The paper “Harm to Nonhuman Animals from AI: a Systematic Account and Framework” by Simon Coghlan and Christine Parker (2023), published in this journal, is a valuable contribution to the debate on the impact of AI technologies on nonhuman animals and invites further discussion on this important topic. One aspect to be discussed is that from an animal ethics perspective, the harm framework presented by Coghlan and Parker should be extended to a harm-benefit framework, so as not to perpetuate a rather reductionist perspective on nonhuman animals. When investigating the impact of AI on other animals, it is very important to focus on how these technologies can harm them, but it is equally important to explore how they can be used to enable animals to live good lives and improve their wellbeing. In the following, I will present the rationale for this claim (Section 1), highlight applications through which AI systems are or can be used to benefit nonhuman animals, and show how these benefits can be classified in terms of the harm framework proposed by Coghlan and Parker (Section 2). I conclude by identifying open research questions that need to be addressed next (Section 3).

1 Why “Do No Harm” Is Not Enough

There is wide agreement that individuals or entities that are sentient do possess interests in the strong sense, i.e., they have interests, contrary to weak interests where something is in the interest of someone or something. Sentient beings have at the very least the interest not to feel pain or to suffer as well as the interest to feel pleasure and other positive emotions. Or with the words of Coghlan and Parker (2023, 9): “Animals, at least sentient ones, have interests in not being harmed and interest in being benefited.” While there is an emerging debate on the possibility of AI’s being potentially sentient (Andrews & Birch, 2023), there is no doubt that certain animals are sentient.Footnote 1 Their interests need to be included in ethical evaluations of actions and social practices for an ethical evaluation to be exhaustive. When evaluating our human treatment of other animals, animal ethics has for a long time focused on the manifold harms that humans inflict upon animals. This is understandable given the—morally wrong—harms we inflict on billions of animals every day (Bar-on et al., 2018 for numbers of domesticated land animals and Singer, 2023 for an ethical evaluation). Yet, animals also require more for a good life in the normatively demanding sense than being free of pain and other negative states.

A good life in a normatively sophisticated sense means considering that an individual needs more for his or her wellbeing than just the satisfaction of basic needs such as access to sufficient food and water, shelter from heat or cold, the ability to satisfy basic instincts such as the instinct to reproduce, and freedom from pain, suffering, or stress. These are all easily measured. A normatively sophisticated understanding of the good life therefore goes beyond the merely measurable aspects of animal wellbeing (cf. Nussbaum, 2006). It is important to include this in one’s approach for at least two reasons. First, it perpetuates a rather reductionist view of animals to ignore all the different kinds of positive states and emotions they can experience, as if all that matters to them is not to feel pain or not to suffer. Most of us would not accept such a simplistic view of what we need for a good life for other humans, where we agree that aspects such as political or social participation and education are important for a good life. Certain animal populations also maintain, for example, cultural practices (Sommer & Parish, 2010), and participation in these may enhance animal wellbeing, even if such increases in wellbeing are difficult to measure.

The second aspect concerns the societal relationship between humans and animals and therefore goes beyond the (perspective on) individual animal wellbeing. After decades of animal ethics arguing that the prevailing treatment of nonhuman animals is morally wrong, visions of what human-animal relations should look like instead are needed. This has been the subject of intense negotiation in recent animal ethics (for many, see Donaldson & Kymlicka, 2013). An understanding of the good lives of other animals is needed to make sense of this.

For these reasons, it is essential to consider how AI can also promote animal wellbeing and do good for animals, as well as showing how AI systems can harm them. It is important to note that Coghlan and Parker do account for benefits and indeed often mention how AI can benefit animals. Among other examples, they mention the use of AI in smart home applications for benefiting or “taking care” of companion animals and the use of AI image recognition to help detect illegal poaching of wild animals (Coghlan & Parker, 2023, 2). Nevertheless, they “only” present us a harm framework that subsumes forgone benefits as harms.

This also falls short in the context of the three “well-known theories of ultimate harm and benefit” that Coghlan and Parker (ibid., 10) briefly discuss: hedonism, desire theory, and objective list theory. In hedonism, pleasure is just as central as pain, so that it should be explicitly investigated how pleasure is to be generated and it cannot be thought of as a mere “absence of pain.” Desires also encompass more than the desire to be free from pain and suffering and often refer to activities or states that can actively be promoted in a way that benefits an individual. In an objective list theory such as Martha Nussbaum’s Capability Approach (Nussbaum, 2006), the aspects listed on the objective list describe aspects necessary for living a good life. Thereby, a normatively demanding understanding of the good life is advocated in which it is just as necessary to benefit individuals (e.g., to help them participate in society) as it is not to harm them.

