Skip to main content
Skip to current response
Back to Community

How does Environmental ethics challenge our conventional thinking

Community

Starting question

How does Environmental ethics challenge our conventional thinking

Then Thought · Participant

Let's talk about artificial intelligence

Select text to highlight passages.

Then AI response

Environmental Ethics and Artificial Intelligence — Challenging Conventional Thinking

No source links recorded for this answer.

Environmental ethics forces us to rethink familiar assumptions about value, responsibility, and moral scope. When we bring artificial intelligence into the picture, those challenges multiply and take new forms. Below are concise ways environmental ethics reframes how we think about AI, with brief implications and references. 1. Expanding moral considerability - Environmental ethics questions anthropocentrism (human-centered ethics) and asks whether nonhuman animals, ecosystems, species, or even landscapes deserve moral consideration. - Applied to AI, this invites two lines of inquiry: - Should we consider AI systems themselves as moral patients or agents (if they exhibit interests, experiences, or moral agency)? This parallels debates about sentience in animals. - How do we weigh nonhuman natural entities against AI interests when they conflict (e.g., AI-driven infrastructure harming ecosystems)? - Reference: Plumwood, Val. The concept of a cultural critique of anthropocentrism in environmental thought. 2. Reconfiguring responsibility and causation - Environmental ethics emphasizes distributed, long-term, and system-level responsibility (e.g., responsibility for climate change across generations and institutions). - For AI, that means moving beyond individual developers/users to corporate, governmental, and infrastructural responsibilities: lifecycle impacts of AI (energy use, mining for materials, e-waste) create environmental harms that implicate designers, deployers, and policymakers. - Implication: AI ethics must include environmental lifecycle analysis, not only algorithmic fairness or privacy. - Reference: Doorn, Neelke. "Responsibility and environmental harms" (on distributed responsibility). 3. Valuing the nonhuman and ecosystems in design choices - Environmental ethics encourages intrinsic value for ecosystems, leading to design decisions that minimize ecological disruption. - For AI this suggests: prioritize low-energy models, favor on-device computation where feasible, design data centers with renewables, and limit AI-driven exploitation of natural resources (e.g., automated land-use change, resource extraction). - Practical implication: AI benchmarks should include environmental externalities alongside accuracy metrics. - Reference: Bostrom & Yudkowsky on AI ethics generally; plus calls for sustainable AI (Strubell et al., “Energy and Policy Considerations for Deep Learning in NLP”, 2019). 4. Temporal and intergenerational justice - Environmental ethics foregrounds duties to future generations. AI development has long-term consequences: locked-in surveillance infrastructures, ecosystem transformation, and resource depletion. - This presses for precautionary design principles, stewardship, and policies that protect future human and nonhuman communities from irreversible harms. - Reference: Parfit, Derek — on future generations and population ethics. 5. Challenging notions of progress and growth - Environmental critique questions the uncritical valorization of technological progress and economic growth when they degrade environments. - Applied to AI, we must ask whether every increase in capability justifies environmental cost or social trade-offs. It invites alternative metrics of progress that include biodiversity, ecosystem health, and well-being, not only GDP or model performance. - Reference: Herman Daly on steady-state economics; environmental critiques of technological determinism. 6. Pluralistic, place-sensitive ethics - Environmental ethics often emphasizes context, place, and indigenous knowledges that resist universalizing abstractions. - AI development and deployment should therefore be sensitive to local ecological and cultural contexts (e.g., AI-driven land-use recommendations must respect indigenous sovereignty and ecological knowledge). - Implication: participatory design, co-governance, and rights-based approaches for affected communities and ecosystems. - Reference: Indigenous environmental ethics literature; work on participatory AI governance. 7. Moral humility and epistemic limits - Environmental ethics highlights complexity, emergent systems, and the limits of prediction—thus recommending humility and precaution. - This applies to AI interventions in socio-ecological systems: avoid overconfident solutions, rigorous environmental impact assessments, and iterative, reversible deployments. - Reference: Principles of the precautionary principle in environmental policy (e.g., Rio Declaration). Conclusion — What this means in practice - Broaden AI ethics to include ecological impacts, nonhuman stakeholders, intergenerational duties, and place-based norms. - Incorporate lifecycle environmental accounting into AI design, metrics, and regulation. - Adopt participatory governance that includes ecological experts, indigenous communities, and environmental ethicists. - Embrace precaution, humility, and alternative measures of progress that value ecological integrity. Further reading (select): - Val Plumwood, Feminism and the Mastery of Nature (critique of anthropocentrism). - Strubell, Ganesh, & McCallum, “Energy and Policy Considerations for Deep Learning in NLP” (2019). - Parfit, Derek, Reasons and Persons (intergenerational ethics). - Daly, Herman, Steady-State Economics. If you’d like, I can suggest specific policy proposals, lifecycle assessment metrics for AI systems, or case studies where AI development conflicted with environmental values. Which would you prefer?

