让我们聊聊聊天机器人

介绍(Introduction)

While surfing a website, you must have come across a popup with an image of a person with headset and a chat window, intriguing you to ask a question. That, as you may already know, is a chatbot. Instead of scanning through the whole website, chatbots provide a quick and easier way of handling our queries and briefing about the services of the company.

在浏览网站时，您一定会遇到一个弹出窗口，其中显示了一个带有耳机和聊天窗口的人的图像，这使您很想问一个问题。 您可能已经知道，那是一个聊天机器人。 聊天机器人无需浏览整个网站，而是提供了一种快速简便的方法来处理我们的查询和有关公司服务的简介。

Chatbots have been trending in the recent years, but it is not a new concept. The term “ChatterBot” was first coined by Michael Mauldin in 1994, but the first chatbot was created in the 1960s. A chatbot is a dialogue system that simulates the behaviour of a human conversing with a user.

近年来，聊天机器人一直在发展，但这并不是一个新概念。 “ ChatterBot”一词最早是由Michael Mauldin于1994年创造的，但第一个聊天机器人是在1960年代创建的。 聊天机器人是一种对话系统，可模拟人类与用户交谈的行为。

那么，一切都是从哪里开始的呢？ (So from where did it all started?)

Inspired by the Turing test, Joseph Weizenbaum, a professor at MIT, developed a program ELIZA which imitated a therapist asking open-ended questions. It constantly interacted with the user by generating a response to the user queries. The program would break up the query into words and assign a value to each of them. The value of the word was determined by its importance within the sentence. It used pattern matching and substitution methodology to reorder the words in the form of a question. Sometimes, it just rephrased the query by emphasizing on the word with the greatest weight. If ELIZA couldn’t find a pattern to respond to, it used some default responses like “Can you elaborate on that?”.

受到图灵测试的启发，麻省理工学院的教授约瑟夫·魏岑鲍姆(Joseph Weizenbaum)开发了一个程序ELIZA，该程序模仿了治疗师提出的开放性问题。 它通过生成对用户查询的响应来不断与用户互动。 该程序会将查询分解为单词，并为每个单词分配一个值。 这个词的价值取决于它在句子中的重要性。 它使用模式匹配和替换方法对问题形式的单词进行重新排序。 有时，它只是通过强调具有最大权重的单词来重新表述查询。 如果ELIZA找不到响应的模式，则会使用一些默认响应，例如“您能详细说明吗？”。

For example, let the input be, “I WANT TO SLEEP ALL DAY.”. ELIZA assigns a weighted value to each word of that sentence. ELIZA attributes low values to pronouns (I), slightly higher values to action verbs (want to), and the highest value to the actual action (sleep all day). This helps ELIZA to flip around the input into a human response, by simply turning the values into a question, flipping the pronoun, and switching the verb to convey meaning. Thereby responding “What if you never got TO SLEEP ALL DAY?”

例如，让输入为“我想整天睡觉”。 ELIZA为该句子的每个单词分配一个加权值。 ELIZA将低值归因于代词(I)，将较高值归因于动作动词(想要)，将最高值归因于实际动作(整天睡眠)。 通过简单地将值变成问题，翻转代词并切换动词以传达含义，这有助于ELIZA将输入翻转为人类的React。 从而回答“如果您整天都没睡觉怎么办？”

Sometimes, the conversation moved forward in meaningful way, while other times a meaningless response was given. Even though ELIZA’s response was limited, it became quite popular and inspired other scientists in the field.

有时，对话以有意义的方式前进，而其他时候却给出了毫无意义的回应。 尽管ELIZA的React有限，但它变得非常流行并启发了该领域的其他科学家。

A timeline of some other notable chatbots:1972 - Kenneth Mark Colby, a computer scientist and psychiatrist at Stanford, developed a chatbot PARRY, which behaved as a paranoid schizophrenic patient. Due to the programmed behaviour of shifting the weightage of the input verbs, its response prompted the user to elaborate himself more. 1984 - William Chamberlain and Thomas Etter developed RACTER (short for raconteur- meaning storyteller), an AI program that generated random English language prose. It even wrote a book, “The Policeman’s Beard is Half-constructed: Computer Prose and Poetry”.1988 - Rollo Carpenter created Jabberwacky, an entertaining chatbot, that learnt new responses based on real-time user interactions, rather than being driven from a static database. 1991 - Creative Labs released DR. SBAITSO, a fully voice operated AI psychologist program for MS-DOS based systems. SBAITSO was an acronym for Sound Blaster Artificial Intelligent Text to Speech Operator.1995 - A.L.I.C.E. (Artificial Linguistic Internet Computer Entity) uses Artificial Intelligence Mark-up Language (an XML dialect for creating natural language software agents), which allowed more sophisticated conversations. Its code is available as open-source.2001- SMARTER CHILD, a precursor to Apple’s Siri and Samsung’s S Voice, provided information from the web. It was deployed on instant messaging services.

