Responsible Use of Artificial Intelligence in ADICFito
The development of ADICFito incorporates artificial intelligence models to support digital analysis of phytoplankton cells. Aware of the importance of transparency, traceability, and responsible use of these technologies, the following measures were adopted:
1. Phytoplankton cell recognition
Model used: Azure Custom Vision models were used for image classification.
Taxonomic scope: The system classifies cells at the Family level.
Data inclusion criteria: Only families with more than 100 available images were included; the rest were grouped into a negative class ("others").
Evaluation metrics: Model performance is reported in terms of accuracy, precision, and recall per family, in order to transparently reflect both its strengths and limitations.
Limitations and responsible use:
- The model does not replace expert validation, but rather serves as support to expedite routine analyses.
- It is recognized that the quality and representativeness of the dataset condition the results, so the image set will continue to be expanded and diversified.
2. Automatic generation of taxonomic descriptions
Model used: GPT-4 was used for generating accessible descriptions of phytoplankton species included in the platform.
Objective: Reduce knowledge gaps for non-expert users through a catalog with clear descriptions, avoiding excessively technical language.
Responsible use:
- Guidelines were established to avoid biases, interpretation errors, or taxonomic inconsistencies.
- The model's role is to complement the work of specialists, not to replace their scientific judgment.
3. Commitment to continuous improvement
ADICFito maintains a commitment to updating its models and databases, promoting transparency in performance metrics and data provenance. The goal is to ensure that the use of artificial intelligence contributes reliably, ethically, and responsibly to the monitoring of harmful algal blooms.