Metamorphic testing
Metamorphic testing helps solve the problem of not knowing what the expected results should be. By defining relationships between inputs and outputs, we can check that AI models behave consistently, making them more reliable.
7th April 2025
Metamorphic testing is a way to test systems, especially AI models, when you don’t know exactly what the correct output should be. Instead of checking for a “right answer,” you define rules or patterns (called metamorphic relations) that should hold true when you change the input in a specific way.
For example:
For example:
- If you slightly brighten an image, the object detection result should stay the same.
- If you replace a word with a synonym in a sentence, the sentiment analysis should still be positive.
Why is this good for us testers?
- Test systems without needing a perfect oracle (i.e., when expected outputs are unknown or hard to define).
- Catch bugs or inconsistencies in non-deterministic systems like machine learning models.
- Improve confidence in model behavior by checking for logical consistency.
- You can design smart test cases even when you don’t have labeled data.
- It’s great for exploratory testing of AI features.
- You can automate many of these checks to scale your testing.
5th November 2025
Discover how integrating SAST with comprehensive test management enables your team to ship safer software sooner.
Explore MoT
Boost your career in software testing with the MoT Software Testing Essentials Certificate. Learn essential skills, from basic testing techniques to advanced risk analysis, crafted by industry experts.
Debrief the week in Quality via a community radio show hosted by Simon Tomes and members of the community