Coghlan and Parker are right in saying that “on any of these three theories of wellbeing, harm can result from the absence or the deprivation of certain positive things (…), as well as from the presence of certain negative things.” (2023, 2) However, focusing on harm and subsuming the absence of positive states under harm fall short regarding animals’ wellbeing and the good life of animals for the two reasons mentioned above.

I can only speculate as to why Coghlan and Parker have chosen to use a harm framework (rather than a harm-benefit framework). One reason may be their use of David Fraser’s (2012) framework. Fraser developed his framework to provide guidance to policymakers making decisions about the treatment of animals in real-world practice. While he mentions that human actions can also bring benefits to animals, his framework focuses on harms. In terms of benefits, Fraser takes a rather “narrow” view of how humans can benefit animals, for example, by providing them with shelter, food, or a painless death (ibid, 735). Here, I do not want to address the question of whether killing an animal constitutes a harm and whether it can be considered a benefit under certain circumstances (for details on these discussions, cf. Višak & Garner, 2016). However, as mentioned above, merely addressing the basic needs of animals, such as shelter and food, is not enough to benefit them in a sophisticated understanding of the term.

2 AI Applications that Benefit Animals

Different AI applications that are used to benefit or at least can be used for benefitting animals already exist. I introduce a few of them and cluster them in the categories Coghlan and Parker use in their framework: (a) intentional, (b) unintentional—direct, and (c) unintentional—indirect.Footnote 2 I thereby focus on existing or soon-to-exist AI application, using narrow, non-sentient AI-systems.

Next to projects using AI-systems for reducing or eliminating poaching (e.g., the Elephant Listening Project), there are many more projects that aim at protecting wild animals, such as BirdVision, Wildbook, TrailGuard AI, or ChimpFace (Bossert & Hagendorff, 2023, 6). AI here is used intentionally to benefit wildlife. To be more precise, in many projects, it is intentionally used for species protection, which is a different aim than the protection of individual wild animals. Nevertheless, these projects truly benefit individual animals.Footnote 3 AI technologies can also be used for protecting non-domesticated animals living in urban areas (Fairbrass, 2023). For example, the HogWatch project in London uses urban camera traps to survey urban mammals and their use of public green spaces and private gardens (ibid.). It intentionally aims at collecting more information about the animals to allow for better protection. Intentional AI applications that benefit domesticated animals are, e.g., enabling personalized medical treatment for individual animals in veterinary medicine (Ezanno et al., 2021, 3) or profiling shelter animals to enable faster and longer-term adoptions. In addition, AI systems can be very beneficial for animals if they allow the number of animals used in experiments to be drastically reduced (Hartung, 2016). Regarding other captive animals, the introduction of AI-based animatronics, such as robotic dolphins,Footnote 4 may benefit the marine mammals if they replace biological dolphins in aquariums and zoos.

Unintentional benefits for animals may come from self-driving cars that lead to fewer roadkills (Singer & Tse, 2022). Since this consequence of saving animal lives is not the focus of the development of such cars, it is an unintended but directly beneficial outcome. Unintended indirect beneficial consequences of AI applications for animals currently seem to be rather speculative. An example might be recommendation systems that recommend plant-based dishes for climate change reasons.

3 Outlook

In this comment, I have argued why the question of how AI applications affect animals should focus not only on harms but also on benefits. To show that AI systems can benefit animals as well as harm them, I have provided some examples in Sect. 2. Admittedly, most of them focus on very basic animal needs, such as shelter, good health, or simply enabling them to stay alive. The more sophisticated understanding of the benefits to animals that I have called for is, to my knowledge, not yet found in current projects. For example, projects would need to be developed that use AI systems to enable animals to participate in society (domesticated animals in our society or all animals in the culture of their own population) or to engage in activities that give them pleasure beyond the satisfaction of basic needs. The Earth Species projectFootnote 5 may be one such AI application, provided that the potential for interspecies communication actually benefits animals other than humans. Developing (more) such AI applications is an open task for computer scientists. An open research gap for ethicists is to develop arguments on how to prioritize between different interests in cases of conflict when applying AI systems. Such conflicts of interests will arise when including animals in ethical evaluations of AI systems.