You're getting more and more distanced from checkable information
From a hill in London where humans have tracked the heavens for three and a half centuries, the Royal Observatory Greenwich has offered a measured caution for our accelerating age: when machines answer before we have fully learned to ask, something essential in the architecture of human understanding may quietly dissolve. The concern is not with technology itself, but with the disappearance of the productive friction — the wrong turns, the redundant measurements, the questions no one thought to ask — that has historically been the hidden engine of discovery. History suggests that the most durable knowledge is often a byproduct of curiosity that could not justify itself in advance.
- The Royal Observatory warns that AI's gift of instant answers may be quietly dismantling the habits of questioning and evaluation that make genuine expertise possible.
- Early astronomers kept meticulous, seemingly wasteful records that an efficiency-optimized AI would have skipped — yet those records proved indispensable to scientists a century and a half later.
- As Google's AI Overviews and similar tools replace lists of sources with pre-digested conclusions, users are drifting further from the verifiable, checkable evidence that anchors reliable knowledge.
- The risk is not simply that AI occasionally fabricates — it is that it erases the scaffolding of evidence and the detours that sometimes lead to unexpected, reshaping discoveries.
- The Observatory's ongoing transformation, First Light, is a deliberate effort to rekindle the passion for open-ended inquiry that has driven 350 years of astronomical work.
The Royal Observatory Greenwich, with three and a half centuries of astronomical history behind it, has issued a careful warning: the convenience of instant AI answers may be eroding the human habit of asking questions in the first place.
Director Paddy Rodgers framed the concern not as technophobia but as historical observation. Early astronomers performed painstaking, seemingly redundant work that a machine optimized for efficiency would never bother with — yet that accumulated data became invaluable over a century later, when researchers used it to verify theories the original observers could never have imagined. Human curiosity pursued what seemed wasteful; the results outlasted their creators.
Rodgers acknowledged AI's genuine power — DeepMind's AlphaFold2 and its mapping of nearly all known proteins stands as proof — and recognized that academics and students find real value in using AI to challenge assumptions and deepen learning. But he drew a distinction between tools that extend human capability and systems that replace human effort entirely.
The subtler risk, he argued, lies in how the path to answers is disappearing. Where a Wikipedia reader could follow citations back to original sources and encounter unexpected context, AI systems deliver conclusions without the scaffolding of evidence. As Google's AI Overviews displace traditional search results, users receive synthesized answers rather than sources to explore — moving from a model where we find information to one where information finds us, pre-digested and final.
Rodgers is not arguing against AI in science or learning. He is arguing for preserving the friction — the dead ends, the detours, the questions we did not know to ask — that human curiosity has always generated, and that has quietly shaped the most durable discoveries in the Observatory's long history.
The Royal Observatory Greenwich, perched on a hill in London with three and a half centuries of stargazing behind it, has issued a quiet but pointed warning: the convenience of instant AI answers may be eroding something harder to replace than we realize—the human habit of asking questions.
Paddy Rodgers, who directs the Royal Museums Greenwich group overseeing the Observatory, framed the concern not as technophobia but as historical observation. The institution's long record of astronomical discovery, he argued, reveals what gets lost when we outsource thinking to machines. "A reliance solely on instant answers risks losing the habits of questioning and evaluation that underpin knowledge, expertise and innovation," he said. The Observatory is currently undergoing a transformation called First Light, designed to capture and reinterpret the passion that has driven astronomers across 350 years of work.
Rodgers acknowledged that technological innovation has always been central to scientific progress. But he drew a distinction between tools that extend human capability and systems that replace human effort entirely. Early astronomers, he noted, did things a machine would never bother doing—painstaking observations, redundant measurements, work that seemed unnecessary at the time. Yet that accumulated data became invaluable a century and a half later, when researchers used it to verify theories about Earth's navigation that the original observers could never have imagined. An AI system, optimized for efficiency, would likely skip such apparently wasteful labor. A human, driven by curiosity or thoroughness, pursued it anyway—and created a resource that outlasted its creator's lifetime.
The tension between AI's promise and its limits has become increasingly visible. DeepMind's AlphaFold2 predicted the structures of nearly all known proteins, a breakthrough that would have been impossible without computational power. Reid Hoffman, LinkedIn's co-founder, has described AI as a transformation of cognitive work itself, useful for testing ideas: "I think X, are you against it?" Academics and students report genuine benefits—the ability to challenge assumptions, work through problems collaboratively, focus attention on deeper learning rather than rote research.
Yet the Observatory's concern points to a subtler risk. When Wikipedia emerged, a user could follow a citation back to a source, verify a claim, encounter the original context. AI systems compress that journey. They deliver answers without the scaffolding of evidence, without the detours that sometimes lead to discovery. Rodgers observed that "you're getting more and more distanced from relatable or checkable information." The problem is not that AI lies—though it sometimes does—but that it obscures the path by which answers were reached.
This matters because the way we search is changing. Google's AI Overviews now appear at the top of search results, replacing the old lists of links. TikTok and X are running similar experiments. Users increasingly receive synthesized answers rather than sources to explore. The shift is subtle but consequential: it moves us from a model where we find information to one where information finds us, pre-digested and final.
Rodgers is not arguing against AI in science or learning. He is arguing for preserving the friction—the questioning, the dead ends, the unexpected discoveries—that human curiosity generates. The Observatory's history suggests that some of the most valuable work happens not when we get quick answers, but when we pursue questions we did not know to ask, and stumble onto results that reshape what we thought we knew.
Notable Quotes
A reliance solely on instant answers risks losing the habits of questioning and evaluation that underpin knowledge, expertise and innovation.— Paddy Rodgers, director of Royal Museums Greenwich
When used responsibly, AI tools enable students to direct their attention to the more important parts of learning, but outsourcing thinking entirely would highlight the technology's limits.— Oxford Brookes University lecturer
The Hearth Conversation Another angle on the story
When Rodgers talks about early astronomers doing "unnecessary" work, what does he actually mean? Why would they do something a machine wouldn't?
Because they were curious. They didn't know what they were looking for. A machine is built to solve a specific problem efficiently. A human might measure something three times when once would suffice, or record observations that seem tangential, because they're not sure what matters yet.
But couldn't an AI system be programmed to do that kind of exploratory work?
Theoretically, yes. But in practice, AI is optimized for efficiency and relevance. It's designed to give you the answer you're asking for. It's not designed to waste time on hunches or to pursue questions that seem unproductive. The human mind does that naturally.
So the real worry isn't that AI is bad at science. It's that we'll stop doing the kind of thinking that leads to unexpected discoveries.
Exactly. If every student uses AI to answer every question, they never develop the habit of sitting with a problem, of being confused, of following a thread that might lead nowhere. Those habits are how you learn to think.
Rodgers mentions that old astronomical data became useful 150 years later. How often does that actually happen?
Often enough that it matters. Science is cumulative. You don't know which observations will matter to future researchers. That's why the Observatory kept such meticulous records. But if we're only keeping the answers that seem relevant right now, we're throwing away the raw material that future breakthroughs depend on.
Is he saying we shouldn't use AI at all?
No. He's saying we should be intentional about when we use it. Use it to test an idea, to challenge your thinking, to work through a solution. But don't use it to replace the work of asking the question in the first place. The asking is where the learning happens.