From Narrow AI to Super Intelligence: Shaping the Future of Animal Health
published on
05/31/2025
written by
B. Dharmaveer Shetty
Dharmaveer is a veterinarian-epidemiologist who oversees the World Organisation for Animal Health (WOAH)’s global wildlife networks and its Emerging Diseases Group Secretariat, contributing to the Organisation’s One Health approach. A strong believer in frontier technologies like artificial intelligence, blockchain and bio-sequencing, he loves applying these tools to disease intelligence and health system design. With a PhD in Epidemiology from the University of California, Davis, and two decades of work across four continents, he has tracked tigers and lions in India, led teams countering COVID-19 misinformation, built cross-regional global networks and co-developed guidelines and strategies adopted by governments and multilateral partners. An inducted member of the Sigma Xi scientific honor society and former Emerging Pandemic Threats researcher, he has co-authored a number of scientific and technical papers.
Abstract
Artificial Intelligence (AI) is transforming animal health across the world. These tools enable early disease detection, personalised treatment and improved animal management. By analysing complex datasets, such tools optimise diagnostics, predict outbreaks and improve veterinary care. This article discusses real-world AI trends and case studies. As these technologies improve, their impact on animal health will continue to increase.
AI is the ability of computers to imitate human cognitive intelligence, such as learning from data, recognising existing patterns and making decisions. Branches of AI include Machine Learning (ML), Natural Language Processing (NLP), Computer Vision, Robotics and Expert Systems. While ML uses data to gain statistical insights and recognise patterns that can be applied to new tasks, NLP uses algorithms to understand and mimic human languages. Meanwhile, Computer Vision uses models to interpret visual inputs, Robotics uses machines to automate tasks and Expert Systems use algorithms to mimic the reasoning and decision-making abilities of human experts.
Real-world Applications, Case Studies and the Road Ahead
Veterinarians and other animal health professionals work with AI experts who specialise in ML, NLP, Computer Vision, Robotics, Expert Systems and other novel branches to improve the health of animals around the world (Table 1). AI can be used to improve disease detection, track antimicrobial resistance, accelerate drug discovery and provide personalised treatment through precision medicine. It can also aid in the real-time monitoring of animals, including wildlife health, as well as improve pandemic surveillance with early-warning systems, amongst other applications. For example, by employing ML, Computer Vision and Expert Systems in medical images and scans (such as radiographs, ultrasounds, CT scans, MRIs and digital pathology images), AI can be used to improve diagnostic capabilities. In addition, ML algorithms and Expert Systems can be used to develop predictive disease detection algorithms using veterinary health records. Similarly, ML and NLP can be used to rapidly detect and respond to wildlife diseases by monitoring unusual wildlife cases, such as a disease seen in a new host species for the first time, across a network of wildlife rehabilitation centres [1].
Table 1. Artificial Intelligence (AI) in animal health: trends, real-world applications and case studies
Case Study 1: Using CNN to interpret dogs’ thoracic radiographs for disease detection
AI is increasingly used to help veterinarians interpret radiographs to diagnose diseases. A group of scientists from the University of Padua in Italy developed a multi-label deep convolutional neural network (CNN) model* – a layered pattern-recognition algorithm that learns to identify patterns in data through a step-by-step approach – to classify thoracic radiographs in dogs [2]. This computer-aided model was trained to classify thoracic radiographs into the various radiographic findings, including unremarkable, interstitial pattern, mass, pleural effusion, pneumothorax and megaoesophagus. Though two different CNNs were trained, the CNN based on the ResNet-50 architecture, a deep learning model with residual blocks, had an AUC (Area Under the Receive-Operator Curve), a measure of diagnostic accuracy, above 0.8 (which is considered a good result) for a majority of radiographic findings.
Case Study 2: Using ML to predict effluent quality and microbial risk in dairy farms
AI has been used to predict effluent quality and its associated risks of microbial pollution at dairy farms in California in the United States of America. In intensive dairy farms like these, significant quantities of animal waste are generated, then treated and repurposed in the farm. While this waste is recycled and repurposed, the risk of pathogens getting recycled back to the farm needs to be minimised using established treatment protocols. To study this risk, a pilot study conducted by a group of scientists, including the author, from the University of California, Davis compared two Alternate Dairy Effluent Management Strategies (ADEMS) which separate and process the solid and liquid components of waste material: (1) the mechanical solid-liquid-separator, a machine that filters solids from liquids, and (2) the gravitational Weeping Wall solid separation system, a wall that drains liquids using gravity.
