It is essential to avoid failures; we tell you how!
AI systems are only as ethical as the data models through which they have been trained. In this infographic, we explain their main biases and how to detect them quickly.
Términos de uso
You can use the resource for personal or informative use with attribution to the entity red.es following our terms of use.
It is essential to avoid failures; we tell you how!
AI systems are only as ethical as the data models through which they have been trained. In this infographic, we explain their main biases and how to detect them quickly.
1. Data biases: Lack of representation of certain groups.
-
Formulate questions varying gender, age, or context and observe if the answers change.
2. Human prejudices: AI can reproduce stereotypes present in their trainers.
-
Verify neutral responses on sensitive topics.
3. Algorithm biases: Certain results may be favored.
-
Repeat similar questions and evaluate coherence.
4. Manipulation biases: Data may be adjusted to obtain positive results.
-
Consult how the response was generated and its sources.
5. Exclusion bias: elimination of data considered irrelevant.
-
Pose uncommon situations and analyze the answers.
6 Labelling bias: Incorrect classification of training information.
-
Check for inaccuracies or mixing of concepts in the responses.
7. Feedback bias: Our interactions influence future responses.
-
Observe patterns or evolution in the answers.
Now that you know the main biases, put them to the test and avoid mistakes!
Other infographics
