GoKawiilThe Bay Area animal welfare movement is closely linked to effective altruism, a philanthropic movement committed to maximizing the amount of good one does in the world—indeed, many conference attendees work for organizations funded by effective altruists. That philosophy might sound great on paper, but “maximizing good” is a tricky puzzle that might not admit a clear solution. The movement has been widely criticized for some of its conclusions, such as promoting working in exploitative industries to maximize charitable donations and ignoring present-day harms in favor of issues that could cause suffering for a large number of people who haven’t been born yet. Critics also argue that effective altruists neglect the importance of systemic issues such as racism and economic exploitation and overlook the insights that marginalized communities might have into the best ways to improve their own lives.
When it comes to animal welfare, this exactingly utilitarian approach can lead to some strange conclusions. For example, some effective altruists say it makes sense to commit significant resources to improving the welfare of insects and shrimp because they exist in such staggering numbers, even though they may not have much individual capacity for suffering.
Now the movement is sorting out how AI fits in. At the summit, Jasmine Brazilek, cofounder of a nonprofit called Compassion in Machine Learning, opened her sticker-stamped laptop to pull up a benchmark she devised to measure how LLMs reason about animal welfare. A cloud security engineer turned animal advocate, she’d flown in from La Paz, Mexico, where she runs her nonprofit with a handful of volunteers and a shoestring budget.
Brazilek urged the AI researchers in the room to train their models with synthetic documents that reflect concern for animal welfare. “Hopefully, future superintelligent systems consider nonhuman interest, and there is a world where AI amplifies the best of human values and not the worst,” she said.