Two letters published by The Guardian in May 2026 capture a useful disagreement about AI-assisted thinking. Richard Thackeray describes AI as a research aid that broadens curiosity. Software engineer and musician Phil Snell warns that polished output can conceal a loss of understanding, ownership and craft. These are personal perspectives, not scientific proof—but the tension is real for organisations adopting AI.
Offloading search, summarisation or first drafts can give people more time to compare options and ask better questions. That benefit appears only when the user remains able to test the result. Faster production is not the same as better reasoning, and fluent language is not evidence of accuracy.
The strongest use of AI is therefore collaborative: a person defines the objective and constraints, the system proposes material, and a responsible person verifies sources, assumptions and consequences.
When teams accept outputs they cannot explain, they accumulate cognitive debt. The immediate task may be completed, but the organisation becomes less able to maintain the work, detect subtle errors or respond when conditions change.
In software, AI-generated code should be treated like an untrusted contribution. It needs review, tests, dependency checks and a maintainer who understands it. In research and writing, the equivalent controls are source verification, clear attribution, independent reasoning and removal of unsupported certainty.
Human-centred AI is not achieved by adding a “human in the loop” label. The person needs enough information, time and authority to disagree with the system. A nominal reviewer who is expected to approve hundreds of outputs cannot provide meaningful oversight.
Teams should also preserve spaces where people practise core skills without AI assistance. Junior staff need opportunities to form mental models, not only to edit generated answers. Experienced staff need incentives to report uncertainty and reject weak output rather than reward speed alone.
AI does not automatically replace thought, and it does not automatically deepen it. The outcome depends on how the tool is used, what the organisation rewards and whether people retain responsibility for the result. The aim should be more capable human judgement—not the maximum possible volume of generated output.
Explore more evidence-led analysis in ITnovati’s AI coverage.
The Guardian: “AI is changing how we think, not replacing it”
Editorial disclosure: AI tools may have assisted research, drafting or editing. ITnovati remains responsible for the published text. Time-sensitive technical, legal and product claims should be checked against the linked primary sources.
ITnovati