其他一些著名的聊天机器人的时间表： 1972年-斯坦福大学的计算机科学家兼精神病学家Kenneth Mark Colby开发了一个聊天机器人PARRY，该机器人表现为偏执型精神分裂症患者。 由于移动输入动词权重的程序化行为，其响应促使用户进一步完善自己。 1984年-威廉·张伯伦( William Chamberlain)和托马斯·埃特(Thomas Etter)开发了RACTER(raconteur意思是讲故事的人的缩写)，这是一个产生随机英语散文的AI程序。 它甚至写了一本书，“警察的胡子是一半构造的：计算机散文和诗歌”。 1988年-Rollo Carpenter创建了一个有趣的聊天机器人Jabberwacky，该机器人基于实时用户交互而不是从静态数据库驱动学习了新的响应。 1991年-Creative Labs发布了DR。 SBAITSO，一个完全语音操作的AI心理学家程序，用于基于MS-DOS的系统。 SBAITSO是Sound Blaster人工智能文本到语音运算符的首字母缩写。 1995年-ALICE(人工语言Internet计算机实体)使用了人工智能标记语言(一种用于创建自然语言软件代理的XML方言)，从而可以进行更复杂的对话。 它的代码可以作为开源代码获得。 2001年-SMARTER CHILD(Apple的Siri和三星的S Voice的前身)提供了来自网络的信息。 它已部署在即时消息服务上。

With advancements in the field of natural language processing and AI, virtual assistants cam into existence.2010 - Apple’s Siri, a voice based personal assistant. It is from here, that the virtual assistants started gaining recognition.2012 - Google now makes recommendations based on user activity.2015 - Amazon Alexa added the concept of home automation through speech processing.2015 - Microsoft Cortana2016 - Tay (discussed later in limitations section)

随着自然语言处理和AI领域的进步，虚拟助手应运而生。 2010年-苹果公司的Siri，基于语音的个人助理。 从这里开始，虚拟助手开始获得认可。 2012年-Google现在根据用户活动提出建议。 2015年-亚马逊Alexa通过语音处理增加了家庭自动化的概念。 2015- Microsoft Cortana 2016- Tay(稍后在“限制”部分中讨论)

Microsoft

微软

An Honourable MentionI hope you remember Office Assistant a.k.a. Clippit, Microsoft Office’s intelligent user interface. Even though, it is not a chatbot but it can be considered as a precursor to them. Working with the Microsoft help section, Clippit offered help for using the Office products.

荣誉奖我希望您还记得Office Assistant(又名Clippit)，这是Microsoft Office的智能用户界面。 即使它不是聊天机器人，也可以视为它们的先驱。 使用Microsoft帮助部分，Clippit提供了有关使用Office产品的帮助。

为什么选择聊天机器人？ 以及在哪里使用它们？ (Why Chatbots? And Where to Use Them?)

Responding to the customer’s query quickly and efficiently greatly enhances customer experience. Many a times, customers get irritated when they can not retrieve the required information. Calling customer support is the last thing they do, knowing the long waiting time and uncertain support. This is where chatbots win the game. Since they are available 24/7 and are not affected by the volume of customer queries, therefore they not only help in retaining customers but also lower the costs of maintaining human customer support.

快速有效地响应客户的查询，极大地增强了客户体验。 很多时候，当客户无法检索所需的信息时，他们会感到恼火。 知道漫长的等待时间和不确定的支持，致电客户支持是他们的最后一件事。 这是聊天机器人赢得比赛的地方。 由于它们是24/7全天候可用的，并且不受客户查询量的影响，因此它们不仅有助于留住客户，而且可以降低维护人工客户支持的成本。

“265 billion customer support requests are made every year, and it costs businesses a whopping $1.3 trillion to service them.” -IBM

