OpenAI CEO Sam Altman recently posted a blog revealing that the average ChatGPT query requires approximately 0.000085 gallons of water, or one-fifteenth of a teaspoon. The quote, included in a longer post about the effects of AI on the world, provides a startling insight into the environmental impact of large language models.
The cost of energy to use AI was also mentioned in the post by Altman, where the author noted that a single query in ChatGPT requires about 0.34 watt-hours. That is equivalent to the amount of energy that an oven would use in more than a second or that a high-efficiency lightbulb would use in a few minutes. Altman also envisioned a future where AI will be even more affordable and effective, stating that the cost of intelligence will approach the cost of electricity eventually.
The post, however, leaves one wondering how these figures were arrived at. OpenAI has not replied to queries seeking an explanation of the model used to arrive at Altman estimates. The environmental price of AI is an increasing issue, particularly as the technology is scaled at an unprecedented rate.
Some researchers estimated that AI systems would soon exceed Bitcoin mining in energy use in 2024. The Washington Post partnered with researchers last year and found that it consumed more than a bottle of water to generate a basic 100-word email with GPT-4. Another important aspect that was brought up by the investigation is that the water consumption differs greatly depending on the location of data centers, which tend to use local cooling systems that influence the overall consumption.
The arguments presented by Altman attempt to put the AI resource consumption into perspective and express it in terms that people can relate to, yet they also demonstrate that it should be more transparent. With the increasing introduction of AI into our everyday lives, one of the crucial factors in sustainable innovation will be the comprehension of—and counteracting—its effect on the environment.
