A typology of four kinds of answers from AI systems for assessing their reliability
A librarian at University of Virginia proposes dividing answers from AI systems into four types – factual, interpretive, constructive and strategic – with a different verification process for each.
Leo S. Lo, university librarian and dean of libraries at University of Virginia, proposes a typology of four kinds of answers that AI systems provide to users in his article for The Conversation. He first described the typology in the Journal of Academic Librarianship. According to the author, AI presents all four types of answers in an equally fluent and authoritative tone, so the differences between them can easily go unnoticed, even though each type requires a different approach to assessment.
The author distinguishes between factual answers (a verifiable claim that can be checked against a source – e.g. the date an institution was founded), interpretive answers (the answer draws on evidence but depends on which evidence was included and which was omitted – for example, the question of how much screen time is too much for a teenager), constructive answers (the answer is created, rather than objectively right or wrong, such as a draft cover letter or eulogy, where what matters is suitability for the purpose, audience and style) and strategic answers (the answer combines information with judgment about goals, risks and personal circumstances – for example, whether to take aspirin daily – and may be well formulated without being factually correct or suitable for the person concerned).
The author illustrates these categories with two of his own searches using AI-generated answers in the Google search engine: a query about the amount of screen time for teenagers and a query about taking aspirin daily. In both cases, according to the author, the answer from Google provided context, highlighted complications and offered more detailed guidance if additional information was supplied, while the type of answer could shift from factual to interpretive or strategic within a single response.
For each type, the author recommends a different verification process: for factual answers, follow the cited source; for interpretive answers, ask what the strongest evidence for a different conclusion is; for constructive answers, assess suitability for the purpose and audience; and for strategic answers, find out what additional information about the user could change the advice and whether a qualified professional should be involved.
Why it matters
Recognizing the type of answer prevents an AI answer from being mistakenly treated as an unequivocal fact when it is actually an interpretation, a creative proposal or strategic advice that depends on personal circumstances. This is especially relevant to answers generated directly in search engines, where the user does not actively request an AI answer and can easily give it the same credibility as a verified fact.
Relevant practical impact
What this means
For individuals
When making a decision based on an answer from an AI system (e.g. to a health or personal question), first identify the type of answer and then choose how to verify it, rather than relying on its fluent and persuasive tone.
Check the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.