OpenAI’s CEO Sam Altman tackled worries regarding AI’s ecological footprint this week during an event organized by The Indian Express.

For starters, Altman — present in India for a significant AI conference — stated that fears about AI’s water consumption are “completely fabricated,” although he conceded it was a genuine concern when “we used to implement evaporative cooling in data centers.”

“Now that we’ve stopped that practice, you come across these claims online that say, ‘Don’t utilize ChatGPT, it consumes 17 gallons of water for every query’ or similar,” Altman remarked. “This is utterly false, completely ludicrous, without any basis in reality.”

He pointed out that it’s “reasonable” to worry about “the total energy usage — not per individual query, but overall, given that the world is increasingly using AI.” In his opinion, this indicates the need for rapid adoption of nuclear or wind and solar energy sources.

There’s no statutory obligation for tech firms to reveal their energy and water consumption, prompting scientists to seek independent assessments. Data centers have also been linked to escalating electricity costs.

Referencing a prior discussion with Bill Gates, the interviewer inquired if it’s true that a single ChatGPT query requires the same energy as 1.5 iPhone battery charges, to which Altman countered, “There’s no way it consumes that much energy.”

Altman also expressed that much of the discourse surrounding ChatGPT’s energy consumption is “unjust,” particularly when it emphasizes “the energy required to train an AI model, compared to the energy expense for a human to do one inference query.”

“However, training a human also requires substantial energy,” Altman noted. “It demands roughly 20 years of life and all the nourishment consumed during that period before one becomes intelligent. Furthermore, it necessitated the extensive evolution of the 100 billion individuals that have ever existed, who learned to avoid predators and to understand science and more, to create you.”

Thus, in his opinion, the just comparison is, “When you pose a question to ChatGPT, how much energy is consumed once its model is trained to respond to that inquiry compared to a human? And likely, AI has already matched energy efficiency when assessed in that manner.”

The complete interview is available for viewing below. The discussion concerning water and energy consumption starts around 26:35.