This week, OpenAI CEO Sam Altman tackled worries regarding AI’s environmental footprint during an event organized by The Indian Express.
Firstly, Altman – who was attending a significant AI summit in India – dismissed concerns about AI’s water consumption as “totally fake,” while admitting it was a genuine problem when data centers employed “evaporative cooling.”
“Since we no longer use that method, you often encounter claims online such as, ‘Avoid ChatGPT; each query consumes 17 gallons of water’ or similar statements,” Altman stated. “These assertions are entirely false, utterly absurd, and completely detached from reality.”
He further commented that it’s “reasonable” to worry about “the overall energy consumption, not just per query, given AI’s widespread global adoption.” According to Altman, this necessitates a rapid global shift “towards nuclear, wind, and solar energy sources.”
As tech firms are not legally obligated to report their energy and water usage, researchers have been undertaking independent studies on the matter. Furthermore, data centers have been linked to escalating electricity costs.
Referring to a prior discussion with Bill Gates, the interviewer inquired if it was accurate that one ChatGPT query consumes energy equivalent to 1.5 iPhone battery charges, prompting Altman to respond, “It’s absolutely nowhere near that amount.”
Altman additionally voiced his dissatisfaction, asserting that numerous discussions surrounding ChatGPT’s energy consumption are “unjust,” particularly when they concentrate on “the energy expenditure required to train an AI model, compared to the energy a human expends for a single inference query.”
“However, training a human also demands considerable energy,” Altman observed. “It involves approximately 20 years of life and all the sustenance consumed throughout that period to achieve intelligence. Moreover, it required the extensive evolution of hundreds of billions of individuals throughout history, who learned to evade predators, master scientific principles, and more, ultimately leading to your existence.”
Therefore, from his perspective, the appropriate comparison is, “Once an AI model is trained, how much energy does it consume to answer a question posed to ChatGPT, versus a human doing the same? And it’s likely that AI has already achieved comparable energy efficiency when measured this way.”
The complete interview is available for viewing below. The segment discussing water and energy consumption commences at approximately 26:35.