TLDR: The AIxAnimals movement should actively support the welfare-first adoption of AI in agriculture. Regulatory restrictions and enforced transparency remain essential, but should be considered in combination with enforceable welfare standards, independent assessment pipelines, and consistent financial and technical support for farmers. This could make positive welfare effects plausible while still restricting harmful applications. The approach is not risk free, and should therefore be periodically reassessed for unintended consequences.
Numerous AIxAnimals articles I came across began with (or contained) the fact that AI might affect magnitudes more non-human animals than humans; which I believe is true. The number of farmed animals alive globally at any given time has gone beyond unimaginable numbers, estimated ~112 billion farmed animals globally. This is a huge deal - more than a dozen times as many as humans!
Human decisions and institutions exert significant control over the situation of these animals. Thus, the integration of AI and its effects depend substantially on our ability to put the welfare of these animals before our monetary motivations. Of course, this has to be done with limitations, timelines, current societal structures, technological development, tendencies, and ecological considerations in mind. Thus heavily limiting our spectrum of plausible approaches, one of which I hope to cover in this piece.
While having lively discussions about AIxAnimals as a part of the Sentient Futures AIxAnimals Fellowship, I encountered numerous arguments and approaches to the problem at hand, which felt exhilarating considering the area’s neglectedness. Nonetheless, much of the discussion concerned either initiatives opposing AI adoption in animal farming, or against animal farming itself, or improving transparency. These include the following:
Generally, I do not believe the above approaches are wrong or that they should not be considered, or that this is an exhaustive list (it definitely is not); however, in this post, I will explore a complementary approach: actively supporting and aiding the adoption of AI applications that improve animal welfare in farming, under strict regulatory guidelines.
This post goes through how AIxAnimals initiatives could contribute to and shape the adoption of AI focusing on protection and support, where it might prove to be useful, and what the remaining risks are to be addressed.
There are numerous factors restricting impactful AIxAnimals initiatives regardless of their adequacy. Our current societal structures limit the courses of action we can take for relatively rapid, consistent and positive change at scale, whether through research, advocacy, or policy initiatives. While they do not pose hard boundaries for better animal welfare or broader changes, I believe the following factors contribute tremendously to these difficulties on the AIxAnimals movement.
The current food supply chains, affordability of products, the employment of millions of people, and regional economies are all organized around the current farming systems. In 2023, about 16% of the global dietary energy supply and 36% of protein supply were accounted for by meat, eggs and milk - these are incredibly substantial numbers. This system is incredibly hard to advocate against or work to change as a significant portion of our global infrastructure depends on this economy. A successful transition has to address potential gaps or distribute transition costs fairly as well as their subsequent societal, economic, and political consequences.
This also introduces substantial economic incentives and unequal distributions of power across different parties, including farmers, processors, retailers, and technology suppliers. A 2021 World Bank overview reported that livestock supported the livelihoods of approximately 1.3 billion people, while accounting for 40% of the global agricultural output value. An additional contributing factor is the largely disadvantageous financial situation of farmers: a 2026 French study showed that the generated agricultural income of 43% of farmers across 2015-2024 was below its minimum-wage-equivalent benchmark. Accordingly, the interests and bargaining power of the aforementioned parties might differ substantially and pose immense financial pressure to welfare improvements despite farmer or public support.
Publicly accessible information about livestock, animal farming, animal welfare and overall conditions, methodologies for handling different situations (e.g. diseases), lawsuits, and reported breaches are vast. The lack of public knowledge, and conclusively inadequate engagement is hardly a result of lack of data. Nonetheless, these data are fragmented, often hard to interpret or find, and not meant for public engagement. This is reflected in a 2024 survey covering eight EU countries, which found that while public concern for animal welfare and support for stronger farm-animal welfare laws are high, most respondents still possess limited knowledge of existing welfare practices.
Meat consumption is a significant part of cultural practices across the world, forming an attachment which influences the required approaches heavily. The IPCC discusses how social norms, culture and individual choices interact with infrastructure and other structural changes over time, including in relation to animal welfare and meat consumption. The main inference for the AIxAnimals movement is that access to information itself plays a relatively small role in societal aptitude towards structural change; interacting motivations, available alternatives, and economic incentives together with information have to be taken into account.
The listed approaches - a non-exhaustive, deliberately short list as there is only so much space in a small blog - that came up during our AIxAnimals Fellowship discussion are standard and promising directions for positive impact. In light of the constraints discussed above, I also find them incomplete and potentially harmful approaches.
Broad opposition to adoption has numerous limitations which might cause slow progress, relatively few positive results, and ultimately undermine other approaches in the future. There is sustained pressure for adoption due to strong commercial and economic incentives; therefore, restrictions require strong enforcement and sustained political support. Nonetheless, I believe preventive approaches should be pursued alongside welfare-positive adoption initiatives, as there need to be clear guardrails guiding adoption to avoid potential negative outcomes.
