Published on | Prototyping: From UX to Front End — Medium Carlos Yllobre

También disponible en español aquí 🇪🇸

I decided to write this article to share what I have learned (so far) while building my own bot, and testing many others. The points I am going to touch upon here, are coming from 2 types of approaches, both with the purpose of making the bot-human interaction as understandable and painless as possible: Context & Social Features. But first there is an essential question we need to ask:

Should bots imitate human behavior?

There has been a lot of discussion about this lately, with very interesting arguments coming from both sides. In order to answer this question, we need to define and understand what a bot is. According to The Tech Terms Computer Dictionary:

A bot (short for “robot”) is an automated program that runs over the Internet. Some bots run automatically, while others only execute commands when they receive specific input.

So a bot performs automated tasks, responding to specific triggers and lines of commands, with no other capability that following a script. The complexity of the bot will be related to the number of tasks it can process and how complex and variable are the responses to each one of these tasks.

From a purely technical point of view, the answer is clear: A bot shouldn’t imitate the human behavior, simply because no matter how complex the script is, a bot will never be capable to reach a minimum of the human behavior complexity. Following this logic, it makes sense not to add too many human qualities to a bot, based on its limitations. But, there are 2 problems with the application of this theory:

1. We are still humans and we don’t know yet how to interact in a non-human-human conversation. The only precedents we have are human-human interaction and it is logic, even if we try not to, that we end up bringing certain social behavior and emotional components to a bot-human talk.

2. We tend to underestimate the complexity of machine learning. An algorithm to identify and respond to human emotions can improve dramatically the overall experience while performing simple tasks, like answering questions, bringing support or identifying issues even before the user asks for help.

Last April, Minlie Huang, a computer scientist at Tsinghua University, Beijing, and his team built a chatbot with an “emotion classifying” algorithm that learned to detect emotion from 23,000 social media posts. The paper, that you can find here showed that 61% of humans who tested the bot, had a preference for the emotional version of the chatbot than for the neutral version.

Having said that, I believe that there should be always a certain grade of a social component in any human-bot interaction. Removing or paying too little attention to very basic behaviors, like, for example, knowing when the user ends up a conversation with a “that’s it, thank you” is not only a bad practice, but also confusing and overall, a bad experience.

❌ Microsoft Teams T-Bot

How can we make a better human-bot interaction Today?

Context:

Start and end

Bots should be able to identify and respond accordingly to the ways humans start and end a conversation. Some people might be more polite and end up the chat with a “Thank you for your help” and some others might end up with a simple “ok, bye’. Keeping certain social parameters is key to the way chatbots respond to the diversity of human personalities.

Continuity

Bots should keep track of the questions already answered and the information already collected about users and issues. In a basic level, it should avoid repeating the same answer to two different ways of asking a question, if the question was already answered and the user is not satisfied, the chatbot should provide more information about that issue, instead of providing the same answer again. On a more complex level, keeping the continuity across different chat sessions, collecting the data about the user, and personalizing the welcome and farewell message as well as the answers will be an important improvement of the experience.

Spring Bot

Multiple requests

Bots should be able to answer effectively to multiple requests/questions, identifying and ordering these in a way that one answer leads to another, if this is not possible due to the diverse nature of the requests, then the chatbot should respond to each one of them following the order presented by the user.

Change

Bots should be able to identify and respond accordingly to the ways humans jump from one subject to another. Question B is not necessarily in the same line of Question A. A complex script with a strong Data integration should do the trick.

Social Features:

Keywords and Behavior

While we wait for a more optimal algorithm to better recognize emotions, bots should be able to identify some behavior patterns in the user’s answers that can indicate their mood, and modulate the tone of voice and possible answers as well as trigger feedback requests.

Emojis / Image / Gif Recognition

Emojis, stickers, pictures and animated gif are part of our communication these days, if we want chatbots to communicate successfully, we need to integrate algorithms capable of reading and understanding these elements in order to give more satisfying answers and understand better how users communicate.

Explain Limitations

There are still many limitations when interacting with bots, they are not humans and they shouldn’t behave like humans, but they need to interact in a way that satisfies the complexity of a human communication.Bots should be prepared to recognize a communication issue and bring a proper answer in order to overcome that issue. If a bot is not capable of understanding a question or a request, then it should give users an ‘honest reminder’ of its limitations and provide with possible solutions.

PingUp Bot

Ask for Feedback

Identifying possible stress and tension points in the conversation flow is critical to trigger a feedback request, this should never interrupt the user’s task, unless that task is specifically related to a clear complaint. Asking for feedback is a simple way to respond to a bad experience and one of the most important procedures to collect data in order to understand what went wrong and consequently, improve the user experience.

Misbehavior

The bot is not a human, but it might still be part of a brand and a company, so it has to reflect the same values. There are certain boundaries in every human interaction that we should apply to Human-bot interaction but remove the emotional component. The answer a chatbot should give to a racist, sexist or disrespectful behavior, should be completely neutral and informative, not judgmental. The action is to interrupt that conversation and explain that is due to a violation of the terms and conditions related to the use of the platform. A link to the terms and conditions should be provided, and in some cases, an invitation to retake the conversation following the conduct code.

Humor

Humor is an essential part and an added value of every communication process. Adding some simple funny lines to a conversation always related to the tone of voice, nature of the task and user’s response will enrich the overall experience and contribute to delivering a more fluid interaction. In order to use something so complex as humor, the chatbot should be able to identify a combination of a simple and positive response, the user’s problem/request and the variable of time to deliver that answer. This answer could be a simple line, an image or the use of emojis to increase the meaning of the message, but this response has to be triggered by the user first.

Google Assistant

Conclusion

I like to see bots as the preliminary step or the Experimental Lab to a more complex/rich (maybe dangerous?) AI with more generic and maybe chaotic purposes. This is what makes it so important in my opinion, not only as an early experiment but also as an excellent and more accessible introduction to what is yet to come. An early definition and improvement of certain parameters in the behavior of Today’s bots will pave the way for the future Human-bot interaction.

Thank you for reading! 😊

También disponible en español aquí 🇪🇸

Originally published at charlieyllobre.com on September 3, 2017.

Building a better Human-Bot interaction was originally published in prototypr on Medium, where people are continuing the conversation by highlighting and responding to this story.

Building a better Human-Bot interaction

Cool stuff, every week

More Articles