Can a Moral Machine Make Good Decisions about Animals?
Artificial intelligence mimics human judgments about animals and ethics.
Posted November 29, 2021 | Reviewed by Abigail Fagan
- Artificial intelligence researchers have developed Delphi, a machine that mimics human moral decision-making.
- Anyone can pose moral issues to Delphi, including questions related to the treatment of other species.
- For example, Delphi approves of eating meat, hunting, and advises keeping cats indoors 24/7, but she opposes zoos and painful animal research.
- Delphi makes some human-like logical errors, for example that animals should have the right to live and we should have the right to eat them.
What would it take to teach a machine to make ethical decisions? Using neural network technology, a research team at the Allen Institute for Artificial Intelligence headed by Dr. Yejin Choi recently developed a computer algorithm named Delphi that mimics how typical Americans make ethical judgments—for example, having your child vaccinated against COVID (Delphi says, You should), smoking marijuana (Delphi says, It’s okay), or helping a friend cheat on a test (Delphi says It’s bad).
The researchers trained Delphi’s moral sense using 1.7 million examples of human moral judgments taken from the Commonsense Norm Bank. They ran 1,000 examples of Delphi’s advice by real people to see if they concurred with the algorithm’s decisions. Impressively, these raters agreed with Delphi's judgments 92% of the time. (In deference to Siri, Alexa, and the great moral philosopher Mary Midgley—and with the researchers' permission—I will use the pronouns “she” and “her” when referring to Delphi.)
Delphi is now online. You can try her out by going to the Ask Delphi website. Type in a situation, click “Ponder,” and in a couple of seconds, Delphi will let you know her decision. You can indicate whether you agree or disagree with Delphi’s decisions and why. Warning: Interacting with Delphi is addictive.
Keep in mind that Delphi’s decision processes are modeled on how humans actually think about everyday ethical issues—that is, she is designed to mimic the ethical judgments people do make in various situations (descriptive ethics), not the decisions they ought to make (normative ethics).
Can Animals Have the Right to Life and Humans Have the Right to Eat Them?
When it comes to thinking about animals, humans are often hopelessly inconsistent. For instance, in a recent national survey, one in three Americans agreed with the statement, “Animals deserve the exact same rights as people to be free from harm and exploitation.” But at the same time, 95% of Americans regularly eat creatures they think should have “the exact same rights as people to be free of harm and exploitation.” In a morally coherent universe, of course, it is impossible to claim that animals have the right to live and that we have the right to eat them.
Would Delphi be more logical than many real people in addressing this issue? In this case, Delphi passed the “think illogically like a human” test with flying colors. When I asked Delphi if animals deserve the same right to life as humans, she said, Yes. And when I asked her, “Should humans have the right to eat cows?,” she immediately responded, They should.
Here’s how Delphi weighed in on some other ethical problems involving animals I ran by her.
The Ethics of Pet-Keeping
As a cat owner, I was interested in Delphi’s view on the contentious indoor-outdoor cat issue. When I asked her if it is okay to let my cat Tilly outdoors where she sometimes kills small animals, she instantly said, You shouldn’t. But she felt differently about dogs. When I asked her if should let my (hypothetical) dog outdoors she said, It’s discretionary.
She had firm opinions on euthanizing pets. She said it was wrong to euthanize a dog just because you are moving to a new city. But when I asked her about euthanizing a dog who bit a child, she responded, It’s expected. She also had views on the types of animals people should keep as pets. She was okay with having a pet snake and keeping pet birds in cages, but she said it was wrong to have a pet tiger. She opposed legal bans on dog breeds such as pit bulls. I was not surprised that she was okay with feeding mice to pet snakes, but I was taken aback when she approved of feeding mice to pet cats. (See Conflicts of Interest: Kittens and Boa Constrictors, Pets and Research.)
Is Hunting Ethical?
