Faith-based computing versus the unnatural science

The new tool, the same old question

It has been reported that developers are being pushed by management to adopt large language models for everyday coding tasks—roughing out frameworks, generating unit tests, even churning boilerplate. The author of a recent discussion reposted on Lobsters and linked to an ACM Queue piece pushes back: whether a human or an LLM types the keys is less interesting than understanding what the code does, how it was built, and when to look under the hood. Helpful? Sure. A cure-all? Not remotely. Tools accelerate work; they don't replace the thinking that makes software reliable.

Belief is not an engineering strategy

The piece drills into a phrase that keeps surfacing: belief. It has been reported that some evangelists frame LLMs with near-mystical language; the writer—summoned in the piece as KV—calls this nonsense and explicitly rejects the notion of "believing" in AI, likening it to faith in the Tooth Fairy or the Flying Spaghetti Monster. Allegedly, even heavyweight algorithmic thinkers who examine these systems are doing rigorous applied work, not preaching doctrine. The reminder is sharp: science and engineering demand reproducible methods, not wishful thinking or luck.

Computing as an "unnatural" science

The argument lands on a neat pivot: computing studies human-made artifacts and should be treated as an "unnatural science"—one that must still obey the scientific method even if its subjects were invented rather than discovered. If outputs vary randomly, that’s not progress; it’s chance dressed up as progress. The takeaway is practical and a little stern: use LLMs as tools, interrogate their results, and don't let hype replace craftsmanship. So what now—double down on curiosity and caution? That sounds like a plan.

Sources: queue.acm.org, Lobsters