The Dual Nature of AI: The Potential for AI to Save the Earth, and What Comes Before — The Hidden Story of Power, Water, and Emissions

The Dual Nature of AI: The Potential for AI to Save the Earth, and What Comes Before — The Hidden Story of Power, Water, and Emissions

The speed at which AI is integrating into our lives has accelerated significantly in recent years. Search, translation, document creation, image generation, customer support—the convenience has undoubtedly expanded, but behind this "normalcy," AI requires a large amount of electricity and water. The expansion of data centers, semiconductor manufacturing, water use for cooling, and changes in power generation configurations to ensure power supply. These factors are accumulating, raising the question, "Is AI a burden on the environment?"


However, the article on Phys.org (reposted from The Conversation) avoids a simple dichotomy of good and evil. While AI can increase environmental costs, the same AI can also be a tool to identify, reduce, and optimize "waste of resources." In other words, AI can be both a "fuel-consuming device" and a "control device to stop fuel waste." The issue is which will prevail and what is needed to make it prevail.


1) First, the "burden" issue: Why is AI bad for the environment (in a negative sense)?

The environmental impact of AI can be broken down into three main areas.

  • Electricity: Both training and inference (calculations during use) require the operation of a large number of GPUs/TPUs.

  • Water: Tied to water use for cooling and power generation, it can conflict with water resource scarcity in certain regions.

  • Infrastructure and Materials: Includes semiconductors, servers, power grid enhancements, building materials, and e-waste.


Even more troublesome is the "rebound effect" (demand induction), where efficiency does not necessarily lead to a reduction in total consumption. Even if AI achieves energy savings, if AI usage explodes, the total electricity consumption increases. In other words, to use AI in an "environmentally friendly way," not only technology but also demand-side design (where to use AI and where not to) and frameworks for measurement, disclosure, and regulation are necessary.

2) Reasons it can still be an "ally": Five field examples from the article

What makes the article interesting is that it doesn't end with abstract discussions about "AI being useful for environmental measures," but instead lists multiple use cases in the field. Here, we focus on "what can be reduced" in each area.


(A) Agriculture: Reducing water = reducing electricity

Agriculture accounts for a large portion of global freshwater use. The article introduces precision irrigation, like that of the climate tech company Kilimo from Argentina, which uses machine learning and meteorological/satellite data to optimize "when, where, and how much water to use." Reducing water is valuable in itself, but even more important is that it also reduces the energy needed to pump and deliver water to the fields.


It also touches on mechanisms to verify water savings and trade them as water-saving credits, translating environmental impact reduction into economic incentives.


(B) Data Centers: Using AI to cut AI waste

It may sound ironic, but AI is effective in data center operations. The article mentions that while data centers have high power demands, operational improvements have led to increased efficiency, and AI can analyze "workloads," "temperature," "cooling efficiency," and "power usage" to adjust computing resources and cooling according to demand.


For example, putting servers into low-power mode during low-demand periods, optimizing cooling and airflow, and changing operations according to weather—these "subtle optimizations" have a significant impact, especially in facilities with large total power consumption. The key is that these optimizations are not just "theoretical savings" but are tied to operational KPIs (power costs, utilization rates, temperature constraints, failure rates).


(C) Energy Industry: Reducing leaks and losses through inspection and monitoring

The energy industry has significant emissions, but also great potential for improvement. The article touches on cases where drones and image analysis are used to detect abnormalities in pipelines, and efforts to use AI for methane monitoring and estimation. Methane's strong short-term greenhouse effect makes leak countermeasures particularly noteworthy.


However, there is also potential backlash here. The suspicion is that "fossil fuel companies will use AI to 'optimize' and ultimately prolong their existence." Since AI can be both an "accelerator for decarbonization" and a "high-efficiency device for fossil fuels," society needs to decide in which direction to use it.


(D) Buildings and District Heating: Reducing waste through "systems," not human behavior

Heating, cooling, and electricity in homes and offices are directly linked to emissions. The article mentions the optimization of district heating in central Copenhagen (using sensors to read building conditions and adjust supply based on 24-hour forecasts) and the potential for AI to reduce energy use in medium-sized offices as a research topic.
The key here is that the system proactively eliminates waste, not relying on "requests." While human energy-saving awareness fluctuates, control continues steadily. AI excels in reading temperature, humidity, and usage conditions to align supply and demand.


(E) Aviation: Reducing contrails and fuel waste

In aviation, not only CO₂ emissions but also the warming impact of contrails are debated. The article cites examples where AI adjusts flight routes and altitudes to avoid humidity conditions that form contrails, and proposes efficient routes from operational data for fuel efficiency improvement.
Aviation is a field where implementation is cautious due to strict safety and regulations, but if done well, it can have a significant impact.

3) SNS Reactions: Empathy and caution run "simultaneously"

Following the reactions on social media to this article, the tone is largely divided into two.


① "AI is not just destroying the environment" camp
Using AI as an "optimization tool" in fields with significant resource waste, such as agriculture, buildings, and aviation, has some support. There is an expectation that "using it where necessary can offset environmental impact." Especially in areas like water and fuel reduction, which align with cost savings on the ground, implementation is seen as realistic and is more positively received.


② "Ultimately, the total amount increases" camp (caution and criticism)
On the other hand, there is a strong concern that "even if efficiency improves, increased usage negates it" and "data centers proliferate under the banner of energy saving." As AI becomes more widespread, the number of inferences increases, models grow larger, and data center investments accelerate. Unless society can resist the "temptation of convenience," the total amount is likely to increase—this is the perspective. Recently, there have been numerous discussions highlighting the power demand of AI and data centers, a return to fossil fuels, and a lack of transparency, and this context is reflected in social media.


③ "Who benefits and who bears the burden" camp (regional and fairness)
Another point gradually gaining traction is the issue of fairness. There is a structure where the regions hosting data centers bear the water and power burden, while the benefits flow elsewhere. If AI is to do "good on a global scale," there needs to be a transparent explanation of how to distribute the benefits and burdens.


Note: On the Phys.org side, the article's comment section is essentially inactive (at least, it shows zero comments), and the spread on social media seems more like "interested parties sharing among themselves" rather than a large-scale uproar. However, since this theme is a topic of ongoing global discussion, the "dual nature" presented in the article is being read as adding fuel to existing debates.


4) Conclusion: AI's environmental issue is about "design" and "governance," not "technology"

The article consistently suggests that the issue is not whether to "use AI or not," but rather where, under what conditions, at what scale, and how to measure its use. There are indeed areas where AI can be beneficial. However, it is also true that the environmental costs of AI are growing in reality, and if left unchecked, "AI for saving" will be swallowed by "AI for consuming."


The realistic compromise will likely be summarized in the following three points.

  1. Measurement and Disclosure: Visualization of electricity/water/emissions for each model/service (creating a "selectable" state).

  2. Prioritization: Emphasizing areas with greater social benefits (power grids, buildings, agriculture, industrial efficiency) over entertainment and excessive uses.

  3. Rebound Countermeasures: Institutional design to prevent efficiency from translating into increased demand (pricing, regulations, procurement standards, agreements with municipalities).


The possibility that AI can "save the planet" is not zero. However, it will not happen automatically. AI, if left unchecked, will consume resources. If properly designed, it can prevent waste. Ultimately, creating AI that is kind to the planet is not up to AI, but to human decision-making.


Source URL

Related Articles