Why people may disagree
People often approach AI’s impact through one lens, making a for-or-against position feel obvious. Looking across the evidence, scales, trade-offs and our own assumptions reveals a more nuanced picture.
The same words can describe different objects.
An “AI footprint” may refer to one request, one model, a product, a provider’s fleet, a data centre or the whole sector. It may include only active accelerators, or extend to cooling, electricity generation, hardware manufacturing and mineral extraction.
Two figures can both be valid while answering different questions. Their disagreement may come from the object and boundary chosen rather than from an error in either calculation.
Per-request and total impact tell different stories.
A service can use less energy for a comparable request while the total electricity used by the wider system increases. Total impact also depends on how many people use AI, how often they use it, which models they choose and how much computation each task requires.
Focusing on efficiency can make progress visible. Focusing on total demand can make growth visible. Neither view is sufficient on its own.
Precision can conceal assumptions.
Some results come from measurements of live systems. Others come from laboratory measurements, engineering models, inferred hardware or scenarios about future demand. Decimal places do not make these forms of evidence equivalent.
Carbon accounting can also change a result without changing the electricity physically consumed. Water figures can change when off-site electricity generation or local scarcity is included. Reading the assumptions is part of reading the number.
Facts do not assign importance.
People differ in how essential they consider AI, which uses they value, what future benefits they expect and which environmental costs they consider acceptable. These differences can remain even when the underlying measurements are understood.
Evidence can challenge unsupported claims and clarify consequences. It cannot, by itself, decide the importance AI should have in our lives.
Nuance does not mean that every claim is equal.
A nuanced view still distinguishes measurements from estimates, historical observations from projections and transparent methods from unsupported assertions. It also refuses to turn incomparable figures into rankings.
The purpose of nuance is not to avoid a position. It is to understand what supports that position, what remains uncertain and where personal priorities enter the reasoning.