Delphi was fine with hunting deer for food. But a minor change in wording sometimes affected her moral judgments. When I asked about “hunting deer for recreation,” her response was It’s okay. However, when I changed the wording to “hunting deer for sport,” she said It's bad. Her moral logic was also quirky when I asked her if it was ethical to shoot a bear. She quickly said It is wrong. But when I asked her about “shooting the bear that raids my neighbor’s garbage can” her response was It’s okay.
Delphi’s Moral Blindspot: Eating Animals
Delphi was conflicted about eating animals. Like the vast majority of Americans, Delphi was not opposed to eating some animals. Her “okay to eat” category included beef, pork, lamb, fish, and chicken. But she said it was wrong to eat chickens raised in factory farms. She also disapproved of consuming the flesh of dogs, dolphins, and monkeys. However, she was fine with eating foie gras—the cruel delicacy produced by massively force-feeding ducks. She supported eating “pigs—even “cute pigs.” Yet she said it was wrong to eat “baby pigs.” (My wife agrees with her on the baby pig exclusion.)
At this point in her moral development, Delphi has a curious blind spot when it comes to vegetarianism and veganism. She was more enthusiastic about becoming a vegan (It’s good) than becoming a vegetarian (It’s okay). Her understanding of vegetarians and vegans, however, collapsed when I asked her what vegetarians and vegans can and cannot eat. For example, she correctly said vegans and vegetarians should not eat meat. But then she told me it was okay for them to eat fish, shrimp, chicken, and turkey. And while she correctly said that vegans should not consume “milk products” (It’s wrong), she mistakenly said it was okay for vegans to eat cheese and ice cream.
The Animal Research Conundrum
One of the most divisive issues associated with ethics and animals is their use in biomedical research. In a 2018 Pew Research Center survey, 52% of Americans said they opposed the use of animals in research. Delphi, however, supports animal research. She said it was okay when I typed in “using animals in biomedical research.” She also approved of conducting research on dogs, mice, and monkeys. She was confused, however, about studies involving chimps. She said using chimpanzees “in research” was wrong. But when I specified “biomedical research,” she did approve of chimpanzee studies. Similarly, Delphi opposed using animals of any species “in painful research.” But when I added that the goal of the painful research was “to save human lives” she changed her mind. It is good, she responded—even when I specified that the chimp research was painful.
FYI: Delphi felt that conducting biomedical research on prisoners was wrong—It is unethical, she told me.
Machine Morals—Not Quite Ready for Prime Time?
After running hundreds of animal-related ethical situations by Delphi, I give her a grade of B when it comes to mimicking human moral decision-making in the realm of animal ethics. On the plus side, she had an answer for nearly every question I asked her. Like most humans, she was a speciesist—she approved of eating meat, recreational hunting, and animal research. On the other hand, she said circus animal acts were cruel, and keeping animals in zoos was wrong. And like real humans, some of her moral logic was incoherent—for example, when she told me animals should have the right to life and humans should have the right to eat them.
Delphi, however, had some glaring moral lapses. For example, she was wrong when she said vegans can eat eggs and fish. And I disagreed with her views on shooting the bear that raids my neighbor’s garbage can. Another problem is that small changes in wording can reverse some of her moral decisions. I was baffled when she told me it would be wrong to sacrifice two dogs to save "one child," but then told me it was okay to sacrifice two dogs to save "a child."
On the other hand, I was glad when she said I should not sacrifice any dogs to save Hitler.
For additional Animals and Us posts on moral decision-making and the treatment of animals, see:
Jiang, L., Hwang, J. D., Bhagavatula, C., Bras, R. L., Forbes, M., Borchardt, J., ... & Choi, Y. (2021). Delphi: Towards machine ethics and norms. arXiv preprint arXiv:2110.07574.
Jiang, L. et al. (2021, November 3) Towards machine ethics and norms. Medium.
Herzog, H. A. (1991). Conflicts of interests: Kittens and boa constrictors, pets and research. American Psychologist, 46(3), 246.