ALG Blog 2: How GenAI Works

Generative AI just a game of guessing
This case study explains how generative AI works and why it is so computationally expensive. It also shows the different kinds of harms it can cause, from misinformation to environmental impact and labor issues. The goal of the study is to make us think critically about both the benefits and the consequences of generative AI, and to question whether the advantages are strong enough to outweigh the costs.

Case Study reading:
How generative AI works and how it fails

Discussion on environmental impact of GenAI

Question 1: Current impact of AI on the environment

The current environmental impact of AI is mainly connected to how much energy is required to generate even a simple response, compared to a search engine like Google, which are far less computation intensive. As the case study mentions, generating just a single token requires about a trillion operations. This scale of computation is only possible because of large data centers.

These data centers create multiple environmental pressures. They take up land that could otherwise be used for housing, agriculture, or even reforestation. They also rely heavily on electricity and on large amounts of water for cooling. This creates local consequences, such as water scarcity in nearby communities, and global ones, such as higher carbon emissions. There are also indirect effects, such as construction and maintenance bringing more traffic and pollution to areas that may have previously been isolated, and eventually the servers and machinery become e-waste.

Another major issue is outsourcing. Many of these facilities are being placed in the Global South, where environmental protections are not that strong. This allows companies to extract resources and externalize the costs of pollution, while the local people carry the burden.

Question 2: How it might change in the future

The demand for AI will most likely continue to increase, as companies view it as a way to reduce labor costs. Some improvements in energy efficiency might occur, but these would happen only when they also generate economic profit, not out of genuine concern for the environment. Even with more efficient hardware, the overall growth in AI usage is likely to cancel out those gains. In addition, the expansion of data centers will generate increasing amounts of e-waste as old machines are discarded. If resources in developed countries become scarce, corporations may turn to less-protected regions to meet their needs, such as the extraction of lithium from the Amazon, further deepening global environmental inequalities.

Question 3: Could AI help solve environmental problems?

AI could be useful for things like predicting natural disasters earlier, making traffic systems more efficient, or helping optimize energy grids. These changes could reduce waste and save resources. But the problem is that the solutions proposed by the use of AI come with their own costs. More data centers mean more electricity, more water, and sometimes even deforestation to clear space or get the minerals needed to build the machines. On top of that, companies are using AI to replace workers, which creates social and economic problems. Helping one side while damaging another doesn’t really solve the bigger issue, it just shifts the harm to a different place.

Question 4: How should policymakers respond to AI’s environmental impact?

Policymakers could prevent big tech companies from outsourcing environmental damage to weaker countries. They could regulate where data centers are built, for example, prioritizing colder climates where less energy is needed for cooling. They could also invest in research to make AI models smaller, or even reduce data dimension, make it more efficient, and less resource-hungry. These policies need to be strongly enforced, with companies held accountable.

New discussion question

Question

How do you personally balance the benefits of using AI tools with the environmental costs behind the scenes? Do you feel that your individual use matters, or do you see it as mainly the responsibility of corporations?

Reason and answer

I thought of this question because the reading mentioned how much resources it takes to generate a single token, which made me think about the gap between the technical side of AI and our everyday experience of it. It made me reflect on how something that can feel abstract, or even a bit like magic, still has real environmental consequences. Even when we can sense that the environment is being harmed, it often feels distant, yet our individual choices can still make a difference.

For my answer, I think individuals should consider the consequences of what they use, just like we do with other things in our lives. But the ultimate responsibility lies with big tech companies. It is similar to fast fashion where we’re encouraged to buy less or recycle clothing, but the real environmental damage comes from large companies producing cheaply and in huge quantities, often exploiting both people and resources.

Reflection

This exercise helped me see how the technical details in the case study connect to big-picture issues like climate change, land use, and global inequality. I realized that even if AI seems virtual, this “magic” thing that feels untouchable, it relies on very physical resources, like land, minerals, water, and electricity. It made me think about the human and environmental cost behind that just recently made its way into our daily life: the energy to train models, the materials for servers, and many others. Even though it feels abstract, our choices, whether as users or as a society, can directly influence these impacts.