The Chatbot Hype

15 Nov, 2018

“How can I help you?” is a phrase you see jumping out the right corner of your screen, visiting any website. It’s 2018, which means, that you no longer have to look for an email or call support if you need an advice or extra info – you can simply message a chatbot. Whether you’re an active user or ignore the buzz around chatbots, they are indeed revolutionizing many industries – banking, insurance, travel, real estate, healthcare, to name a few. For many people, chatbots have become a real passion, and, most importantly, a favorite job.

Botscrew has been developing custom AI chatbots in Lviv since 2016, and people who work here cannot be more excited about what they do. Why do we even need chatbots? How are they developed, communicate, and how will the technology change in the future? Let’s dive into the chatbot world!

We’ve started with chatbots quite spontaneously, while in a car to Berlin. We were driving from Ukraine to Germany to participate in THack Berlin 2016, a travel hackathon. On the spar of the moment, we decided to make a travel chatbot with no idea how to actually do it. During three days of the hackathon, we managed to develop our first chatbot – a travel assistant that helps to organize personal trips. While working on the chatbot, we understood that this technology is likely to become very popular in the future as it offers a whole range of benefits. Chatbots don’t need any additional interfaces to be developed, you can simply use your favorite messengers. It’s really cool, since you don’t need to download a new app and get used to a completely new interface. By the way, we finished second at THack Berlin, and since then decided to start developing chatbots seriously.

bots4

We met our first client – an Italian ticketing platform and travel company Musement at another hackathon in Mallorca. For them, we developed a Facebook Messenger chatbot that helps to plan trips, look for best travel options, buy tickets, etc. We also added a few extra functionalities – searching for attractions and events in selected destinations and a possibility to buy tickets for them too. All three of us quit our jobs at local IT companies and moved to Bali for a few months to concentrate on chatbots, learn more about the technology, and surf.

How do chatbots work?

The end user of any chatbot is a human who communicates with a bot in order to get some information or solve problems. Chatbots are amazing solutions for young people, who definitely prefer texting to calling. If there are some problems with your Internet provider, mobile operator, or bank account, simply asking a bot is much easier, than calling customer service, waiting in line, and trying to explain what happened to a helpline representative.

For example, when you want to plan your next trip, you can simply add a travel assistant on Facebook and message it saying “I want to go to Switzerland”. Super easy. This question goes directly to a service based on Natural Language Processing (NLP) and Machine Learning. The service identifies the sentence with two indicators – “intent” and “entity”. “Intent” means the main goal of the sentence, here it’s the verb “to go”, while “entity” recognizes other important parameters, like Switzerland in this sentence. After recognizing these two indicators, the service understands that there is a lack of information and starts to generate more questions to clarify all details – When do you want to travel? How (plane, train, bus)? How much money would you like to spend on the tickets? The chatbot asks questions until it gets all the information needed to book tickets for your trip. It works very fast – within mere seconds the system understands your message and writes back.

Chatbots are much faster than human agents in a call center. According to 2018 Customer Service Benchmark Report, chatbots answer in a second while an average response time of any client support service is 12 hours and 10 minutes. A huge difference!

You don’t have to build a chatbot from scratch. There are many datasets you can buy to train a system – datasets from Google, IBM, Facebook, etc. We have even created a new position at our company – chatbots trainer. It’s a person responsible for choosing the right datasets for our projects.

Goodbye, routine work

The most successful case we have worked on is the Law firm without lawyers. We developed a chatbot Ailira to generate typical template documents for FB Messenger and web page of one Australian law firm. Approximately, 20% of lawyers are doing rather routine jobs filling out standard documents, so why not make their lives easier? Ailira starts with a question – How can I help you? At the same time, there are a few buttons in the messenger such as “create a will”, “create other documents”, “talk to a lawyer”, etc. Therefore, you can start to create your document immediately by filling out a form and choosing the right descriptions. It’s a rather boring job for a human, but very easy and fast one for a bot. If more complicated questions occur, you can always contact a real lawyer.

bots5

Ailira made a real buzz in the media. As a part of the marketing campaign for this law firm, we built an office with computers, but without people in a shopping mall. Everyone could come and ask the chatbot a question, instantly get an answer and create a will, for example. Ailira was actively discussed in many Australian media afterwards, moreover, the case was even studied at The University of Sydney.

Chatbots come in handy in almost every business since routine tasks, as well as complicated searches and technical problems, exist everywhere. The idea of chatbots is to make human-computer interaction easier. Chatbots simplify the user’s choice on every step, the only thing you have to do is to answer the questions.

User customer support is a whole industry focused on providing answers to the most common questions. Employees learn typical answers, spend some time looking for right instructions, etc. Vice versa, chatbots are trained just once, and can communicate with many people simultaneously. And, what is more important, way faster.

We’re convinced that the humans deserve more interesting, intellectual job, where they could apply their creativity. Why spend so much time and human resources on the job that bots, robots, and computers can do? Our ultimate goal is to get rid of routine tasks and give people a chance to focus on something that really matters.

Conversational design

Every chatbot is a character with its own name, age, and conversation style. Ailira, the lawyer chatbot, is a 40-year-old woman who speaks rather seriously, using typical legal language. The travel chatbot that we developed for a startup from California, recommends places to visit and things to do, and communicates in the appropriate way – using emoji, slang words, stickers, gifs, etc. Creating chatbot’s “personality” and the way it communicates is a very important part of the job. Usually, we discuss such details with our clients on the first stage of cooperation. It is essential that chatbot’s character complements the subject area.

Once, we’ve developed an eCommerce chatbot for online shopping, which can change the way it communicates based on a user’s character. It synchronizes with your social network accounts including Spotify and chooses how to speak with you, and what to offer you to buy online. For example, if someone is a big fan of Eminem, the chatbot will offer them to buy suitable clothes and accessories, communicating with you like a rap fan.

Sometimes it happens that a chatbot doesn’t understand something. Spelling correction is a very pressing problem today. The NLP industry is working hard to solve this problem and improve the machine’s language recognition skills. Humans can easily understand a context and build connections between words. If instead the word car someone types cat in a message, a chatbot will recognize the word cat as the right word, because the word by itself is correct. Unlike humans, who will definitely think the word doesn’t suit the context,  a chatbot won’t notice any mistake.

Maintaining the context is one of the biggest challenges that the NLP industry and, especially, chatbot developers are now facing. The more connections between words a chatbot can build, the smarter and more similar to human beings it becomes. The dream of all developers is to build the human-like chatbot what would be able to pass the famous Turing test. It’s a test that checks the machine’s ability to generate human-like responses through the natural language. As of today, no chatbots have passed this test yet. It’s a future challenge for both scientists and developers.

It might sound strange, but creating an ideal human-like chatbot isn’t our main goal. Ultimately, we want to help people, reduce the number of routine jobs, provide useful information (including recommendations and tips), and solve business problems. The technology of chatbots has become our own tool for doing this.  

Comments are closed.

Інші новини