被贩卖的 AI 片面性,你需要认识人类自己还是认识机器?

上周,《华尔街日报》一篇关于 Gmail 邮件被第三方厂商随意查看的报道却意外揭开了科技行业另一个谜团:当下所谓各家公司所标榜的「人工智能」,其核心竞争力不是机器,而是人

当然,这句话的理解方式有很多,比如,你当然可以认为,过去几年人工智能领域所掀起的「抢博士」热潮,人工智能的竞争,也是人才竞争;而与之相对的则是另一个事实,在这个看似高大上以及被媒体热议要时刻替代人类的行业,更需要愿意重复劳动的廉价劳动力。看下面这段叙述:

……Return Path assigned two data analysts to spend several days reading 8,000 emails and manually labeling each one, the person says. The data helped train the company’s computers to better distinguish between personal and commercial emails.

Return Path declined to comment on details of the incident, but said it sometimes lets employees see emails when fixing problems with its algorithms……

还有另一家公司,他们要解决的邮件的智能回复,当机器无法从海量邮件里提取智能回复的样本时,人类就上场了:

Two of its artificial-intelligence engineers signed agreements not to share anything they read, Mr. Berner says. Then, working on machines that prevented them from downloading information to other devices, they read the personal email messages of hundreds of users—with user information already redacted—along with the system’s suggested replies, manually indicating whether each made sense.

随后《卫报》的报道则将这种现象一分为二地做了分析。首先,在保证数据隐私的前提下,人工智能需要人类智能这件事本身就很合理,不管是搭建神经网络还是参数调优,都需要人类智能的高度介入,而这一部分人类智能绝大多数来自该领域的专家、博士。

其二,人工智能领域还隐藏着一些「潜规则」,他们会将人类智能包装为人工智能产品进行售卖,《卫报》列举了其中一家公司:

In 2017, the business expense management app Expensify admitted that it had been using humans to transcribe at least some of the receipts it claimed to process using its “smartscan technology”. Scans of the receipts were being posted to Amazon’s Mechanical Turk crowdsourced labour tool, where low-paid workers were reading and transcribing them.

这种做法不仅是对投资人的欺骗,也是对用户的欺骗,而当涉及到与人类交流时,类似的做法也让技术陷入到另一个道德困境中,一个很简单的问题:当你认为你在和一个智能程序聊天时,你的言谈方式会和与人类(客服)聊天一样吗?

答案显然是否定的。事实上,早在上世纪 70 年代,MIT 科学家约瑟夫·魏泽鲍姆就发现了这个现象,当时,他创造了有史以来第一个 Chatbot 伊莉莎。

img

伊莉莎原本是用于在临床治疗中模仿心理医生。尽管伊莉莎的实现技术仅为关键词匹配及人工编写的回复规则,导致对话是单向的而且也会产生一些很奇怪的回复,比如,当用户提到自己的妈妈时,伊莉莎会以「你说你妈妈?」这样的句子来回复,但依然有很多人沉迷与伊莉莎的聊天,甚至会透露一些私密的个人信息。

《连线》也曾在 2017 年报道了一个研究发现,从而进一步佐证了这个事实:

“People are very open to feeling connected to things that aren’t people,” says Gale Lucas, a psychologist at USC’s Institute for Creative Technologies and first author of a new, Darpa-funded study that finds soldiers are more likely to divulge symptoms of PTSD to a virtual interviewer—an artificially intelligent avatar, rendered in 3-D on a television screen—than in existing post-deployment health surveys. The findings, which appear in the latest issue of the journal Frontiers in Robotics and AI, suggest that virtual interviewers could prove to be even better than human therapists at helping soldiers open up about their mental health.

硅谷资深记者约翰· 马尔科夫在《与机器人共舞》一书曾这样评价伊莉莎:「这证明人类习惯在与自己互动的对象中寻找人性存在的迹象,从没有生命的物体到提供虚拟人工智能的软件程序,无一不是如此。

img

这也就不难理解,为何当 Google 展示一个可以给人类打电话聊天的 Duplex 时会引发如此大的争议1,硅谷资深记者 Steven Levy 当时在 Twitter上写道:

Is it ethical to have a human-sounding robot interact with someone without informing the other party that he or she is in conversation with an it? Real question.

