To address these issues, the article suggests several solutions, including the creation of checklists for reporting AI-based science, making methods and data fully open, and changing cultural norms about how data are presented and reported. However, it also acknowledges that these solutions may not be sufficient and that the field of AI and ML still has a long way to go in terms of maturity and reliability. The article concludes by expressing hope that, with time, the research community will be able to better understand and use AI and ML, much like how the aerospace industry has evolved to make airplanes trustworthy.
Key takeaways:
- Artificial Intelligence (AI) and Machine Learning (ML) are powerful tools for scientific research, but their misuse or misunderstanding can lead to misleading claims and irreproducible results. This is particularly problematic in biomedical research, where misclassification could have serious consequences.
- Common issues include data leakage, where there is insufficient separation between the data used to train an AI system and those used to test it, and the use of AI models that do not reflect real-world data. These problems can lead to AI systems that perform well in tests but fail in practical applications.
- Researchers have proposed solutions such as checklists for reporting AI-based science, making methods and data fully open, and changing cultural norms about how data are presented and reported. However, these solutions face challenges such as privacy concerns, lack of understanding about AI among researchers, and reluctance to release code for public scrutiny.
- Despite the problems, some researchers believe that the issues with AI and ML will resolve themselves over time, as the field matures and researchers gain a better understanding of how to use these tools. They argue that the current issues are similar to the teething problems faced by other new scientific methods in the past.