How Might AI Harm Animals?
Throughout history, human technologies have tended to both help and harm animals. For example, while advanced farming machinery eliminated the need for animals to pull heavy loads, the development of modern factory farms has led to the mass confinement, exploitation, and slaughter of billions of animals annually.
As humans explore the use of artificial intelligence (AI), the authors of this paper express concern that such technologies could disregard the interests of animals. In this paper, they develop a “harm framework” that outlines the intentional, unintentional, direct, and indirect ways that AI may negatively impact animals. With this framework, they intend to provide a clear basis to design and regulate AI technologies, considering animal welfare.
They first outline three main reasons to investigate the risks that AI technologies pose to animals:
- Ethically, they point out that animals are sentient and possess moral value.
- Humans and non-humans are interdependent, so AI technologies that harm non-humans may harm humans (and vice-versa).
- Thus far, AI ethics has largely overlooked animals or only considers their well-being as far as it’s instrumental to humans’ well-being.
Their framework is based on the work of the animal welfare scientist David Fraser and includes five types of possible harms that AI can inflict on animals:
- Intentional Illegal Harms: Under this category, illegal AI technologies may be designed to harm animals or legal AI technologies may be misused in an illegal or socially condemned way. For example, drones may trace animals for the illegal wild animal trade, or drones used to track animals for conservation purposes could be hacked by poachers.
- Intentional Legal Harms: AI technologies may be designed to harm animals in a way that society generally accepts. For example, the agriculture industry might use AI that supports or reinforces factory farming methods. Similarly, researchers may create technologies to enable animal testing.
- Unintentional Direct Harms: AI technologies that are designed to benefit society may unintentionally harm animals because they disregard or overlook them. For example, AI devices that collect underwater garbage may accidentally harm certain aquatic species. Another type of unintentional direct harm occurs when AI makes a mistake (e.g., an automated feeding device may fail to provide enough food for the animals who rely on it).
- Unintentional Indirect Harms: Generally beneficial AI can also indirectly harm animals. This can occur materially (e.g., AI may emit carbon and contribute to climate change) or epistemically (e.g., algorithmic systems may reinforce the anthropocentric bias of meat-eating or the use of animals for entertainment). Furthermore, if AI technologies distance animals from their caretakers, humans may struggle to read and interpret the animals’ needs.
According to the authors, choosing not to invest in and develop AI to benefit animals results in another type of harm. For example, AI has the potential to eliminate the use of animals in research and to find suitable plant-based meat alternatives, but such technologies can only happen with investment and support.
The authors argue that whether or not AI harms animals depends on who has a seat at the table as these technologies are being developed and regulated. For example, farming technologies created by big agriculture will likely have a much different purpose than technologies created by farmed animal advocates. As such, it’s necessary for advocates to play an active role in the AI sector to speak on behalf of animals and their needs.
As we continue to make society aware of the full breadth of harms that AI can pose to humans and animals, the authors also recommend that experts design specific AI principles addressing animal welfare; include animals in AI guidelines; and address animal welfare in laws and regulations targeted at AI systems.