A Geologist on Working with AI: How Not to Destroy the Science You Love

In an article in Science, a geologist explains how he integrates AI into his work while maintaining scientific rigor. This is not a manifesto claiming that “AI will replace scientists,” but rather a pragmatic analysis of where language models truly speed up routine tasks and where their use becomes a pitfall.

The main value of this experience for engineers and developers lies in establishing clear boundaries for its application. AI excels at processing large datasets, detecting anomalies, and generating hypotheses for testing. But it does not replace interpretation, domain knowledge, or critical evaluation of results. When a model produces a plausible answer, the temptation to accept it as fact is greatest—and this is precisely where science begins to erode.

Practical conclusion: AI in research acts as an amplifier, not a replacement. You can delegate synthesis and the exploration of options to it, but the responsibility for the conclusion always remains with the human. If you skip the critical verification stage to move faster, you’re not optimizing the process—you’re simply producing subpar science.

Source: Hacker News - Newest: ""AI" "LLM""