Judgment in the Age of AI · Article 24
A summary does more than reduce word count; it reallocates attention. It may emphasise conclusions, compress evidence, remove objections and turn tentative language into certainty. The result need not be obviously false to make the author appear simpler and more absolute than the original.
Run a compression-loss test
| Compressed element | Possible distortion |
|---|---|
| Qualifications | “In some circumstances” becomes a universal conclusion |
| Strength of evidence | Correlation becomes causation |
| Opposing views | A disputed issue appears settled |
| Authorial tone | A hypothesis, suggestion or question becomes an assertion |
| Order of argument | The conclusion is detached from the conditions that produced it |
Compare claims, not merely length
Extract three to five central sentences from the summary and locate support in the original. Search especially for “but”, “however”, “may”, “limitation” and “unless”. Rewrite any confident summary sentence that lacks equally strong wording in the source.
A two-summary method for long texts
- Generate a summary of the author’s claims
- Generate a separate summary of evidence and limitations
- Have a person check whether both preserve the same boundaries
A good summary is not simply shorter. It preserves the force and limits of the original while becoming shorter.
References
- NIST: Generative Artificial Intelligence Profile
- UNESCO: Guidance for Generative AI in Education and Research
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.