高级AI助手的伦理
计算机与社会
2024-04-30 v2
摘要
本文聚焦于高级AI助手带来的机遇以及伦理和社会风险。我们将高级AI助手定义为具有自然语言接口的人工智能体，其功能是根据用户的期望，在一个或多个领域内代表用户规划和执行一系列行动。本文首先考虑技术本身，概述了AI助手、其技术基础和潜在应用范围。然后探讨了AI价值对齐、福祉、安全和恶意使用等问题。进一步扩展探究范围，我们更详细地考虑了高级AI助手与个体用户之间的关系，探讨了操纵与说服、拟人化、适当关系、信任和隐私等主题。基于这一分析，我们考虑了高级助手在社会规模上的部署，重点关注合作、公平与可及性、错误信息、经济影响、环境以及如何最好地评估高级AI助手。最后，我们为研究人员、开发者、政策制定者和公共利益相关者提供了一系列建议。
引用
@article{arxiv.2404.16244,
title = {The Ethics of Advanced AI Assistants},
author = {Iason Gabriel and Arianna Manzini and Geoff Keeling and Lisa Anne Hendricks and Verena Rieser and Hasan Iqbal and Nenad Tomašev and Ira Ktena and Zachary Kenton and Mikel Rodriguez and Seliem El-Sayed and Sasha Brown and Canfer Akbulut and Andrew Trask and Edward Hughes and A. Stevie Bergman and Renee Shelby and Nahema Marchal and Conor Griffin and Juan Mateos-Garcia and Laura Weidinger and Winnie Street and Benjamin Lange and Alex Ingerman and Alison Lentz and Reed Enger and Andrew Barakat and Victoria Krakovna and John Oliver Siy and Zeb Kurth-Nelson and Amanda McCroskery and Vijay Bolina and Harry Law and Murray Shanahan and Lize Alberts and Borja Balle and Sarah de Haas and Yetunde Ibitoye and Allan Dafoe and Beth Goldberg and Sébastien Krier and Alexander Reese and Sims Witherspoon and Will Hawkins and Maribeth Rauh and Don Wallace and Matija Franklin and Josh A. Goldstein and Joel Lehman and Michael Klenk and Shannon Vallor and Courtney Biles and Meredith Ringel Morris and Helen King and Blaise Agüera y Arcas and William Isaac and James Manyika},
journal= {arXiv preprint arXiv:2404.16244},
year = {2024}
}