Trivial AI use can have devastating climate consequences

THE future of humanity, we are told, will be irrevocably shaped by artificial intelligence (AI). That same future is also under existential threat from global warming and its disastrous consequences – a threat that is being exacerbated by the unthinking use of AI.

Of course, there are many powerful and life-changing uses for AI, from cancer research to predictive data analytics.

Perhaps the most widespread use today, however, is arguably the most trivial: generative AI (GenAI), not least in the form of large language models (LLMs) that spit out text based on spotting linguistic patterns, rather than engaging with meaning.

Even if one sets aside the legal and ethical issues surrounding GenAI models and their use of copyrighted material, their meteoric rise has had regrettable practical consequences.

Some have decried how LLMs encourage intellectual laziness, with students or entry-level employees outsourcing what used to be fundamental work, and thus never developing the required skills: understanding and summarising information, or turning rough ideas into polished text.

Due to the looseness of the term “AI”, LLMs are also drawing attention – and perhaps resources – away from more useful forms.

At an SG60-themed conference on Tuesday (Jul 29), Prime Minister Lawrence Wong noted that not all AI applications are equally useful, saying: “Most of us use AI the way we use Google – that is not exactly the best way to use AI or to harness the potential of AI.”

He stressed that LLMs are just one small part of AI, with other areas having far more potential.

Even as Singapore encourages broad-based adoption of AI, the country must “think equally hard about applying technologies like AI in a meaningful and deliberate manner that creates jobs for Singaporeans”, he said.

The climate cost

For those with no attachment to the old-fashioned notion of thinking for oneself, the widespread use of GenAI might not be an issue.

The true problem is that trivial GenAI use exacts a real environmental cost, for often dubious gains.

Take Grok, the GenAI model created by X, formerly known as Twitter. Since late 2024, the data centre powering this LLM has made headlines for guzzling electricity and water, polluting nearby waterways and emitting greenhouse gases.

And this is just one of many resource-hungry data centres that are crucial for powering AI models. Such facilities consume high amounts of water and electricity to keep their servers cool and running.

In the current climate crisis, the proliferation of data centres might seem almost profligate. Even as nations try to cut emissions and energy use, every ChatGPT query adds to the carbon burden.

Of course, as with anything else, AI use should be subject to cost-benefit analysis. Many data centres may be supporting truly meaningful work: AI applications that reap concrete gains, for companies or for the government.

When smart factories use data analytics to reduce energy consumption and cut waste, for instance, the gains from doing so should more than offset the cost of powering such AI tools.

In contrast, it is depressing to consider the carbon cost of the trivial GenAI queries that are being made en masse each day.

And even when AI use is purported to improve productivity, we should question the real savings being achieved. Is the time saved in getting ChatGPT to produce a corporate-speak e-mail, for instance, really worth the resources burned?