Acknowledging AI's dark side.

摘要: The 17 July special section on Artificial Intelligence (AI) (p. [248][1]), although replete with solid information and ethical concern, was biased toward optimism about the technology. The articles concentrated roles that military government play in “advancing” AI, but did not include opinions of any political scientists or technology policy scholars trained to think unintended (and negative) consequences governmental steering technology. interview Stuart Russell touches these concerns (“Fears an AI pioneer,” J. Bohannon, News, p. [252][2]), as a computer scientist, his solutions focus improved training. Yet even best training will protect against market incentives stay ahead competitors. Likewise double-edged M. I. Jordan T. Mitchell's desire “that society begin now consider how maximize” benefits transformative (“Machine learning: Trends, perspectives, prospects,” Reviews, [255][3]). Given grievous shortcomings national governance weaker capacities international system, it is dangerous invest heavily without processes place allow those who support oppose engage fair debate. The implied we are all engaged common endeavor, when fact dominated by relative handful mostly male, white east Asian, young, affluent, highly educated technoscientists entrepreneurs their affluent customers. A majority humanity outside looking in, past time for working be frank it. The rhetoric also loaded positive terms. presents risk real harm, serious analysis its potential future would do well unflinchingly acknowledge fact. The question posed collection's introduction—“How ensure rise machines entirely under human control?” (“Rise machines,” Stajic et al. , [248][1])—is wrong ask. There no institutions adequate “ensure” it. procedures which humans can take part decision process. more important this: Should slow pace research applications until people, representing world's diversity, meaningful role deliberations? Until debate, there debate worth having. [1]: /lookup/doi/10.1126/science.349.6245.248 [2]: /lookup/doi/10.1126/science.349.6245.252 [3]: /lookup/doi/10.1126/science.aaa8415