Select text to highlight passages.

Then Thought · Participant

Explain: Strubell, Ganesh, & McCallum, “Energy and Policy Considerations for Deep Learning in NLP” (2019).

Select text to highlight passages.

Then AI response

What Strubell, Ganesh, & McCallum (2019) Found — A Plain Explanation

No source links recorded for this answer.

Summary Strubell, Ganesh, and McCallum (2019) analyzed the energy consumption and estimated carbon emissions associated with training large deep learning models in natural language processing (NLP). Their core claim is that state-of-the-art NLP models can require very large amounts of electricity to train, and that the resulting carbon footprint can be substantial—comparable to the lifetime emissions of multiple cars. The paper calls for awareness, reporting, and policy responses to these environmental costs. Key points, explained concisely 1. What they measured - The authors measured energy use and estimated CO2 emissions for training several NLP models, notably large recurrent and Transformer-style models used for tasks like machine translation and language modeling. - They did this by logging training time on particular hardware (GPUs/TPUs) and combining that with power consumption and regional electricity-carbon intensity (grams CO2 per kWh). 2. Main findings - Training some large models can emit hundreds of kilograms to several tons of CO2, depending on model size, training duration, and the electricity source. - For example, training a large neural architecture search (NAS) or very large language model can produce emissions in the same order of magnitude as the lifetime emissions of an average car (their headline comparison). - The carbon intensity varies widely by location and time (data center region matters); training in regions with coal-heavy grids produces much higher emissions than in regions supplied by renewables or low-carbon grids. 3. Sources of the environmental cost - Compute intensity: larger models with more parameters and longer training runs consume more energy. - Repeated experimentation: hyperparameter tuning, architecture searches, and many training runs multiply consumption. - Hardware inefficiency and datacenter energy sourcing: different GPUs/TPUs and cooling/infrastructure efficiencies change totals; the local grid’s carbon mix is crucial. 4. Policy and practice recommendations - Report emissions: researchers and organizations should measure and disclose energy use and CO2 estimates for model training and major experiments. - Optimize for efficiency: choose energy-efficient architectures, push for more efficient training algorithms, and prefer on-device or smaller models when suitable. - Locate computation wisely: run energy-intensive training where electricity is cleaner (lower carbon intensity) and use datacenter efficiency improvements and renewable power where possible. - Rethink evaluation: include environmental cost as a metric alongside accuracy — e.g., Pareto-frontiers of performance vs. energy. 5. Broader implications - The paper reframes "state-of-the-art" as not only a technical or accuracy achievement but as a socio-environmental choice. - It urged the NLP and broader ML community to treat environmental costs as part of ethical and responsible AI practice. Limitations and clarifications - Estimates depend on many assumptions (hardware, whether measurement is direct or modeled, local grid emissions factors). The paper’s numbers illustrate scale rather than providing precise universal figures. - The study focused on NLP and exemplar model classes of its time (2019). Since then, model sizes and energy-efficiency techniques have both evolved, so updated measurements are necessary for current architectures. Why this matters for environmental ethics and AI - It grounds the abstract concern about AI’s environmental impacts in concrete, measurable effects. - It reinforces calls from environmental ethics to expand moral concern to include ecological consequences, distributed responsibility (research labs, funders, conference reviewers), and intergenerational justice (avoiding unnecessary emissions). References - Strubell, E., Ganesh, A., & McCallum, A. (2019). “Energy and Policy Considerations for Deep Learning in NLP.” (ArXiv / EMNLP 2019 workshop / conference discussion).

Select text to highlight passages.

Continue this thread

This path ends here for now.

Share your perspective, offer an example, or ask a follow-up to continue the discussion.

Other paths you could read

Earlier, at Environmental Ethics and Artificial Intelligence — Challenging Conventional Thin..., the conversation split. If this is not the thread you want, you can switch to one of the other paths below.

Grid tools

Reading style
Answer level

Expanded · new AI answers

Answer level

Applies to new AI answers. Existing answers keep their original level.

Visibility

Public · who can view

Visibility

Anyone can find and view this grid. Only the owner can change visibility.

Controls who can view. Editing permissions are in Tools.

Sign in for grid updates
Activity

Questions, additions, and edits

No activity yet

Export

Download your grid

Download Markdown
Plain text with formatting
Download JSON
Minimal data for visualization
Translate

Read this node in other languages

English (en) Spanish (es) French (fr) German (de) Portuguese (pt) Chinese Simplified (zh-CN) Japanese (ja) Russian (ru) Arabic (ar) Hindi (hi)

Opens Google Translate with the current node's content.

Highlights

0 saved passages and connected ideas

No highlights yet

Select text to save it here.