Artificial intelligence, a miracle of science, that knocked the socks off everyone. To explain it the best, Artificial intelligence is the power to create something almost human. Machines that do not just calculate figures within seconds, but also perceive human emotions, expressions and also respond to them appropriately.
Machine learning is a prime tool in the hands of AI. It derives information and carries out functions such as learning, analyzing, etc., from the data provided.
Artificial intelligence finds its application in various fields. Scientists and developers envision developing bots that could one day be used in the activities that require the most precision, like surgeries. They are already being used in the manufacturing sector, across the world. Artificial intelligence has already conquered the customer service sector with its bots and how.
More than 60 percent of users of the ‘cloud’ encounter a bot, a chatbot to be specific, at least once a day. Every time you visit a website and find yourself communicating with its customer service department, there is a great possibility, that you are interacting with a chatbot.
Initially, chatbots were these robots that relied solely on human guidance and were known for their monotony during interactions.
However, over the years, with the help of tools such as machine learning and natural language processing, they have become more expressive and comprehend better, therefore making their interactions with the customers, closer to human interactions.
ChatBots AKA these conversation agents are meant to carry out a conversation with real people, in the form of spoken words or texts. These software applications are generally available in two forms. Either via stand-alone applications or Web applications.
Most of the platforms including an online retail store or a social media page, every other industry has employed chatbots, to satisfy their customers. Incorporating Chat Bots has become a growing trend, also because it is no more a computer science miracle that requires extraordinary skill. In present times, any ordinary person, without much experience or coding skill, and a CS degree, can make a chatbot of his own. Several websites offer some drag and drop interfaces to create your very own conversational agents.
Now, if you’re wondering, how these chatbots function, here’s the plot. These chatbots process the data that they receive from the user, via a natural language processing tool. Before responding to the user’s text, they try to understand the meaning that the user wishes to convey. After understanding what has been said, they respond to the user, according to how they have been programmed.
These substitutes for customer service reps have completely changed the scenario of the market. Here are some stats to prove the way these have taken several industries by storm.
The rate at which the chatbots have conquered the market, suggests that by the year 2022, these AI chatbots shall overtake 90% of customer interaction. The bank sector has been receiving great feedback from their customers. The bots have been praised for being more responsive and precise at helping them than the human force. Also, incorporating these chatbots in its infrastructure can help a company at cost-cutting by almost 30%.
It is being predicted that in the coming years, nearly 85% of customer interaction shall be conducted by AI chatbots.
Talking of investments, 50% of businesses are willing to take a leap and employ chatbots, instead of advancing their mobile apps. These bots have been acknowledged for their 24-hour services and 37% of people have chosen bots over humans, in the case of customer service emergencies.
Talking of accomplishments, Facebook employed over 3 lakh bots for user interaction, in the year 2018.
The blue bot has been known, to have sent about 2 million messages to 500000 customers. A study proposed that as many as almost 27 percent of people in the world are enthusiasts of technology, especially artificial intelligence.
You might be surprised to know that almost 1.4 million people use or encounter chatbots on a routine.
These chatbots offer a stack of benefits to companies and streamline customer interactions. Most chatbots happen to address about 80 percent of standard questions and problems with the utmost efficiency.
This feature led to, 34 percent of consumers of an online retail company, choosing to interact with artificially intelligent bots, instead of the human force of customer service executives in the year 2017.
With the data provided above, it is quite evident that chatbots are no more a fancy and rare service that is yet to seep into the market. A multitude of customers (almost 67%) used chatbots in the year 2019. Its fast and efficient services have managed to amaze a lot of people.
Amongst the various ‘awe’ features of the chatbots, including their efficiency, consumers are most ‘awed’ by the 24 – hours service feature of these bots.
A survey that interacted with users of the ‘cloud’, revealed that about 64% of them said that the 24-hour service was the best feature of the chatbots. The rest 36 percent felt that the ability of the bots, to respond quickly and throw prompts to help the users and make their search better, was their most beneficial feature. As per the recent statistics published by ‘Drift’, 32 percent of customers used chatbots to answer a question or regular searches. 29% of consumers found them useful for brief researches and fetching detailed information on a subject.
- 27 percent of consumers used these to resolve their problems or to get their complaints addressed.
- 23 percent of people used them to pay bills at a restaurant, hotels, bars, pubs, etc.
- 22 percent of people use them to make reservations at hotels, restaurants, for dinner dates, etc.
- 19 percent found them helpful in scheduling meetings to carry out some professional commitments, etc.
- Other 19 percent found them useful while communicating with a brand or an online service provider.
