持续几年的“聊天机器人革命”带来了巨大的期望。营销人员和未来主义者幻想着智能虚拟自主代理人，他们会比我们更好地了解我们自己，从谈话片段中收集我们的愿望，并将我们的每一个命令转化为即时行动。这个想法有点像《2001：太空漫游》中的哈尔(在他发疯之前)和漫威宇宙中的贾维斯的混合体。
2018年初，我们可以遗憾而又相当坦诚地说，聊天机器人没有实现这些伟大的期望。即使微软的Twitter 机器人成为新纳粹的支持者，其他的机器人也出了问题，但在聊天机器人行业的整体进展并没有带来我们所承诺的巨大利益。2016年和2017年不是“对话式商务”的年份，聊天机器人也不是“新应用程序”，常人可以通过Messenger或Amazon Echo等平台访问的人工智能似乎比智能化更具人工性。Facebook最近不得不关闭其虚拟个人助理“M”，据称是因为“M”过分依赖人类来提供有意义的答案。
现在是时候去理解为什么我们的期望和预测（主要是由具有良好记录的专家做出）完全没有达到预期目标。我坚信，聊天机器人仍然是人们在线互动的一种令人兴奋的新方式。这就是为什么我认为，对于营销人员、开发人员和企业家来说，了解我们所犯的错误以及我们需要修复哪些错误才能使技术发挥其真正的潜力是极为必要的。
关于机器人和人类
这次半惨败的核心的最大误解是机器人作为虚拟助理、客户服务运营商或个人购物者，可以无缝地、完全取代人类。
了解笑话，熟悉书面和口头语言的细微差别，运用直觉来选择正确的词语或行动，这些都是人工智能代理者未来几十年需要学习的。这些特质中的某些特质可能永远无法在非人类的“大脑”中复制。当我们被引导相信：我们可以与机器人互动，就像我们与真人互动一样(或者在某些情况下，甚至没有被告知，我们正在与一个自动化系统聊天)，再加上对整个技术的普遍不信任，结果令人沮丧。
机器人是机器人，人类是人类。计算机程序和系统确实非常擅长某些任务——快速查找信息、进行繁重的计算和存储成千上万字节的数据——但在其他方面却非常糟糕。他们尤其不擅长产生任何与人打交道，操作员必须感到的最低程度的同理心。这两者混合起来可能是致命的。
谈话VS信息
2017年，聊天机器人大多脸上都是扁平的第二个主要原因是人们通常使用这些界面。营销人员和开发人员被这种新玩具冲昏了头脑，认为他们最终会创造出“对话式”代理人，人们可以像和朋友或人类助手一样进行对话。我们梦想着像托尼斯塔克一样，和一个无所不在、看似无所不知的网络助理聊聊诙谐的笑话和交流深刻的思想。但那是错误的。对于非人类来说，对话是很难维持的，一旦一台计算机失去了它在双向互动中所处的位置，结果就会很快打破整个局面的“魔力”。
聊天机器人应该有明确定义的路径和漏斗，通过这些路径和漏斗他们引导访问者走向明确或隐含的目标。这个过程应该包括一系列的问题和答案，甚至可能包括链接和按钮。这并不意味着机器人不应该实现和提供自然语言处理，这意味着我们应该明确界定机器人可以执行的领域。我们应该建立一个机器人无法跨越的界限，并支持和控制这个系统，以便它能够提供更有效和更充实的体验。
此外，明确的目标可以让营销人员和企业家更精确地跟踪和衡量这些工具的投资回报率。我们对这项技术感到非常兴奋，以至于忘记了我们所做的任何事情都需要时间、金钱和精力，而且我们大多数时候忽略了分析的一部分。这是另一个巨大的问题，因为我们没有公开(或在许多情况下，甚至是私人)的数字告诉我们，在聊天机器人上的电子商务交易或用户参与如何与其他更传统的渠道比较。这种盲目的实验可能只有大品牌才有能力做到。
象牙塔
这就引出了我的最后一点。小型企业并没有完全接受聊天机器人，因为他们认为这种技术很复杂，很难实施，而且几乎无法衡量。像人工智能、机器学习和自然语言处理这样的名字随处可见，让一般小企业的所有者感到害怕。
例如，餐馆老板可能已经在他们的公司的Facebook页面上拥有大量的关注，并且在他们的带领下进行了几次成功的广告活动，但他们希望更好地为潜在客户和通过Messenger提问的客户提供服务。如果他们不得不花费巨资去运行一个机器人，但是没有一个简单的方法来理解它是否有效，他们可能会认为聊天机器人不在他们的控制范围之内，甚至可能是一个骗局。
我们需要提出更简单但同样有效的解决方案，将这种技术应用到我们的日常生活中，并帮助小企业的企业家利用自动化社交互动的力量。只有在我们掌握了这一点之后，我们才能专注于建造天网，如果我们还想的话。
每天都有数十亿人在使用 Messenger。他们将它放在口袋中，并准备使用它来接收非常简单，但立即可行，且高度个性化的信息——这项技术已经使我们能够做到这一点。我们确定要用有缺陷的聊天机器人执行，而浪费这个机会吗？
原文
3 reasons chatbots didn’t meet industry expectations in 2017
The dawn of the “chatbot revolution” less than a couple of years ago ushered in great expectations. Marketers and futurists fantasized about intelligent virtual autonomous agents that would understand us better than we understand ourselves, glean our desires from snippets of conversations, and turn our every command into an immediate action. The idea was something like a mix ofJARVIS from the Marvel Universe and HAL from 2001: A Space Odyssey (before he went crazy).
