As AI systems proliferate in society, the AI community is increasingly
preoccupied with the concept of AI Safety, namely the prevention of failures
due to accidents that arise from an unanticipated departure of a system's
behavior from designer intent in AI deployment. We demonstrate through an
analysis of real world cases of such incidents that although current vocabulary
captures a range of the encountered issues of AI deployment, an expanded
socio-technical framing will be required for a more complete understanding of
how AI systems and implemented safety mechanisms fail and succeed in real life.

AI 安全的概念在社会中日益普及，AI 社区越来越关注 AI Safety 的概念，即防止在 AI 部署过程中系统行为意外偏离设计意图而导致的故障。通过分析真实案例，我们展示了当前的词汇虽然能够涵盖 AI 部署中遇到的一系列问题，但仍需要一个扩展的社会技术框架来更好地理解 AI 系统和实施的安全机制在现实生活中的成功与失败。

人工智能安全性的具体问题再探讨

Concrete Problems in AI Safety, Revisited

Increasing the number of cyclists, whether for general transport or
recreation, can provide health improvements and reduce the environmental impact
of vehicular transportation. However, the public's perception of cycling may be
driven by the ideologies and reporting standards of news agencies. For
instance, people may identify cyclists on the road as "dangerous" if news
agencies overly report cycling accidents, limiting the number of people that
cycle for transportation. Moreover, if fewer people cycle, there may be less
funding from the government to invest in safe infrastructure. In this paper, we
explore the perceived perception of cyclists within news headlines. To
accomplish this, we introduce a new dataset, "Bike Frames", that can help
provide insight into how headlines portray cyclists and help detect
accident-related headlines. Next, we introduce a multi-task (MT) regularization
approach that increases the detection accuracy of accident-related posts,
demonstrating improvements over traditional MT frameworks. Finally, we compare
and contrast the perceptions of cyclists with motorcyclist-related headlines to
ground the findings with another related activity for both male- and
female-related posts. Our findings show that general news websites are more
likely to report accidents about cyclists than other events. Moreover,
cyclist-specific websites are more likely to report about accidents than
motorcycling-specific websites, even though there is more potential danger for
motorcyclists. Finally, we show substantial differences in the reporting about
male vs. female-related persons, e.g., more male-related cyclists headlines are
related to accidents, but more female-related motorcycling headlines about
accidents. WARNING: This paper contains descriptions of accidents and death.

通过一个新的数据集 “自行车框架”，本研究探讨新闻标题中对骑行者的观感，引入了多任务正则化方法提高事故相关帖子的检测准确性，并通过与摩托车相关标题的比较来证实结果，发现一般新闻网站更有可能报道自行车事故，女性摩托骑士更容易有关事故的标题。