- 17% of people used them while buying stuff online.
- On the other hand, some customer care executives also acknowledged the assistance they are provided with, by the chatbots.
- 34 percent of them believed that the time they saved with their service, helps them invest in deep thinking and coming up with more creative ideas for their businesses.
Case Study – 1
With the data, presented before this, it can be easily comprehended that the popularity and engagement of chatbots are growing meteorically. Chatbots have occupied most industries and have handled the customer service sector with real dexterity.
However, most of the chatbots that we come across, function in English, irrespective of the region, to which the customer belongs. Therefore, developing a chatbot that could communicate in a language other than English, was a real task and also, a major necessity.
A prime underprop for the English language being the language of communication, was the availability of NLP libraries.
Natural language processing, abbreviated as NLP, is a subdivision of linguistics, Artificial intelligence, information engineering, and computer science. It is aimed at carrying out interactions between computers and human beings, in natural languages. This is used to train computers to understand commands and respond better. There are several NLPs available, but only for English.
Other substructures that work along with NLPs to comprehend a language are POS taggers, word 2 vec models, etc.
POS tagging refers to a part of speech tagging. This process involves marking verbs, subjects, helping verbs, objects, and other parts of sentences, to identify, which part of speech they belong to. This is also known as grammatical tagging.
Word 2 vec is a technique employed to process a natural language. This follows a neural network model to understand the relation of a word, to a text. It can identify synonyms and can prompt words to complete a sentence.
Creating a new NLP for a new language can be extremely time-consuming.
As part of their plan to develop a second language for bots, Smart Cat resorted to the Serbian language and implemented a different approach.
They used a dataset of sentences that were unlabelled and in Serbian and performed machine learning procedures on them.
These sentences were processed and the texts were first converted to lower case. The extra white spaces, punctuations, and unnecessary numbers were removed.
The above-mentioned actions are performed in most NLP cases and can be performed on texts that have been written in any language.
Had the text been in the English language, you could continue by performing stemming, lemmatization, removing sparse terms, and stop words, which occur quite often in a sentence.
Stemming is a process that involves, reducing a text or a sentence to its root form, by stripping off its unnecessary parts.
Lemmatization refers to a process, as part of which words are analyzed morphologically and their inflection endings are removed, without creating much difference to the meaning represented by the sentence.
Stop words are words like a, an, as, such, that can create difficulty in communicating with a computer or a machine.
Thus, a tf-idf statistic was performed on the dataset. A large vocabulary was created for Serbian and the words that were included in it had tf-idf index assigned.
Tf-idf is the abbreviation for the term frequency-inverse document frequency. It reflects the importance of a word in a text. It can be understood as a term weighting scheme.
Stemming and lemmatization were avoided for two reasons:
There are many suffixes, rules, and exceptions in Serbian which makes it difficult to perform the above-mentioned processes on sentences.
Performing the above processes could lead to changed dialects and grammatical errors, therefore causing wrong conclusions.
They finally had some processed sentences to work on. Had it been English, they would have opted for POS tagging, but no POS taggers are available for the Serbian language. Therefore, they decided to create a word 2 vec model for it. It helped them understand the relation and similarity between words, in a better way.
They trained the model by using 400,000 sentences and word embeddings were created. They then used K means to cluster together, the retrieved embeddings. They received clusters with greetings, complaints, personal information, gratitude, etc.
K means are meant for classifying ‘n’ observations in ‘k’ groups or clusters.
Word embeddings are words that derive the same meaning and have a quite similar presentation.
This clustering algorithm was employed only as a tool for classification.
Whenever a new message was received by the system, the chatbot analyzed and classified the text according to the class that it belonged to.
It then reverted or sent a response assigned to the class that was predicted, only if the accuracy for prediction was greater than the threshold value set for it.
After the execution of all the techniques employed, the conclusions were recorded by analyzing the conversations.
Out of the 100 conversations that it participated in, 210 messages were exchanged. Out of these 210 messages, the chatbot was unable to handle 23 messages and it happened to mishandle or handle incorrectly 11 messages. It was therefore derived that in about 84% of cases, the bots were able to respond and in about 95% of the cases, they were able to handle the conversations appreciably.
However, better results were reported when they used n-grams. Here, n was the value of class in the range of 1 to n. There were a large number of classes, for example, some were greetings, while some were greetings with questions. The process took a bit longer than normal but served better results.
The chatbot could quite easily understand the messages that it received in Serbian, and could efficiently respond to them also.
The translation of a conversation has been provided below :
Customer / user : Hello
Chat Bot: Hello! Thank you for connecting with us. How may I help you?
