Building the Perfect Chatbot

Iren, in this contribution, focuses on the perfect chatbot. If not familiar with the term, a chatbot is software capable of conducting conversations that pick up on auditory and textual clues from the human voice, or typed text, to engage in a conversation.

The ultimate chatbot is one that passes the Turing Test, developed by Britain’s famed Alan Turing, the father of modern computing and AI. Turing’ designed his test to determine if a human engaged in a conversation with a machine intelligence would find it indistinguishable from talking to a person. The question “are we there yet?” has still to be answered satisfactorily in 2019, but we are getting closer.

I hope you enjoy the read and if you want to learn more about chatbots, please visit the BOTXO platform where you can find out about their chatbot builder.

Do you think about how many times you have chatted with a bot on a website to get help when shopping or to answer your customer support request? Do you even know if the entity answering your call is human or an AI? Sometimes it is obvious particularly when you find yourself frustrated by the lack of comprehension at the other end of the connection. This frustration is proving to be happening less these days as automated customer interactions continue to evolve. Gartner, the high technology market research firm states that, by 2020, 85% of customer interactions will involve chatbots. That’s because humanizing chatbots is now well within the realm of the norm when running a business online or dependent on over the telephone interactions. That’s why Maruti Techlabs proposes that we move beyond merely questioning the need for conversational AI, and instead, develop the intelligence quotient of chatbots to bring the human touch to interactions as learned from their operators and users.

Four Different Types of Chatbots Divided into Two Groups

There are four types in which bots can be divided into stateless, semi-stateful, stateful, and loyal. These four come in two groups. The first two fit in the category of scripted bots. The last two are conversational. Even though both types can address an issue raised by a user in conversation, it is the way they get there that is significant.

In a scripted or sequential bot, the questions anticipated by users are pre-scripted by topic. The bot works well as long as the caller sticks to the script and the questions match the answers. When an unexpected scenario occurs and a caller feeds the bot unexpected words and phrases, the conversation fails because the bot cannot deal with unknown and unanticipated inputs. Here is one example of two different outcomes.

A. A user asks a sequential bot to “please find cheap flights.” The bot understands this keyword phrase since its developer has scripted this into its program. The bot then follows the script with additional questions like “to which destination?”, “for what dates?” and “for how many people?”

But what happens if a user adds lots of extra data like “dear bot, get my friend and me out of here for a long weekend to the city of love, max for $700 per person,” the chances are that scripted bot won’t understand the request since the user has input more information than the scripted program can decipher.

Scripted or sequential bots are designed to lead conversations with a user and persist in getting answers to pre-defined questions that in turn sequence to follow-up questions. There is little wiggle room in such apps. Hacker Noon’s guide to chatbots describes scripted bots as fairly dumb and not designed to learn anything from users which leads many times to very frustrating conversations.

On the other hand, there is the conversational AI bot, a smart bot that engages a conversation more effectively because it is built to understand complexity. Conversational bots can remember users and previous conversations with them.

Maruti describes other advantages including an ability to perform more requests and continuously learn, and remember every interaction. A conversational bot is not built to define the path of a conversation but rather works best when receiving more input from users. Providing lots of data then allows the algorithms upon which the bot is built to calculate how close a user’s inputs are to his or her intent. This means users don’t have to input exact, pre-scripted sentences to be understood. This result is a conversation that is more natural and personal.

So How Do You Get to the Perfect Smart Chatbot?

Here are 10 suggestions:

1. A Bot Needs to Become a Good Learner to Know Its Users

Smart bots can keep some knowledge about users in terms of who they are and other facts of identity. A chatbot needs to be constantly learning to recognize patterns in data it receives and whether this happens or not is up to we humans. Machine learning algorithms built into the intelligent bot allow operators behind it to make sense of the streams of data coming from users. Bot operators, therefore, need to be good at looking at and learning and giving the bot data to work on to perform better in interacting with users.