联想到刚刚完结的《西部世界》第二季里的一个场景:一个妹子在园区被男子搭讪,两人在上床前,女子要用游戏枪来「检验」这个男子到底是不是机器人。

随着剧情的发展,当机器人接待员混杂到人类之间,你如何确定和你聊天的这个「东西」是人类还是接待员?而当这个场景延伸到现实里,就像 Google 展示 Duplex 时那样,你如何确定电话那头到底是纯粹的机器还是人类操控的机器抑或纯粹是人类?

如果将上述讨论继续下去,会涉及越来越多的心理学、哲学以及伦理学知识,不过有趣的是,这些看似是围绕机器该做什么的讨论,其落脚点都在人类身上,或者可以这样理解,上述讨论只回答了问题的一个答案:人类到底是什么?

但这个问题还需要另一个答案,那就是怎么回答「人工智能/机器学习到底是什么?」

a16z 合伙人 Benedict Evans 最近的一篇文章就试图回答这个问题,这篇名为Ways to think about machine learning的文章读起来并不容易,但 Evans 把握住了当下人工智能/机器学习热潮的关键要素:数据

正因为数据,所以新型的自动化(或者说被包装的人工智能)才得以展开;同时,数据让人工智能/机器学习具备了上世纪 70 年关系型数据库的角色,成为驱动企业发展的新力量;这一系列的叠加,可能会让人工智能成为一种技术基础设施。

img

其中,Evans 对于关系型数据库和人工智能/机器学习之间的相似之处,以及公众如何误解自动化的叙述非常值得一读:

An important parallel here is that though relational databases had economy of scale effects, there were limited network or ‘winner takes all’ effects. The database being used by company A doesn’t get better if company B buys the same database software from the same vendor: Safeway’s database doesn’t get better if Caterpillar buys the same one. Much the same actually applies to machine learning: machine learning is all about data, but data is highly specific to particular applications. More handwriting data will make a handwriting recognizer better, and more gas turbine data will make a system that predicts failures in gas turbines better, but the one doesn’t help with the other. Data isn’t fungible.

This gets to the heart of the most common misconception that comes up in talking about machine learning – that it is in some way a single, general purpose thing, on a path to HAL 9000, and that Google or Microsoft have each built one, or that Google ‘has all the data’, or that IBM has an actual thing called ‘Watson’. Really, this is always the mistake in looking at automation: with each wave of automation, we imagine we’re creating something anthropomorphic or something with general intelligence. In the 1920s and 30s we imagined steel men walking around factories holding hammers, and in the 1950s we imagined humanoid robots walking around the kitchen doing the housework. We didn’t get robot servants – we got washing machines.

Washing machines are robots, but they’re not ‘intelligent’. They don’t know what water or clothes are. Moreover, they’re not general purpose even in the narrow domain of washing – you can’t put dishes in a washing machine, nor clothes in a dishwasher (or rather, you can, but you won’t get the result you want). They’re just another kind of automation, no different conceptually to a conveyor belt or a pick-and-place machine. Equally, machine learning lets us solve classes of problem that computers could not usefully address before, but each of those problems will require a different implementation, and different data, a different route to market, and often a different company. Each of them is a piece of automation. Each of them is a washing machine.

上述三段略显啰嗦的话可谓戳穿了围绕人工智能行业的诸多谎言,这些谎言在过去几年时间被巨头公司、媒体、资本甚至好莱坞包装起来贩卖,公众在一次次被诸如「人机大战」、「机器取代人类」的内容冲击之中,除了陷入到「我是谁、我从哪里来、我到哪里去」的终极提问,已然失去了对于技术发展的正确认知,或许,这是机器给人类的一道诅咒吧。


  1. 我曾在之前的会员通讯里对此做过分析,详见这里 ↩