User: Your website seems to have frequently occurring glitches, why so?
Chat Bot: We will immediately look into the matter and let you know about the updates.
User: Okay, thanks.
Chat Bot: Thank you for connecting with us. If you have any queries, we will be pleased to help.
Case Study: 2
Chat Shopper is a leading fashion engine, employing AI-aided conversational agents. It is a Germany-based company, that visions to improve the quality of fashion searches for their customers and make each product, available online, and just a text away.
Initially, the company only had its chatbots functioning in German. The company informed that they use wit.ai as their NLP system and have Node.js for bot logic.
Wit.ai is a Y-combinator company with its headquarters located in Palo Alto. It is a platform that enables developers to make apps that can communicate to the users effectively.
In the process of adding multiple languages to the chatbot, they started by substituting the answers that were saved primly in the bot engine of the wit.ai. These were replaced by placeholders that they had resolved in their bot logic, separately for separate languages. They then created a clone of the German bot, and trained this with an instance in English, to understand messages from the users and respond to them.
They then began the process of feeding answers to their bots. The process started with translating the file that was initially in German, to English or any other language. This file was then presented before a native speaker of the language, to go through, and do the necessary corrections.
The Users who connected were answered in languages, according to the locales that they had set on their Facebook profile. Also, the users were aided with an option to change the language of the bot, according to their preference.
Meanwhile, they needed to assure that both the wit.ai instances were equally complex at developing and providing new features.
The bot devs even believed that translating a chatbot was very much similar to translating the output of a web application. Since, in both the cases, the front-end is occupied with the placeholders, while the back end has language files and also includes the strings that are supposed to be used or have been used in the application or the chatbot, according to the user’s language preference.
However, there is a basic difference between the two. In the case of the interface used in the chatbots, since it is conversational, the input provided to it by the user, is also in the form of a complex text. An efficient Natural language processing engine is required to analyze this input and then respond to it.
Therefore, it is never just about responding or translating the output, but also requires a series of NLP procedures for every new language of operation.
The bot developers also believe that in the present times, creating a chatbot is not a very systematic or structured process. We do not have the kind of environment and the functional tools to aid the streamlined process of developing bots, like we have Android Studio and x code, for web applications.
Whenever you receive a message and respond to it, there is a complete process that goes between these two tasks. However, this process varies for different bots and machines. Bot developers can choose a programming language of their choice, like Python, etc. They could design the entire bot and it’s working, in whatever way they wish to, and employ the existing ii8n patterns and plugins in different ways, depending upon the language that they have chosen for programming.
The demand for chatbots in the fashion industry especially online retail stores has grown exponentially. In the year 2014, about 77% of customers of these online retail stores desired to get their problems addressed and solved via email or phone. While 23% preferred to use social media and found chatting, a more comfortable alternative.
As per the reports published by Forrester, 44 % of people seek the help of an expert, to guide them through the process, whenever they are making an online transaction. Companies naturally are choosing chatbots over human employees to work out these processes, since bots are more precise and efficient enough at handling such matters, and above all, help them at cost-cutting.
Talking about some of the well-known failures in the field of AI, was a great lesson for bot developers across the world, a developer asserted, that Microsoft’s AI Twitter Conversation Bot suffered a major set back and the program had to be shut down, shortly after its release. The reason given for its failure is believed to be the malfunctioning of the chatbot.
Many experts believe that the team of Microsoft should have hired human assistants for each bot, similar to the ‘We Chat’s’ team, who believed in the fact that machines were man’s creation after all, and could suffer a malfunction, and therefore they provided each of their bots with a human assistant, to supervise their functions.
Case Study : 3
Visa Bot is a leading company, with its headquarters in San Francisco’s Bay area. The company works to streamline the process of visas, for the USA. The company uses Artificial Intelligence technology incorporated in its infrastructure, making the process more efficient.
The developers at Visabot wanted to incorporate multiple language features into their bots. They began with developing a bilingual bot that would communicate with users in two languages, and they developed it from scratch. Developing the bot from scratch was preferred by them, since it would keep them away from any kind of limitations and constraints, that they would have faced, had they used a bot builder to ease their work.
The bot devs asserted that the bilingual bots were quite easy to be operated. The users were provided with a list of languages, from which they could choose the language that they preferred to communicate in. This was done during the initial interaction with the user. The bots already had their scripts, saved in the database that they used. This process is quite similar to the one that takes place on multilingual websites. This guarantees that the database architecture has been selected correctly and the execution of the language selection feature is the least complicated. They chose to work on a very simple approach for developing multilingual bots. However, the developers also asserted, that one can be as creative as possible without a drawback because the server is completely in the hands of the developer. Had they been working on the builder side, it would have been quite difficult to achieve the same extent of creativity. Builders are great options for those devs who wish to avoid coding.