“每年提出2650亿个客户支持请求，为企业提供服务的成本高达1.3万亿美元。” -IBM

1. Chatbots are mainly used in assisting customer service and in marketing automation.
   聊天机器人主要用于协助客户服务和营销自动化。
2. They are being widely used in e-commerce, financial institution, educational, news, and health websites.
   它们被广泛用于电子商务，金融机构，教育，新闻和健康网站。
3. These are also being used for internal activities in companies, such as when requesting sick leave.
   这些还被用于公司的内部活动，例如在请病假时。
4. It is also being used in toys, e.g. Hello Barbie.
   它也被用于玩具中，例如Hello Barbie。
5. A chatbot on WhatsApp has been deployed for queries related to Covid-19.
   已经部署了WhatsApp上的聊天机器人，用于与Covid-19相关的查询。
6. Malicious Use: Spamming, befriending for illegal and harmful activities.
   恶意使用：散布垃圾内容，成为非法和有害活动的伙伴。

一切看起来都很好！ 可是等等… (It all looks so nice! But wait…)

Even though there has been a lot of improvement in this field, but still there are some limitations which include:

尽管在该领域已经有了很多改进，但是仍然存在一些局限性，其中包括：

1. Limited Knowledge Base, thus unable to process unsaved queries.
   知识库有限，因此无法处理未保存的查询。
2. Limitations in NLP(accent problems, language barriers), thus affecting its user base.
   NLP的局限性(重音​​问题，语言障碍)，因此影响了其用户群。
3. Difficulty in managing non-linear conversations. Chatbots cannot handle the situations when the conversation goes back and forth on a topic.
   难以管理非线性对话。 当在主题上来回对话时，聊天机器人无法处理这种情况。
4. Microsoft created an AI chatbot, Tay, which simulated a teenage girl that learns from previous interactions. They experimented on it by deploying it on Twitter. Tay’s algorithm was trained on a anonymized public dataset and was programmed to discover patterns of language through social media interactions. Initially, Tay interacted harmlessly but in few hours, she started tweeting highly offensive things. Tay had to be removed within 16 hours of its deployment. This behaviour could be a proof of the social media’s inherent toxicity, where people bring out their worst. Even though Microsoft was criticized for its AI bot, the event opened up the difficulties of an AI bot that could be easily corrupted when deployed in the open world. Learning from its failures, some months later, Microsoft released Zo, a “politically correct” version of the original bot.
   微软创建了一个AI聊天机器人Tay，该机器人模仿了一个从以前的互动中学到东西的少女。 他们通过将其部署在Twitter上进行了试验。 Tay的算法在匿名的公共数据集上进行了训练，并经过编程以通过社交媒体互动来发现语言模式。 最初，Tay进行了无害的互动，但在几个小时后，她开始发布推文，发动令人反感的事情。 Tay必须在部署后的16小时内删除。 这种行为可能证明了社交媒体固有的毒性，人们在其中表现出了最坏的表现。 尽管微软因其AI机器人而受到批评，但这次活动为AI机器人带来了很多困难，而这些AI机器人在开放世界中部署时很容易遭到破坏。 从失败中吸取教训，几个月后，微软发布了Zo，这是原始bot的“政治上正确的”版本。

虚拟助手和聊天机器人之间的区别 (Difference between Virtual Assistant and Chatbot)

Chatbots are mainly used in customer services, for knowledge acquisition. Whereas, virtual assistant is more of a personal agent, assisting in daily activities such as scheduling, typing, getting directions, booking reservations, etc. So, it’s mainly the application area where these two differ.

聊天机器人主要用于客户服务中，以获取知识。 而虚拟助手更像是个人代理，可以协助日常活动，如日程安排，打字，获取指示，预订预订等。因此，这是两者之间主要区别的应用领域。

结论 (Conclusion)

Looking from a particular business perspective, chatbots have greatly evolved, there are a lot of examples where chatbots have made life easier and more efficient. They efficiently do the tasks assigned to them in a particular domain, but the problem arises when looking from the perspective of AI in chatbot. Due to limited knowledge base, they cannot respond meaningfully to all queries. Since we are still unclear of what AI really is, we cannot achieve it’s full potential.

从特定的业务角度来看，聊天机器人已经有了长足的发展，有很多例子使聊天机器人使生活更轻松，更高效。 他们可以高效地执行在特定域中分配给他们的任务，但是从聊天机器人中的AI角度看时会出现问题。 由于知识库有限，他们无法对所有查询做出有意义的响应。 由于我们仍然不清楚AI到底是什么，因此我们无法充分发挥AI的潜力。

I would like to quote John McCarthy here, who said, “when something starts to work, people will cease to call it artificial intelligence”.

我想在这里引用约翰·麦卡锡(John McCarthy)的话，他说：“当某些事情开始起作用时，人们将不再称其为人工智能”。

翻译自: https://medium.com/swlh/lets-chat-about-chatbots-98d23a4694ed