Will artificial intelligence transform the way we assess animal welfare?

Current methods of assessing animal welfare 

Author: Sarah Babington

In recent decades there has been a growing focus on ethical and sustainable farming practices of which animal welfare is a critical component. Although there is no formally agreed upon definition of animal welfare, there is some consensus that the welfare of an animal is influenced by their experience of the world which is influenced by both internal and external factors (Fernandes et al., 2021). Frameworks such as the Five Freedoms have been extensively used to help define and assess animal welfare in different contexts, with a focus on minimising and preventing negative experiences during an animal’s lifetime (Duncan, 2019). More recent frameworks, like the Five Domains, have expanded upon the concept of the Five Freedoms to recognise the importance of mental state on an animal’s overall welfare and the need to provide opportunities for positive experiences in addition to minimising negative experiences over an animal’s lifetime (Mellor, 2016). 

The assessment of animal welfare typically relies upon a collection of several measures that may be animal-based, resource-based, and management-based. Animal-based measures mainly include behavioural, physiological, and neurobiological indicators (for examples see Table 1) which are used to assess how an animal is responding to a situation, which can infer their experience and welfare state. While these current animal-based measures provide important information on the welfare of an animal, they have limitations as they can be non-specific, subjective, and mostly reflect negative rather than positive welfare states (Babington et al., 2024). Resource-based and management-based measures (for examples see Table 1) can also be used in addition to animal-based measures to understand the influence of external factors on animal welfare. 

 

 

Table 1. Examples of measures used to assess animal welfare (adapted from (Babington et al., 2024). 

Type of measure  Examples 
Animal-based measures  Behavioural  Behavioural observations (e.g., stereotypies, aggression, behavioural diversity)Fear testsCognitive bias tests Preference and motivation testsQualitative Behavioural Assessment Facial expressions (e.g., grimace scale) Vocalizations
Physiological  Catecholamines (e.g., epinephrine and norepinephrine) Glucocorticoids (e.g., cortisol and corticosterone)DehydroepiandrosteroneReproductive hormones (e.g., luteinizing hormone, gonadotrophin-releasing hormone)Fatty acid intermediates (e.g., prostaglandins)Metabolic enzymes (e.g., glucose, lactate dehydrogenase, creatine kinase)Inflammatory markers (e.g., serum amyloid A, C-reactive protein)Immune system markers (e.g., total white blood cells, IL-6, IgA)Cardiovascular function (e.g., heart rate, heart rate variability, blood pressure)Respiratory function (e.g., respiratory rate)Core body temperature 
Neurobiological  Neurochemicals (e.g., dopamine, serotonin, endogenous opioids)Brain activity via changes in blood flow with functional magnetic resonance imagingBrain activity via changes in blood oxygenation with functional near infrared spectroscopyElectrical brain activity with electroencephalography
Resource-based measures   Housing system Feed and water quality and quantity Space allowance Indoor housing environment (e.g., air quality, temperature and humidity, lighting programme) Outdoor or pasture environment Transport vehicle 
Management-based measures  Personnel training and competency Handling practices Routine husbandry procedures (e.g., vaccinations, dehorning, castration) Biosecurity management Business Standard Operating Procedures 

 

The need for objectivity when assessing animal welfare

Current animal welfare measures can be non-specific and subjective resulting in a need for new measures or ways of collecting data that are more objective and specific. Having objective and standardised measures when assessing animal welfare is critical to ensure consistent, reliable, and quantifiable data to inform decisions around management practices and regulatory changes. This is where technological advances can provide some advantages. Artificial intelligence systems have shown some promise in addressing some of the current limitations associated with how animal welfare is assessed. The various visual and sensor technologies available can reduce the reliance on human observation when assessing welfare, which not only addresses the risk of subjectivity but also provides personnel more time to focus on tasks they are uniquely qualified to perform. These technologies are also able to collect much larger amounts of data than ever before which artificial intelligence systems can analyse in real-time allowing for improved efficiency and ability to establish a welfare benchmark that progress can be tracked against or corrective actions can be implemented. 

Artificial intelligence systems 

Artificial intelligence encompasses various types of models, of which some of the ones commonly used in the context of animal welfare include machine learning, deep learning, and neural networks. Briefly, machine learning uses algorithms which are trained against large datasets to identify patterns and create predications and decision models that can then be applied to analyse new unseen datasets (Choi et al., 2020). Deep learning is a type of machine learning that can automatically recognise patterns in datasets from images, video, or audio, by using multilayered neural networks to mimic the decision-making process made by humans (Choi et al., 2020). Deep learning algorithms are often used for tasks too complicated for standard machine learning. Neural Networks are a type of machine learning that are used in deep learning algorithms, which are designed to process data like the neuronal system in the brain to mimic how decisions are made by humans (Choi et al., 2020). These different types of artificial intelligence may be used by themselves or in hybrid models where the different algorithms are integrated to minimise the limitations of the models or improve performance (Bao & Xie, 2022). 

Benefits and challenges of using artificial intelligence to assess animal welfare

Artificial intelligence systems applied to assessing animal welfare are able to collect large amounts of data via images, video, audio, or sensor data which can then be analysed in real-time to identify patterns, specific factors, form predictions, and create decision-making models. Some of the possible benefits of applying artificial intelligence in farming systems to assess animal welfare include (Zhang et al., 2024): 

 

    • Increased objectivity and standardisation of data collection and analysis. 

