TWIFT | Digital | How to Break a Chatbot

How to Break a Chatbot

Ever thought of how to break a chatbot? Chatbots were made for short talks and quick answers. They know how to maintain a conversation, chat with people on various topics or teach them while using very natural communication and language.

When you have dialog boxes with real people who can consult a client at any time of the day and similar chats with bots, you cannot immediately know if you are talking to a person or a chatbot.

Nevertheless, computers and bots still cannot process and recognize certain information as effectively as real people can, and, accordingly, cannot give a clear answer to it. Therefore, there are some tricks to “break” the chatbot and tell them from real people.

This article will tell the secrets that will help you understand whether you are talking to a real person or a bot, and how to trick a chatbot.

At the same time, we will learn some important lessons and get closer to understanding conversational UX and find out how users are processed using a simple conversation. 

Remember that the article is not about hacking chatbots. It is illegal and we advise you not to try it.

Can Chatbots Pretend to Be People?

First thing first, let’s point out that bots are not supposed to pretend to be people.

If artificial intelligence tries to convince you that it is a real person, and it will turn out to be a lie, users will feel uncomfortable after they find out the truth. Nobody likes being deceived.

It is better to think of the chatbot as a tool that helps others and gives them access to important and interesting information.

How to break a chatbot

Top 6 Questions That Will Break a Chatbot

The desire to break something that others have built is very inherent to people. That is why from the moment such technologies were created, we have been trying our best to find a way to break them. Does the Turing test ring familiar to you? This is a series of tasks that are aimed at revealing the artificial nature of the interlocutor you are talking to.

Instead of the usual list of questions, we decided to analyze the approaches that can be used to find out if your interlocutor is a bot.

Truth

Are you a chatbot?

This is perhaps the most obvious question that you can ask. Despite that, it really works, since a good bot has settings that make it impossible to lie. That’s why a real person’s answer will probably be “no”, but a chatbot is likely to tell the truth.

There is nothing bad or special about having a chatbot, that’s why this fact is rarely hidden. BUT, if the bot tries to hide the truth from you, you may want to carefully consider the security of your personal data and in no case share it.

Note: if you are an entrepreneur, be honest with your customers and do not hide the nature of chatbots. 

Talk about Emotions

How are you feeling?

People will try to describe their emotional state, mention some real-life situations, tell stories, and so forth. The answer that a real person will give you is absolutely impossible to predict since each individual has their own story.

Chatbots are programmed for a certain series of standard answers, and such questions for them are another thing that needs to be answered briefly. Perhaps an answer will even be repeated word for word several times.

We have already talked about the honesty of robots, that’s why it is likely that the answer will be along the lines of “I am a bot and I do not experience emotions.” If the main task is to maintain a conversation, it can answer with something short, like “The weather today is just fine/terrible.” 

Note: the chatbot can be programmed in a certain way for fun, so you can add a comic answer to such a question. Make your clients smile.

Best Phrasing

What is that?

Any chatbot is initially programmed to give answers. Often the answers to the same question are almost the same, the variability of answers depends on the scale of the software. If the answer that the bot gave you is not good enough, ask again. Most likely, it will not be able to change it for something more intelligible, the meaning and wording will remain almost identical.

On the other hand, real people will try to use different wording, relay the sense in other words, or admit that they made a mistake.

Note: if you find that many users do not like the phrasing of some answer and they ask you to explain, rephrase or change it, consider which answer they get. Are you sure that the information you provide is correct and reliable?

Ask It to Tell a Story

Have you been working here for a long time? 

This partly overlaps with the emotional state answers. Ask the interlocutor about something that could provide an endless number of answers and go in any direction. A real person may tell you a funny incident related to an interview, share some personal experience and how it helped them get there.

If you talk to a chatbot, it is, of course, unlikely to tell you any fascinating personal story.

Note: it is better for the chatbot to reject such questions since it is absolutely impossible to predict the wording of these questions and what they will be about. Programming a robot to answer any of your personal questions simply will not work, so do everything possible to reject it and transfer the conversation to another topic.

Joke

Why did cats learn to see in the dark?

For chatbots, humor is a whole other, incomprehensible, universe. Well, for some people too. However, some kind of joke, even a simple one like “Why did cats learn to see in the dark? Because they couldn’t reach the switch” will help you to understand if your partner on the other side is a bot.

Artificial intelligence can never understand a joke in the way people do, that’s why this method is a perfect example of how to break a chatbot. If the bot is not programmed to answer a specific question, it will not be able to do so.

A real person may not even understand that you are checking them and give you a clear and logical answer. They are more likely to think that it is a slightly strange beginning of a conversation.

Note: humor is an important part of any conversation, even with artificial intelligence. Be sure that if the interlocutor makes a joke, the chatbot will be able to reject the joke and continue the conversation.

Ask Something Weird

Why does a lily look like a pen?

Such questions will completely confuse any bot. A real person is likely to take this as a joke and will try to look for funny explanations.

A bot will try to respond without perceiving the context of the phrase. Thus, you will immediately understand that you are dealing with a bot.

Note: just like all the previous questions, try to reject them and focus the conversation on something else. Move the conversation to the aspect where the robot can really help the user.

What if Someone Breaks my Chatbot?

Do not forget that your chatbot should first of all help users in finding answers, maintain a conversation, and not make them think that they are communicating with a real person. 

Breaking is not always a bad thing, as it can show users that you are honest with them and increase loyalty.

Comment it, btch   Comment it, btch   Comment it, btch   Comment it, btch   Comment it, btch   Comment it, btch   Comment it, btch   Comment it, btch

Write your fu**ing article right now!