals we need." — Elsa Dominish: On solutions to the EV mining dilemma

Implications: For listeners and industry, the episode argues for transparency, restraint, and systems thinking: measure tech’s impacts, choose lower-carbon options, redesign transport and space systems, and deploy AI where it clearly helps climate goals rather than adding avoidable harm.