One of the chatbot developers asserted that when the users used an emoticon to express themselves, or typed their text in English, it became quite complex to comprehend the text.
After they succeeded at this initial accomplishment, they planned at developing bots that could communicate in Spanish, as well as Chinese.
The bot often encounters cases, when the users in the middle of a conversation, alter their language. In such situations, there are language detectors present in the bots. These language detectors work to identify the altered language, with the help of the percentage of words that are pronounced. The chatbot then provides the user with an option to change the language of the bot, based on the highest proportion.
Case Study: 4
Vic Yankoff, an indie developer shared his experiences and approaches to create a multilingual bot. He has had a great experience at creating Polyglot bots.
Polyglot bots are chatbots that speak multiple languages and can communicate with the user, in his or her preferred language. They not only make the user experience better but help to increase the customer base quite rapidly.
Vic suggested that if you wish to localize your bot, Google Translate’s API can be an amazing tool. Google translate is not very efficient in translating sentences with grammatical accuracy, but it very successfully translates the keywords that make it extremely easy for the chatbots to comprehend what the user desires to convey with his or her texts. He also shared, how he prefers buttons too, as they can be quite easily localized. However, many people don’t agree with the same. The Buttons guide the developer and allow him to know what action must be taken. The only complication that remains, is to translate the button labels, as the bot dev already knows the postbacks and has an idea of the same.
Talking about some of the popular platforms, he asserted that Facebook provides the dev with the language that the concerned user is using on the social networking site and therefore, localization is quite easy when it comes to Facebook.
Whenever a user resorts to a chatbot, Facebook tells you what language is used by the user. The information is usually provided in form of abbreviations, such as en is used to depict English, de stands for German, sp represents Spanish, etc. With this information, the bot developer can set the bot to interact in the same language.
A multilingual bot is quite prone to bugs, and so debugging is a major necessity. The bot needs to be efficient enough at understanding what the user is expecting from it and should be able to fulfill the expectations. Hence, a professional is required to debug the entire system, carefully and with precision.
Vic also affirmed that creating a multilingual bot could be quite a task, but it has a stack of benefits, once you’ve accomplished it. It helps you conquer a larger customer base and therefore, seep into the market better. The growth rates are exponential.
He also suggested that your translation files must be stored at the correct place. The files should be stored at places where the linguists can access them with ease. Do not store them in code repositories.
When you are going through the process of NLP training, your linguists should be able to reach the input or output files with ease, so that they can amend them further, as per requirements.
He suggested that Google spreadsheets could prove an efficient tool to be involved in the process of translation. This helps you to stay updated and at peace with whatever is being taught by the linguists and every training that becomes a part of the process.
One of the experts, working on these multilingual bots, Klemens Zleptnig talked about the correct time, when you should add the languages to the system. Addressing the same question, he asserted that, firstly it must be analyzed that where do most of your customers come from. After identifying your customer base, you must select the languages that your customers would want to communicate in. Post this, these languages can be added to the bot’s system.
There are various companies and start-ups that have presented an amalgamation of technology and creativity, and the results are quite impressive. Following is a list of companies that have Incorporated AI and chatbots and have come out with flying colors.
A Russian Company called Endurance developed a ChatBot for patients who suffer from Dementia.
Dementia is generally, an old age disorder that prominently is about people losing their memory. People with dementia find it difficult to remember even the simplest information and struggle to carry out a conversation, even with a family member. As the disease advances, their condition deteriorates.
So this chatbot by endurance keeps a record of the conversations that the patient has participated in. It records deviations during these conversations. The family members can check the rate of loss of memory and identify the deterioration of the patient’s state. The bot is available in the Russian version. However, the devs and researchers are still working on its English version.
This bot can be a savior for the insomniacs.
Insomnia is a disorder that is characterized by the loss of sleep. People suffering from insomnia tend to express a feeling of frustration ad suffocation. When the world rests its mind in peace, its mind forces it into thinking, panicking and therefore leads to loss of sleep.
Insomniacs also find themselves alone at night, since they often do not find someone, they can talk to.
Insomnobot 3000 by Casper could be a help, in such cases. It offers a conversation to insomniacs, making them feel less alone. However, the conversations are quite imaginative, which could at times backfire, but in many cases, it does help.
