A classic classification model trained to assign claims: Considering model-type differences, which choice best
A bounded, auditable classification into fixed categories is best served by a classic classification model.
The question
An insurer needs to predict whether claims fall into one of five predefined fraud-risk categories, with outputs that are consistent and straightforward to audit. Considering model-type differences, which choice best fits this task?
Preparing for AIGP? Take the free 5-min readiness quiz →
- A generative language model prompted to compose an open-ended narrative describing each claim's riskA generative model produces variable free-text and is harder to audit for a fixed-category classification need.
- A multimodal generative model that also interprets the images attached to each claim fileMultimodal generation adds unneeded capability and variability for a task with fixed textual categories.
- A classic classification model trained to assign claims to the fixed, predefined risk categories ✓A classic classifier maps inputs to a fixed label set with consistent, auditable outputs, matching this bounded task.
- An open-ended generative model tuned to draft entirely new fraud-investigation policiesPolicy drafting is a different task entirely and does not perform the required categorical prediction.
The trap
Defaulting to a generative model for a task that a classic, more auditable classifier handles better. How to remember it
A bounded, auditable classification into fixed categories is best served by a classic classification model.
How many of these would you get right?
One of 1581 AIGP questions on Certsqill. Take a free five-minute check and see your score per domain — not one number, but which section to open tonight.
Test your AIGP readiness — freeMore Understanding How to Govern AI Deployment and Use questions
- Ethical considerations: Which use-case context factor most directly frames this concern in the deployment →
- An on-premise deployment that keeps the model and all: Considering deployment options, which choice best →
- An impact assessment evaluating how the system could: Which assurance activity is most appropriate to perform →
- All 424 Understanding How to Govern AI Deployment and Use questions →
Part of the Certsqill AIGP question bank · Understanding How to Govern AI Deployment and Use ·
Every answer, right and wrong, comes with its own explanation.