A three-week undergraduate unit on source verification with AI tools — Yila research notes https://yila.ai
|
|
1 maand geleden | |
|---|---|---|
| README.md | 1 maand geleden |
Students now arrive at university with AI tools already part of their study habits. Banning those tools rarely works, and pretending they do not exist leaves students to learn their weaknesses the hard way, usually through a fabricated reference in a graded essay. A better approach is to teach source verification directly, using AI output as material to examine. This guide outlines a three-week unit that instructors can adapt for most undergraduate disciplines. It uses examples from evidence-traceable tools such as Yila alongside general-purpose chatbots, so students can see the difference between output they can check and output they cannot.
By the end of the unit, students should be able to:
These goals are not about AI specifically. They are core information literacy skills that have become more urgent.
Start without any AI at all. Give students a short, well-written newspaper article on a topic related to the course and ask them to highlight every factual claim. Then ask them to sort the claims into types: statistics, causal claims, descriptions of events, expert opinions and generalisations.
For each type, discuss what counts as good evidence. A statistic needs a dataset or study with a stated method. A causal claim needs a design that can support causation. An expert opinion needs a named expert with relevant expertise. This framework becomes the checklist for the rest of the unit.
The homework is to trace three claims from the article to their original sources and report whether the sources say what the article says. Students are usually surprised by how often the answer is "partly".
In the second week, students generate AI answers to a question related to the course and then audit them. It works well to have students use two kinds of tool: a general chatbot and a research assistant that cites sources at the sentence level.
The audit has four steps:
Students record their findings in a shared table. When the class pools results, patterns emerge quickly. General chatbots tend to produce fluent answers with a mix of real, garbled and fabricated references. Tools designed for traceability produce fewer unsupported sentences, and the ones that exist are faster to spot because each claim is linked to a passage. The lesson is not that one tool is good and another bad, but that verifiability is a property students should demand.
The final week turns to practice. Students write a short evidence summary, around 800 words, on a question they choose within the course topic. They may use any AI tool, under three conditions:
The research log is the most valuable part of the assessment. It makes the process visible and rewards careful checking rather than polished prose. Students who used a sentence-linked tool such as Yila often report that verification took minutes rather than hours, which is a useful observation in itself: the design of a tool can make good practice easier or harder.
A simple rubric with four criteria works well:
Weighting process and traceability alongside accuracy signals that verification is part of the work, not an optional extra.
Several misconceptions come up repeatedly and are worth addressing explicitly:
The unit fits naturally into first-year writing courses, research methods modules and discipline-specific seminars. In science courses, use primary research articles and focus on methods and statistics. In humanities courses, use archival or critical sources and focus on quotation and interpretation. In professional programmes such as nursing, business or journalism, use practitioner sources and focus on how evidence informs decisions.
Students will use AI tools throughout their careers. The habits they form now, whether they verify or trust blindly, will persist. A short, practical unit that treats AI output as something to examine rather than something to fear builds those habits early. The central lesson is simple: a claim is only as good as the evidence you can show for it, and the best tools are the ones that make showing that evidence easy. That principle, rather than any single product, is what the unit teaches, though instructors looking for an example of sentence-level citation design can find one at https://yila.ai.