Photo from pexels.com
Artificial intelligence (AI) is revolutionizing various industries, including healthcare. The use of AI capabilities, such as natural-language generation, computer vision, and robotic process automation, is growing exponentially.
In a recent McKinsey report for example, it has been shown that organizations are increasingly making use of AI capabilities, with the average number of AI technologies used expected to double from 1.9 in 2018 to 3.8 in 2022.
This growth is reflective of the widespread use of AI in fields like natural language generation and computer vision. Natural-language text understanding has advanced rapidly, moving from a mid-tier position in 2018 to ranking just behind computer vision in 2022, while robotic process automation and computer vision have consistently been the most widely adopted among these various capabilities.
In the field of herpetology and global health, AI can play a vital role in identifying snake species, which could have a significant impact on snakebite victims and conservation efforts.
AI in Snake Identification
Molecular methods such as the use of immunoassays for identifying snakes has its limitations, especially in resource-poor areas. Identification of snake species based on pattern recognition, on the other hand, although it is essential for medical professionals to do so in order to provide appropriate care, can be challenging. This gap can be closed with the help of AI models built on top of computer vision methods. While there are already AI models that can recognise common birds, fish, and butterflies, few have attempted to do the same for snakes, and those that have have focused on narrow taxonomic or geographical niches.
A recent study by Bolon et al. (2022) developed an AI model to identify snakes worldwide. The model achieved an impressive macro-averaged F1 score of 92.2% and demonstrated accurate classification of venomous and non-venomous lookalike species from Southeast Asia and sub-Saharan Africa. This technology could support snakebite victims, healthcare providers, zoologists, conservationists, and nature lovers across the globe.
F1 score is a metric used to evaluate the performance of classification models, particularly in situations where there is an imbalance in the number of samples between different classes. It is a combination of two other metrics: precision and recall.
Precision basically means: of all the positive predictions I made, how many of them are truly positive?
Precision = Number of True Positives (TP) divided by the Total Number of True Positives (TP) and False Positives (FP)
Whereas recall means: of all the actual positive examples out there, how many of them did I correctly predict to be positive?
Recall = Number of True Positives (TP) divided by the Total Number of True Positives (TP) and False Negatives (FN).
The F1 score balances both precision and recall by taking their harmonic mean, providing a single value that represents the model's performance. The F1 score ranges from 0 to 1, where 1 indicates perfect precision and recall, and 0 means the model fails to make any correct predictions.
Limitations of AI ModelsAddressing AI Bias
Human, systemic, and computational biases can affect AI models, impacting their usefulness and trustworthiness. Organizational leaders need to ensure AI systems improve human decision-making and reduce bias. Two imperatives for action include responsibly using AI to improve traditional human decision-making (this is where the human brains are still very much relevant) and the need of accelerating progress in addressing biases in AI.
Researchers also need to work on various techniques to ensure AI systems meet fairness definitions. One promising technique is counterfactual fairness, which guarantees that a model's decisions remain the same in a counterfactual world where sensitive attributes are changed.
Conclusion
AI has the potential to transform the medical and conservation fields, particularly in snake envenomation. The AI model developed by Bolon et al. (2022) represents a significant step forward in snake identification, ultimately benefiting snakebite victims, healthcare providers, and conservationists. However, addressing the limitations and biases in AI models remains a critical concern to fully harness the power of AI in these fields.
References
- Bolon, I., Durso, A. M., Botero Mesa, S., Tollefson, S., Omori, R., Zurell, D., & Alcoba, G. (2022). An artificial intelligence model to identify snakes from across the world: Opportunities and challenges for global health and herpetology. PLOS Neglected Tropical Diseases, 16(2), e0010647. https://doi.org/10.1371/journal.pntd.0010647
- Leong, K. (2022). Micro, macro & weighted averages of F1-score clearly explained. Towards Data Science. Retrieved from https://towardsdatascience.com/micro-macro-weighted-averages-of-f1-score-clearly-explained-b603420b292f
- Manyika, J., Silberg, J., & Presten, M. (2019). What do we do about the biases in AI? Harvard Business Review. Retrieved from https://hbr.org/2019/10/what-do-we-do-about-the-biases-in-ai
- McKinsey & Company. (2022). The state of AI in 2022 and a half-decade in review. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-in-2022-and-a-half-decade-in-review#/
- McKinsey & Company. (n.d.). What is generative AI? https://www.mckinsey.com/featured-insights/mckinsey-explainers/what-is-generative-ai#
- National Institute of Standards and Technology. (2022). There's more to AI bias than biased data: NIST report highlights. https://www.nist.gov/news-events/news/2022/03/theres-more-ai-bias-biased-data-nist-report-highlights