Where on one hand, different industries are introducing their chatbots for customer interactions, Disney has taken a step forward to engage its audience and how.
Disney has proposed a game, in which its audience can solve imaginary crime cases with lieutenant Judy.
Lieutenant Judy Hopps is a character from Disney’s creation, Zootopia.
The users can communicate via the chatbot and let their suggestions reach Judy, for the investigation. It’s a fun way to engage their young audience and have them interact more. The younger fan base of Disney’s characters is loving this chatbot and of course the game that it is a part of.
This superhero science fiction has created a fan base of its known. It is known for its larger-than-life experience and impressive casting. Several film critics have even said that the Marvel universe is expanding at a greater pace than the original one. To add to its glory, the Marvel series has involved Chat Bots to engage its fans even more. The chatbot has a lot of features. One of the most features is how it lets you converse with the character of Star-lord. It even lets you chat with the boy next door, Spiderman. However, it isn’t a freestyle bot. Instead, it has a few limitations. Many users even got replies stating that their responses weren’t valid. What makes it worth an experience, is the colorful artwork, resembling the marvel comics, while you chat.
Unlike most other applications of chatbots, mentioned earlier, UNICEF’s purpose isn’t recreational. It intends to raise its voice for people living in the underdeveloped and developing nations of the world and throw light on their issues. It works to eradicate the lack of essential services that they are suffering from.
U-report is a bot that collects data via surveys and polls. It conducts polls on different social issues, regularly and the U-reporters answer these polls or update data as per their knowledge.
Recently, a poll was conducted to know if Liberian teachers were asking for sex, in return for good grades, from students. As many as 84 % of reporters confirmed this news and the Liberian education minister was asked to take strict steps against them.
Med What :
If you’re one of those people who tend to buzz around google, getting misdiagnosed by the plenty of false knowledge available on the internet, Med What is a rather safe option. It intends to provide a faster medical diagnosis to the patients who resort to it.
Med What is assisted by a machine learning tool that helps it go through the research papers and scientific papers, from which it derives its relevant knowledge. It is more of a virtual assistant than a chatbot. It is an application that provides you with better and more accurate results and diagnosis, based on the symptoms that the patient suffers from. The tools such as Natural Language Processing and Machine Learning add to its customer (patient) interaction and improve the quality of its responses to them.
Roof AI :
Marketing professionals often suffer the challenges of generating and assigning leads. Generating the right lead, at the right time, is quite a task. Enter Roof AI is a conversational bot that eases the work of digital marketers by automating their interaction with potential leads, and all this is done with the help of social media.
With the help of Facebook, the conversational bot identifies the potential leads, and then it provides the user/marketer with its responses, in a friendly and almost human way.
Before assigning it leads to the agent, it prompts it encourages more information out of the potential leads. Amongst the stack of benefits that it provides the users, the best feature of the roof AI is that it allows the estate agents to respond quickly to their clients or the users.
It’s no more a beautiful morning and a cup of coffee served with the newspaper. The way our world has developed a network or the ‘web’, in the past few decades, has open so many doors for knowledge. It is a boon, but with certain drawbacks. If you step forth to look out for news updates, you’re sure to get entangled in the unnecessary ones.
To help you sort out stuff, NBC launched their chatbot in the year 2016. It operates with the help of Facebook and customizes your search. So whenever you need an update, it provides you with the headlines of breaking news, that might interest you.
It works like an assistant for you and of course, helps you save a lot of time.
One of Britain’s leading companies, PG tips advertised itself on television, starring the famous comedian Johnny Vegas and of course, the monkey. The monkey became so popular, the parent company of PG tips, Unilever used this primate character, incorporating it with a chatbot. This extremely interactive chatbot was created not just to advertise the tea brand, but also to raise funds for ‘Red Nose Day.
The bot was developed by the famous Ubisend company, which is known for the chatbots it creates for different brands. Users even claimed that the character of the monkey was not as adorable in the advertisement, but the bot is very efficient. It not only responds well to the users but also sends them jokes daily. The bot has quite evidently given a rise to the company’s customer base and is bringing in a lot of donations.
Alice was amongst the initial bots that were launched in the market and proved their potential. Even though it was launched about 20 years ago, its efficiency is something everyone admires.
ALICE is the abbreviation for, Artificial Linguistic Internet Computer Entity.
Comparing it with perfection would be like limiting the potential of a bot that works on a 20 years old codebase, yet performs impressively.
Similar to other bots that have been developed in the present decade, ALICE too comes across questions, to which it responds better than most other bots. Great progress has been observed in the responses of ALICE, in the past two decades, which makes it stand out.