The team developed a heuristic AI model called E-C-MAN, which combined the results of 26 foundational prediction models into one final output: a super-learner model that learns from its independent foundational models [3] using data from 17 chemical, physical, structural and seasonal input variables to predict E. coli levels (Figure 1). This ML model demonstrated that the separated solid stage in both the ADEMS had significantly lower E. coli levels: an indicator organism that predicts effluent quality and coliform risk. This could have important downstream implications where the separated solid stage can be processed as a manure source for the farm.
Figure 1. The E-C-MAN stacked ensemble Generalized Linear Model was built by a group of scientists from the University of California, Davis, to predict effluent quality and microbial risk in dairy farms that employ effluent processing plants
The model uses various chemical, physical, structural and seasonal inputs to estimate E. coli concentrations, which were used as an indicator organism for predictions. To make accurate predictions, the model combines simpler foundational models (called ‘base learner models’), such as decision trees and neural networks. The output from these base learner models was used to train the super-learner model, which has improved overall prediction accuracy by learning from each individual model. This figure is obtained from Shetty et al. 2023 under the CC BY 4.0 International License.
Challenges and Ethical Considerations
Employing AI in the real world presents multiple challenges and ethical considerations. These include data privacy and quality concerns, issues with unbalanced data leading to output biases, challenges with implementation and accessibility, as well as the high implementation costs and complexities required to deploy AI systems in resource-limited settings. In addition, many AI models, especially those that employ deep-learning systems, lack transparency in how the models make decisions. These challenges and ethical considerations could influence the level of trust that animal health professionals, veterinarians, regulators and farmers place in AI systems, hindering their adoption in real-life settings.
A Promising Future
The potential for using AI in the animal health world is immense, since we are only in the early stages of this technological revolution. In terms of capabilities, we are still in the Artificial Narrow Intelligence (ANI) stage, where systems like ChatGPT or Claude are designed to perform specific tasks (or a narrow range of tasks) extremely well. Once we are able to design next-level Artificial General Intelligence (AGI) systems, which can perform a wide range of tasks that match or exceed human capabilities in the same model, and Artificial Super Intelligence (ASI) systems, which can surpass human intelligence in all respects, including thinking, reasoning, learning, making judgements and possessing superior cognitive abilities, the possibilities would be virtually endless. There would be no limit, in human terms, to the system’s capabilities. In terms of functionality, we have already surpassed the Reactive Machine AI stage, where AI systems lack memory and cannot learn from past experiences, and we are currently at the Limited Memory AI stage, characterised by systems like ChatGPT and Claude that can learn from historical data to improve future decisions. The next step towards the future is developing a Theory of Mind AI, which would be capable of understanding emotions, beliefs and intentions, and finally, a Self-Aware AI, which would possess both consciousness and a capacity for self-perception.
To illustrate a hypothetical future trajectory for AI applications in animal health, one might imagine a time when ASI and Self-Aware AI become technically feasible. In such a foresight scenario, an AI platform built by the World Organisation for Animal Health (WOAH) could, in theory, continuously process real-time data from billions of sensors in farms, veterinary hospitals, clinics and wildlife habitats around the world, thus anticipating disease outbreaks, such as high pathogenicity avian influenza, weeks in advance, and supporting WOAH Members (countries and territories) with timely and coordinated responses. This platform could autonomously assist in coordinating vaccination campaigns and negotiating data-sharing protocols, adapting its decision-making based on ethical feedback from veterinarians and other stakeholders. This platform would also be self-aware, as it would understand its roles, limitations and the ethical consequences of its choices in the real world. While such abilities currently remain hypothetical, they demonstrate the importance of engaging with AI’s evolving capabilities to forge a responsible path forward.
*A convoluted neural network is a type of AI algorithm designed to identify patterns in data through a layered approach. Data is processed by the algorithm through a series of steps (or layers), with each layer detecting increasingly complex features, in order to understand detailed patterns.