At the beginning of 2018, we can sadly yet quite confidently say that chatbots have not met these great hopes. Even discounting Microsoft’s Twitter botbecoming a neo-nazi supporter and other bots gone wrong, overall advances in the chatbot industry have failed to deliver the huge benefits we were promised. 2016 and 2017 were not the years of “conversational commerce,” chatbots are not “the new apps,” and the AI that regular people can access through platforms such as Messenger or Amazon Echo still seems much more artificial than intelligent. Facebook recently had to shut down its virtual personal assistant “M,” allegedly because it relied too much on humans to provide meaningful answers.
It’s time to understand why our expectations and the predictions — made mostly by experts with proven track records — totally missed the mark. I firmly believe chatbots still represent an exciting new way for humans to interact online. This is why I believe it is essential for marketers, developers, and entrepreneurs to understand what mistakes we made and what we all need to fix for the technology to fulfill its true potential.
Of bots and men
The biggest misunderstanding at the core of this semi-fiasco is the idea that bots can seamlessly and entirely replace humans as virtual assistants, customer care operators, or personal shoppers.
Understanding jokes, picking up the nuances of written or spoken languages, and using intuition to choose the right words or actions are things AI agents are decades away from learning. And it’s possible some of these qualities may never be replicable in a non-human “brain.” When we are led to believe we can interact with a bot the way we engage with a live person (or in some cases are not even told that we are chatting with an automated system) the result is frustration, coupled with a general mistrust of the technology as a whole.
Bots are bots, humans are humans. Computer programs and systems are really good at certain tasks — finding information quickly, doing heavy computations, and storing petabytes of data — but are very bad at others. They are especially bad at generating that minimum level of empathy any operator dealing with people must feel. Mixing the two can be deadly.
Conversation versus information
The second reason chatbots mostly fell flat on their faces in 2017 is the interface that was commonly used. Marketers and developers got carried away by this new toy and thought they were finally going to create “conversational” agents that humans could dialogue like they would with a friend or a human helper. We dreamed of being like Tony Stark and exchanging witty jokes and deep thoughts with an omnipresent and seemingly omniscient cyber assistant. Well, that was a mistake. Conversations are hard to sustain for non-humans, and once a computer loses track of where it is in a two-way interaction, results can quickly break the “magic” of the whole situation.
Chatbots should have clearly defined paths and funnels through which they lead visitors toward explicit or implicit goals. The process should include a series of questions and answers, and possibly even links and buttons. This doesn’t mean bots shouldn’t implement and offer natural language processing, it means we should clearly define the realm in which a bot can act. We should create boundaries that the bot cannot cross and that support and contain the system so it can offer a much more effective and fulfilling experience.
In addition, well-identified goals allow marketers and business owners to track and measure the ROI of these tools more precisely. We were so excited about this technology that we forgot that anything we do costs time, money, and energy, and we mostly skipped the analytics part. This is another huge problem, since we do not have public (or in many cases, even private) numbers that tell us how ecommerce transactions or user engagement on chatbots compare against other, more traditional channels. This level of blind experimentation is something only brands with big pockets can afford to do.
The ivory tower
And this leads us to my last point. Small businesses have not fully embraced chatbots because they see the technology as complex, difficult to implement, and nearly impossible to measure. Throwing around names like artificial intelligence, machine learning, and natural language processing scares the heck out of the average small business owner.
For example, a restaurant owner may already have a hefty following on their establishment’s Facebook page and a couple of successful ad campaigns under their belt but would like to better serve prospects and customers asking questions via Messenger. If they have to spend big bucks to implement a bot but don’t have a simple way to understand whether it’s working or not, they’ll probably think chatbots are out of their league and maybe even a scam.
We need to come up with simpler yet equally effective solutions that bring this technology into our daily lives and help small business owners leverage the power of automated social interactions. Only after we master this can we focus on building Skynet, if we still want to.
Literally billions of people are on Messenger every day. They have it in their pockets and are ready to use it to receive very simple yet immediately actionable and highly personalized information — something this technology already allows us to do. Are we sure we want to waste this opportunity with faulty chatbot execution?
