穆斯克为何要捐数百万美元来提高人工智能安全性？

    史蒂芬·霍金、比尔·盖茨、以及特斯拉汽车公司创始人埃隆•穆斯克有什么共同点？

    他们都担心人工智能技术的发展可能给人类带来难以预料的灾难。作为计算机科学的一个分支领域，人工智能可以让机器模仿人类行为，并做出决策。

    这就是为什么穆斯克、非盈利机构“开放慈善项目”，以及旨在减轻新技术灾祸的研究机构“生命未来研究所”决定携手合作，资助那些防止人工智能带来灾难的研究人员。

    到目前为止，该项目共给37个研究团队提供了总计700万美元的资金。

    要记住，穆斯克和其他人设想的灾难并不是好莱坞大片的典型场景：终结者或机器人获得了一定的意识，开始攻击它们的人类主人。实际上，这一情景就本质而言并不难实现。

    这些资助旨在探索因机器和机器人开始在社会上独立运转（比如帮你挑拣商品的机械购物助手和自动驾驶汽车）而出现的法律问题。一个已经获得拨款的研究团队正在研究的课题是，机器人对人身安全和财产造成危害后，需要负哪些法律责任？比如说，研究人员想要搞清楚，如果一个机器人闯了红灯，警察应该给谁开罚单。

    另一个研究团队则在尝试制定一套指南，帮助拥有人工智能的电脑合理化自己的行为，并向人类作出解释。其理念在于让人类询问机器为何要做出某个决定，以解决任何潜在的问题。

    比方说，假设有一台拥有人工智能的电脑，它能在股市中进行交易，为公司实现最大经济收益。如果这台电脑表现出色的部分原因在于它进行了某些非法交易，那么人类可以询问电脑为何选择进行该交易，从而制止这种行为。

    “生命未来”项目的拨款官员丹尼尔·杜威表示，所有这些研究项目都是为了做好基础工作，帮助科学家和工程师创建一个“能以前所未有的方式与人类合作”的智能系统， 该资助行动是电动汽车制造商特斯拉公司首席执行官穆斯克开展的又一个超前项目。除了特斯拉之外，穆斯克还是民用航空航天公司SpaceX的首席执行官。他还大胆构想出了Hyperloop计划，这一运输系统让乘客坐在类似火车的客舱里，在隧道中高速往返于旧金山和洛杉矶之间。

    穆斯克于今年1月首先提出了投资人工智能研究的计划。上周三宣布的700万美元赞助源自穆斯克向“生命未来研究所”捐赠的1000万美元。

    尽管在人工智能领域已有大量研究和资金投入，尤其是大型科技公司，如谷歌和中国搜索公司百度，但杜威表示，这类研究大部分都是以提高人工智能技术的效果为目标的。比如Facebook研发的技术和算法，可以让这家社交网络能够识别照片中的人，即便他们的脸被挡住了。

    杜威把不同类型的人工智能研究比作核电站的员工，每个人都要扮演不同的角色。某个核电站的工程师可能会负责研究如何改善反应堆的效率，而安全工程师则要保证反应堆不会爆炸。（财富中文网）

    译者：严匡正

    审校：任文科

    What do Stephen Hawking, Bill Gates, and Tesla Motors founder Elon Musk have in common?

    They all fear that advances in artificial intelligence, an area of computer science in which machines can mimic human behavior and make decisions, could lead to potential unforeseen disasters down the road for humanity.

    This is why Musk, the non-profit Open Philanthropy Project, and The Future of Life Institute, a research organization that aims to mitigate possible catastrophes from emerging technology, are teaming up to reward researchers working to prevent calamities caused by artificial intelligence.

    So far, 37 research groups have received a share of the $7 million in grants made by the initiative.

    Keep in mind, the types of disasters imagined by Musk and others aren’t the typical Hollywood fare involving terminators or robots gaining some sort of consciousness and turning on their human overlords. Rather, they are more practical in nature.

    The grants are aimed to explore issues like the legal ramifications that could arise from machines and robots operating independently in society. Automated personal shopping assistants that pick up your groceries and self-driving cars are just a few possible examples. One of the research groups receiving grant is looking into ways to “manage the liability for the harms they might cause to individuals and property.” Essentially, if a robot runs a red light, the researchers want to know who should get the ticket.

    Another research group is looking to develop guidelines on how a computer embedded with artificial intelligence could rationalize and explain its actions to humans. The idea is that people would be able to ask a machine why it is making a specific decision to troubleshoot any potential problems.

    For example, consider a computer using artificial intelligence that is programmed to make trades on the stock market to achieve the best possible financial returns for a company. If part of the reason the machine is doing so well involves some form of illegal trading, a human could potentially stop the activity by asking the computer why is it making the trades it chooses to do.

    The goal for all of this research is to lay the groundwork for scientists and engineers to create intelligent systems that can “work with humans in ways that wouldn’t have been possible before,” said Daniel Dewey, a Future of Life program officer that oversees the grants.

    The grants are just another example of the type of far-out projects that Musk, the CEO of the electric-car maker Tesla, likes to be involved with. Besides Tesla, Musk is also the CEO of commercial aerospace maker SpaceX and he dreamed up the idea of Hyperloop, a transportation systemthat involves shooting train-like capsules through tunnels at high speed between San Francisco and Los Angeles.

    Musk first outlined his plans to invest in artificial intelligence research in January. The $7 million in grants announced Wednesday are part of the $10 million Musk donated to the The Future of Life Institute at that time.

    While there’s a lot of research and funding taking place in artificial intelligence, especially from big tech companieslike Googleand the Chinese search company Baidu, Dewey explained that a lot of that research is geared toward projects that are boosting the performance of artificial intelligence technology. Facebook , for example, has developed technology and algorithms that have led to the social network being able to identify people in photos even if their faces were covered up.

    Dewey likens the different kinds of artificial intelligence research to staff at a nuclear power plant, each of whom have different roles. While one power plant engineer might be working on ways to improve the efficiency of the reactor, a safety engineer would be responsible to make sure the reactor doesn’t blow up.