Main image copyright: gorodenkoff
References
[1] Kelly TR, Pandit PS, Carion N, Dombrowski DF, Rogers KH, McMillin SC, et al. Early detection of wildlife morbidity and mortality through an event-based surveillance system. Proc. Biol. Sci. 2021;288(1954):20210974. https://doi.org/10.1098/rspb.2021.0974
[2] Banzato T, Wodzinski M, Burti S, Osti VL, Rossoni V, Atzori M, et al. Automatic classification of canine thoracic radiographs using deep learning. Sci. Rep. 2021;11(1):3964. https://doi.org/10.1038/s41598-021-83515-3
[3] Shetty BD, Amaly N, Weimer BC, Pandey P. Predicting Escherichia coli levels in manure using machine learning in weeping wall and mechanical liquid solid separation systems. Front. Artif. Intell. 2023;5. https://doi.org/10.3389/frai.2022.921924
[4] Xiao S, Dhand NK, Wang Z, Hu K, Thomson PC, House JK, et al. Review of applications of deep learning in veterinary diagnostics and animal health. Front. Vet. Sci. 2025;12. https://doi.org/10.3389/fvets.2025.1511522
[5] Yu H, Lee IG, Oh JY, Kim J, Jeong JH, Eom K. Deep learning-based ultrasonographic classification of canine chronic kidney disease. Front. Vet. Sci. 2024;11. https://doi.org/10.3389/fvets.2024.1443234
[6] Jeong J, Choi H, Kim M, Kim SS, Goh J, Hwang J, et al. Computed tomography radiomics models of tumor differentiation in canine small intestinal tumors. Front. Vet. Sci. 2024;11. https://doi.org/10.3389/fvets.2024.1450304
[7] Fraiwan MA, Abutarbush SM. Using Artificial Intelligence to Predict Survivability Likelihood and Need for Surgery in Horses Presented With Acute Abdomen (Colic). J. Equine Vet. Sci. 2020;90:102973. https://doi.org/10.1016/j.jevs.2020.102973
[8] Haghofer A, Fuchs-Baumgartinger A, Lipnik K, Klopfleisch R, Aubreville M, Scharinger J, et al. Histological classification of canine and feline lymphoma using a modular approach based on deep learning and advanced image processing. Sci. Rep. 2023;13:19436. https://doi.org/10.1038/s41598-023-46607-w
[9] Liu Z, Deng D, Lu H, Sun J, Lv L, Li S, et al. Evaluation of Machine Learning Models for Predicting Antimicrobial Resistance of Actinobacillus pleuropneumoniae From Whole Genome Sequences. Front. Microbiol. 2020;11. https://doi.org/10.3389/fmicb.2020.00048
[10] American Veterinary Medical Association (AVMA). Artificial intelligence poised to transform veterinary care. Schaumburg (United States of America): AVMA; 2024. Available at: https://www.avma.org/news/artificial-intelligence-poised-transform-veterinary-care (accessed on 11 April 2025).
[11] Vlasiou MC. Vet informatics and the future of drug discovery in veterinary medicine. Front. Vet. Sci. 2024;11. https://doi.org/10.3389/fvets.2024.1494242
[12] University of Cambridge. Researchers design AI system to assess pain levels in sheep. Cambridge (United Kingdom): University of Cambridge; 2017. Available at: https://www.cam.ac.uk/research/news/researchers-design-ai-system-to-assess-pain-levels-in-sheep (accessed on 11 April 2025).
[13] Halev A, Martínez-López B, Clavijo M, Gonzalez-Crespo C, Kim J, Huang C, et al. Infection prediction in swine populations with machine learning. Sci. Rep. 2023;13:17738. https://doi.org/10.1038/s41598-023-43472-5
[14] Turner J. The potential for tracking wildlife health & disease via bioacoustics is great (commentary). Menlo Park (United States of America): Mongabay; 2025. Available at: https://news.mongabay.com/2024/04/tracking-wildlife-health-disease-via-bioacoustics-is-effective-more-research-is-needed-commentary/ (accessed on 30 April 2025).
Continue reading
Receive the latest updates from our Organisation, network, and partners.
This includes Members’ self-declarations, news articles, scientific insights, publications, and more.