2. A Bot Should Become Aware of Users’ Needs

Users can have at times complex needs, so the technology behind chatbots needs to be complex as well. As in our above example, a human user may ask a chatbot to fly him or her somewhere specific at a particular time while having limitations on the travel budget or the number of accompanying companions. So how does a smart bot make sense of it all? 

Smart bots have the human capability of gaining information efficiently, approximating their human operators. Efficiently refers to the bot receiving assistance from operators in understanding the intention of every part of a request and then making connections between the different parts to respond most appropriately. Unlike humans receiving a complex set of dependencies in a single request, smart bots consider how users interact but unlike humans, they don’t forget and they don’t get stressed. Smart bots can even be smart enough to gauge a user’s stress levels and respond accordingly.

3. A Bot Should Have the Ability to Sense the Environment

Users’ needs are connected to their environment or to context. Understanding this is critical to smart bot interactions. Before a bot can perform a particular task, it needs to integrate with the physical and language environment of the conversation. It is not only the conversational data that matters but rather the importance of understanding the kind of requests and the intentions specific to the environment which triggers a user’s request. BotXO builds chatbots that do this serving a user best by understanding requests in time, by channel, and through behavioral observation.

4. A Bot Should Be Sharp to Think

A bot whose operator thinks sharp does not just understand the environment the user resides in but goes a step further. The smart bot makes decisions based on how it interprets acquired knowledge. According to Maruti, these decisions are made by leveraging pre-existing knowledge from user interaction and new knowledge continuously conveyed to the bot. Using neural networks, a form of machine learning, the bot thinks sharp and gets sharper over time with every user interaction.

5. A Bot Should be Quick to Act

Progress towards a defined goal in the interaction is only reachable once a smart bot has gone through a sense-think-act cycle. Environment-sensitive decision making makes it possible to keep a user engaged in a conversation, and gives the bot the means to respond to new data inputs.

6. A Bot Should Remain Open to a User’s Change of Mind

Smart bots allow the dialogue to jump between contexts with users able to navigate without a defined path. Without scripted decision-making, the smart bot remains open to new inputs or additional parameters even if it means replacing or adding to existing information it already knows. In real life, people change their minds at any time. A smart bot respects this human attribute giving users the feeling of greater independence and freedom during their bot interactions.

7. A Bot Should Formulate Coherent and Convincing Responses

The smart bot is built to converse naturally. It does this using Natural Language Understanding (NLU), a technology that empowers conversations the bot to write in the user’s natural language during a chat. At Bot-X-O, this multilingual capability includes the ability to converse in English, Danish, Swedish, German, Spanish, and more.

The BotXO algorithm checks for misspellings, recognizes individual names, company names, places, and the rules of grammar. With NLU a smart bot can understand the intent of every sentence typed or spoken. This capability gives the bot the means to reach users more effectively through interactions that are akin to communication between two humans.

8. A Bot Should Understand the Mood of the User

What makes smart bots extra cool and intelligent is their ability to understand the user’s mood. They do this through sentiment analysis, a part of NLU. The bot can sense whether a user perceives a subject in the conversation as being more or less urgent. Sentiment analysis allows the measuring of how users feel about interacting with a bot. According to Jens Dahl Møllerhøj, Lead Data Scientist at Bot-X-O, “being able to measure how happy customers are with the bot means we know how to change the bot to move into the right direction.”

9. A Bot Should Talk Casually to Help Strengthen its “Personality”

Most humans in a conversation rarely limit it to achieving one response or one goal. Humans are social. That means a smart bot needs to integrate social talk into its conversation. Enriching a chatbot with a “personality,” as close to a person as it can be, makes conversations between users and bots better.

10. A Bot Should Act as a Helper Not a Collector

The biggest difference between dumb and smart chatbots is that the first act as collectors, while the second act as helpers. A sequential or scripted bot sticks to pre-defined questions with the expectation of specific answers. It cannot determine intent. Tech professionals from Maruti weigh the quality of a bot’s smartness based on its helpfulness. A smart bot lets users lead in the conversation learning more about that person. The more data the smart bot receives, therefore, the better the results.