Does the use of artificial intelligence in farming help or hinder animal welfare? Natasha Boyland, Kristina Kiminiute and Jonathan Birch argue that while AI brings exciting possibilities for better understanding what animals need, this will be futile without a policy framework that ensures their needs are actually met.
Walk into an egg farm for the first time, and you might be surprised by the cacophony of clucking hens and moving sea of brown feathers covering every inch of the floor. Perhaps not quite how you imagined a cage-free system to look.
But the moment that really strikes you may be when, instead of a farmer, you see a small AI-driven robot moving through the barn. A robot built to find and disrupt hens determined to lay their eggs on the floor instead of in dedicated nest boxes – a “problematic” behaviour to the egg industry.
This technology is already being tested and is just one example of AI’s increasing integration in farming. While technically impressive, it raises serious ethical issues. In this case, AI detects a hen pursuing a preference to nest on the ground and the “solution” is to physically prevent the behaviour, rather than consider why hens prefer those locations and find ways to better provide for their needs.
On 7 July 2026, the European Commission published its Livestock Strategy, which promotes the uptake of “decision-support systems, sensors, big data applications, robotics, early-warning solutions and precision livestock farming” as a way of building a resilient livestock sector.
This is one of its five priorities and includes “encouraging the adoption of digital technologies to improve animal health and welfare”. But is the optimism in the Commission’s Livestock Strategy about AI capabilities improving animal welfare justified?
AI for welfare vs AI for production goals
The laying hen robot example comes from one of over 80 presentations given at the recent AI4Animal Science (AI4AS) conference, one of the largest international gatherings of researchers using AI applications in animal science.
The conference focuses on addressing animal husbandry challenges using advanced technologies, which have potential to “profoundly transform the sector, improving efficiency, sustainability, and animal welfare”. While the word “welfare” featured in many talks, detecting negative health issues in service of improving production goals appeared the dominant theme.
Consider another example from AI4AS: a study investigating causes of piglet mortality in farrowing crates – narrow cages designed to confine a sow (mother pig) before and after giving birth. They are used primarily to reduce the risk of sows accidentally crushing their piglets, but it’s not always effective and comes at great cost to sows’ welfare.
A computer-vision system was used to analyse the number of times and duration that sows stood, sat and lay down. Restricting these intelligent, social animals to these three most basic behaviours is an upsetting thought to many. So much so that over 1.4 million citizens and various animal welfare NGOs have called for an end to the use of these cages in the EU.
In this study, the finding that individual sows differed in the number of times they stood and lay down was presented as a breeding opportunity. Farmers could breed more of the so-called “calm” sows in order to lower the risk of piglets being crushed. But what would the welfare impacts be for the sows? And what degree of welfare improvement can really be achieved while they remain in such severe confinement?
The goal here is to change animals to fit restrictive systems, rather than to improve the environment to better meet the animals’ natural behavioural needs. In industrial farming, production efficiency takes precedence over the needs of sentient beings.
Centring positive welfare
But it doesn’t have to be this way. We can wield the powers of AI for more positive outcomes. The animal welfare field itself has evolved from purely focusing on reducing negative experiences, to emphasising the importance of positive ones – captive animals should have opportunities for joy, pleasure and contentment in their lives.
A third example from AI4AS shows the potential of this research approach. Jorge Vázquez-Diosdado and colleagues at the University of Nottingham used AI to measure play behaviour in dairy calves, harnessing technology to automatically measure a positive welfare indicator.
Their model was able to detect the frolicking of calves, which happened more often when they were in good health and when given access to cow brushes. This demonstrates how AI can be used to look beyond minimising suffering and help research new ways to make animals’ lives better.
Therefore, the optimism in the EU Livestock Strategy is not, in principle, misplaced. AI has unprecedented potential to accelerate our understanding of farmed animals’ behaviour, overhaul welfare assessment and provide novel solutions. But the risks are equally significant. The question is how to steer the sector towards uses that genuinely centre animals’ needs and away from those that help intensive farming practices to exploit them to an even greater degree.
Remembering animals in the EU AI Act
There is a clear and urgent need for animals’ interests to be represented in ethical discourse and governance of AI. Technologies that could harm individuals – humans and other animals – must be monitored and used responsibly.
In the EU, regulation of AI falls under the EU AI Act (2024), which introduces requirements based on the level of risk posed. But while the Act is built on the principle that AI should be used safely and in accordance with Union values, it fails to address animal welfare.
This is a major oversight, considering the EU’s recognition that animals are sentient beings, meaning institutions must “pay full regard to the[ir] welfare requirements” when shaping policy. It is also misaligned with the Livestock Strategy’s focus on improving animal welfare, which it admits is “at the heart of citizens’ concerns when it comes to livestock farming”.
The Commission has set the direction for EU farming, including the role of future technologies to improve welfare. Now is the moment to close the gap between the EU’s commitments to animals and its AI governance. Updating the Act and associated guidance would be a major step forward in ensuring AI works to realise those commitments and does not merely support further intensification of low welfare systems.
AI brings exciting possibilities for better understanding what animals need, but this will be futile without a policy framework that ensures their needs are actually met.
Note: This article gives the views of the authors, not the position of LSE European Politics or the London School of Economics.
Image credit: Labellepatine provided by Shutterstock.
Facts Only
* Natasha Boyland, Kristina Kiminiute, and Jonathan Birch argue that AI requires a policy framework to ensure animal needs are met.
* An AI-driven robot was tested to find and disrupt hens attempting to lay eggs on the floor instead of nest boxes.
* A study investigated piglet mortality in farrowing crates using computer vision to analyze sow behavior (standing, sitting, lying down).
* Restricting sows' behaviors is presented as a breeding opportunity for more 'calm' sows to reduce crushing risk.
* AI was used by Jorge Vázquez-Diosdado and colleagues at the University of Nottingham to measure play behavior in dairy calves.
* The European Commission's Livestock Strategy promotes decision-support systems, sensors, big data applications, robotics, and precision livestock farming to improve animal health and welfare.
* A computer-vision system detected frolicking in dairy calves, correlating with good health and access to cow brushes.
* Regulation of AI in the EU falls under the EU AI Act (2024).
Executive Summary
Full Take
The narrative structure presents a critical tension between instrumentalist goals—efficiency and production maximization—and intrinsic ethical considerations regarding animal sentience. The core implication is that technological advancement, when unguided by ethical foresight, risks accelerating systems that exploit animals for greater intensification, regardless of stated welfare objectives. A significant pattern involves the deflection of focus: the shift from addressing the *causes* of negative experiences (like confinement) to merely mitigating their acute consequences using technology. This suggests a systemic tendency to treat animal welfare as an outcome to be managed within production parameters, rather than a foundational input for system design. The contrast between AI used for "problematic" behavior correction versus AI used for measuring positive states highlights a crucial divergence: the potential for technologies to enforce negative constraints versus the potential for them to reveal positive states. This points toward a systemic gap where policy frameworks fail to embed genuine welfare as an independent goal rather than a byproduct of production metrics. The necessary step is not just regulating the technology itself, but fundamentally redefining what constitutes value within livestock systems.
Bridge Questions: What specific metrics could effectively quantify and prioritize sentient interests over production targets in AI deployment? How can governance structures be established that mandate the exploration of positive welfare indicators alongside risk mitigation when developing agricultural AI? What are the long-term social consequences if AI implementation prioritizes efficiency over inherent animal experience?
Sentinel — Human
The text presents a nuanced argument about the ethical and regulatory implications of using AI in animal farming, effectively balancing technological optimism with necessary policy intervention.