However, focusing on preventive approaches might result in radically different outcomes without parallel support and focus on welfare-positive adoption: Efforts to stop technological development often encounter strong resistance concerning economic concerns, employment, and competitiveness. The welfare aspect of these initiatives can easily get lost in translation and public as well as political interest might not reflect the desired effects.
Focusing on animal welfare-positive approaches does not eliminate this opposition, but it does create a stronger basis for welfare favored discussion around the proposed initiatives. Providing a workable route to compliance for farmers and emphasizing the welfare-positive aspect of these initiatives could serve as a counterweight against economy motivated opposition; and potentially steer public and thus political discussion and interest towards improving animal welfare through safe adoption.
As mentioned above, publicly accessible data is extensive and covers a wide range of concerns; however, it is scarce, not uniform and hard to interpret. Therefore, existing online information cannot serve as a reliable and consistent foundation for auditing.
I believe the AIxAnimals movement could have a major impact by establishing reliable data reporting pipelines, standards, and public-facing platforms. These initiatives do exist across the AIxAnimals space, and especially in an approach supportive of adoption, they could form the basis for adequate conformity assessment and third-party auditing. With strict regulatory structures and guidelines around implementation, structured data collection is an indispensable factor.
In accordance with the previously mentioned limitations, approaches and possibilities, completely ending factory farming is an insufficient operational strategy on its own and is hardly viable in the short term. Structural transition requires strong alternatives, comparable existing investments in them, public support, alternatives for livelihoods, etc. Nonetheless, ending factory farming is a viable long-term goal and should be kept as such while focusing on a welfare-positive adoption approach.
With the eventual phaseout of factory farming in mind, a movement which encourages adoption might yield positive results, which could serve as a potential counterargument for further measures in the future. I believe this is a considerable risk of this approach, and should be treated as such with periodic reassessment against welfare outcomes, production changes, and whether support for adoption establishes dependence on continued farming in its current structure.
While a broad objective might be to completely phase out animal farming (or initially factory farming), the mentioned limitations hardly pose strict boundaries for more positive welfare conditions in the current farming establishments, although they do raise immense challenges of navigating investments and their economic effects across the mentioned limitations. Other significant challenges introduced by the above are the heavy concentration of power, economic incentives, and lack of public knowledge and participation, all of which require intricate navigation and approaches.
Welfare-first adoption should emphasize a combination of enforceable welfare standards, supported by independent assessment, and practical as well as monetary support for implementation. Future international coordination could establish common principles, while - more viable as a short-term intervention - national and regional initiatives and authorities can implement and enforce requirements. Initiatives such as the proposed UN Convention on Animal Health and Protection serve as great examples of international frameworks, although more specific standards would be required for a framework specific to AI adoption.
Measurable, species-appropriate welfare standards and requirements need to be defined by respective authorities. Duties to respond to AI-detected problems also require clear communication, and usable, case-specific guidelines. Systems for exhaustive pre- and post-deployment assessment by independent evaluators using standardized reporting are essential for supporting and verifying welfare-positive outcomes. Authorities should meaningfully penalize any breaches found during these inspections (adjusted to the offender’s situation and financial means) and have powers to suspend harmful applications in case of serious or repeated breaches.
Additionally, economically beneficial applications - that do not inflict welfare-negative effects - should be identified by respective research organizations and included as part of the adoption programs to incentivize the adoption of welfare-positive technologies. The AIxAnimals movement can also prove to be highly efficient in this respect.
Following common welfare guidelines, varying assistance and administrative support can efficiently accommodate the large spectrum of parties that farming covers. Providing grants, shared infrastructure, training, and accessible technical advice forms the basis of fast adoption and consistent maintenance over time. Of course, ongoing assistance with maintenance and periodic independent monitoring with support adjusted to resources and implementation needs are essential for lasting welfare-positive effects. I believe this type of support requires major resources and substantial political lobbying is needed to achieve such results.
In-ovo sexing is a great example of supporting the adoption of a technology where AI applications might also be relevant. France provided €10.5 million for hatchery investment in in-ovo sexing resulting in a cost-sharing agreement intended to reduce transition costs on farmers. This example demonstrates how policy can guide welfare-oriented adoption with ongoing support.
I believe the AIxAnimals field has a strong future in designing future data-handling protocols and tools, advocating for safe adoption, identifying use cases that are both economically beneficial and welfare-positive, as well as ensuring adoption does not result in unintended bad outcomes. I hope this reading might serve as a motivation for further research into how AI can be adopted in animal farming.