    • Non-invasive collection of data at both an individual animal and group level. 

    • Data collection and analysis can be automated and scaled as not limited by personnel resourcing.  

    • Data can be analysed in real-time to allow for early detection of issues and improved efficiency in decision making. 

    • Increased database of information for analysis to make more accurate predictions and inform decisions. 

An example of how artificial intelligence is already being used in farming systems and research is through using video, microphone, and accelerometers to monitor the behaviour of animals on farm. The collected data can be processed in real-time by artificial intelligence to categorise the different types of behaviours (e.g., feeding, general activity, aggression) performed by an animal and identify positive or negative trends (e.g., animals not eating or displays of aggression). This information is then available for earlier detection than is likely from human observation of potential animal welfare and productivity issues that can be further investigated and addressed as required.  

While the use of artificial intelligence shows great potential to improve both the welfare and efficiency of farming systems through its evidence-based decision making, there remains a number of challenges that should be considered. Current challenges and possible risks associated with the widespread use of artificial intelligence in farming systems include (Al-Lataifeh et al., 2024): 

 

    • Requirement for very large datasets to accurately train the artificial intelligence systems. Limited data available or where data is difficult to access will limit the effectiveness of these systems.

    • Ongoing training for personnel to ensure up to date knowledge and competency to manage and effectively use the artificial intelligence systems.  

    • Significant upfront cost which could prevent implementation of artificial intelligence systems in smaller businesses and lead to increasing economic disparities between businesses that are unable to afford new technologies and those that can. 

  • Data security and privacy concerns. 

 

Real-life examples of artificial intelligence currently being used to assess animal welfare 

 

    • Argus is a video and sensor monitoring system with artificial intelligence that is used to monitor and identify animal welfare risk in real-time at slaughtering establishments. 

    • ClearFarm Project uses precision livestock farming for pigs and dairy cattle on farms. Project partners include Universitat Autonòma de Barcelona, Wageningen University & Research, Aarhus University, Università Degli Studi di Milano, Universidad de Murcia, Luke, COVAP, Elpozo, Eshuis, DOI Sensors, Herd-itt, and Hämeenlinnan Osuusmeijeri. 

    • WUR Wolf is a real-time facial recognition platform created by Wageningen University & Research that has been validated to detect 13 facial actions and an inferred 9 mental states in pigs and cattle. expression detection for pigs and cows. 

  • AI4Animals is a video monitoring system with artificial intelligence that identifies animal welfare risk in real-time at slaughtering establishments. The system is owned by Deloitt and was created in partnership with Dutch Society for the Protection of Animals and Vion Food Group. 
 

Impetus Animal Welfare

At Impetus Animal Welfare we are deploying the Argus technology into abattoirs in Australia. This technology is already commercially used in Europe and so there are datasets already being built under commercial conditions.

We believe that better objective measures in abattoirs will have a significant impact on the ability of businesses to make faster and more specific decisions to reduce risk, such as training for particular issues, or changing stock handling to avoid bottlenecks and optimise flow.

It is an exciting future, contact us to find out more or how you can be involved. admin@impetusanimalwelfare.org 

References

Al-Lataifeh, F., Tarawneh, R., Al-Taha’at, E., Dayoub, M., Sutinen, E., & Al-Najjar, K. (2024). Smart technologies for livestock sustainability and overcoming challenges: A review. Hunan Daxue Xuebao/Journal of Hunan University Natural Sciences, 67, 196-204. https://doi.org/10.5281/zenodo.13744058 

Babington, S., Tilbrook, A. J., Maloney, S. K., Fernandes, J. N., Crowley, T. M., Ding, L., Fox, A. H., Zhang, S., Kho, E. A., & Cozzolino, D. (2024). Finding biomarkers of experience in animals. Journal of animal science and biotechnology, 15(1), 28. https://doi.org/10.1186/s40104-023-00989-z. 

Bao, J., & Xie, Q. (2022). Artificial intelligence in animal farming: A systematic literature review. Journal of Cleaner Production, 331, 129956. https://doi.org/10.1016/j.jclepro.2021.129956 

Choi, R. Y., Coyner, A. S., Kalpathy-Cramer, J., Chiang, M. F., & Campbell, J. P. (2020). Introduction to machine learning, neural networks, and deep learning. Translational vision science & technology, 9(2), 14-14. https://doi.org/10.1167/tvst.9.2.14 

Duncan, I. J. H. (2019). Animal welfare: A brief history. In S. Hild & L. Schweitzer (Eds.), Animal Welfare: From Science to Law (pp. 13-19). La Fondation Droit Animal, Éthique et Sciences. 

Fernandes, J. N., Hemsworth, P., Coleman, G. J., & Tilbrook, A. J. (2021). Costs and benefits of improving farm animal welfare [Review]. Agriculture, 11(2), 104. https://doi.org/10.3390/agriculture11020104 

Mellor, D. J. (2016). Updating animal welfare thinking: Moving beyond the “Five Freedoms” towards “A Life Worth Living”. Animals, 6(3). https://doi.org/10.3390/ani6030021 

Zhang, L., Guo, W., Lv, C., Guo, M., Yang, M., Fu, Q., & Liu, X. (2024). Advancements in artificial intelligence technology for improving animal welfare: Current applications and research progress. Animal Research and One Health, 2(1), 93-109. https://doi.org/10.1002/aro2.44