Aytishnik married the girl with whom ChatGPT was talking. And how to teach a robot to communicate with customers?

Aytishnik married the girl with whom ChatGPT was talking. And how to teach a robot to communicate with customers?

Hello, Habre! My name is Oleksiy, I develop mobile applications on Flutter at Pirobyte. Recently, I went to Twitter and did not know whether to be happy or cry, because the epic about itishniks and cringe threads got a new round of development.

In short, the guy talked about the fact that he married a girl with whom ChatGPT was not talking to him. And it wasn’t just a few messages, but a streamed scheme where a chatbot searched for, chatted with, and invited thousands of Tinder girls. So far, I have not found the only one to which the bot was reconfigured. And then, thank God, he started communicating on his own. He made money from other girls through referrals (referring them to different companies). What a story, Mark!

Although the story is interesting, in general it is not clear why to write about it here… Not really? Let’s understand each other.

To begin with, let’s analyze the author’s experience in creating ideal Tinder guy from ChatGPT. In order not to repeat the whole thread in its entirety, I will describe the main points:

  1. The initial scheme was based on a request: you are a guy, you meet a girl, your goal is to invite her on a date. For this, the bot tried to get to know the girl better, but he did it clumsily and constantly called for a date in the forest (literally, in Bittsevsky)

  2. After the update, the bot has a memory, or more precisely, a context. And there was an opportunity to teach him through messages, which our hero took advantage of

  3. Using FlutterFlow (finally Flutter!) and API, ChatGPT independently swiped profiles and chatted with girls. All based on the author’s preferences

  4. When it came to going on a date, the bot made a short squeeze of the dialogue and sent it to its owner, who, in turn, went on a date himself (well, at least something)

  5. The advanced version recorded all appointments in a Google calendar so that there were no overlaps, and consulted with the creator if she was not sure of the answer

And here we exhale and pause.

What was it anyway?

In fact, while the author of the thread was describing his experience of finding the perfect girl with the help of a neural network, he unknowingly told us about one interesting profession today – dialogue designer. And no, what follows will not be advertising, but a small introduction to a relatively new and interesting profession.

It’s like a joke here – how many designers does it take to screw in a light bulb? Taking into account as much as the business needs.

And what does a dialog designer do? Approximately the same as our hero (but without the use of code), that is, he thinks about the purpose of the robot, conversational flow (the logic of the conversation) and improves the quality of his communication with the end user.

The path to creating the perfect chatbot can be very complicated due to 2 things:

And if you read my previous article on ChatGPT, then you know that several things are important for artificial intelligence: understanding the essence of the conversation, context, variables, and clear intentions (tasks). Whereas natural consciousness requires empathy, support, and other things that are not available to a cold machine.

Okay, the base is sorted out.

Let’s try to design our robot?

Just as a concept for now. Where do we start?

Of course, from the purpose and sphere of responsibility of our work. And let’s take something simple. Let’s say that we want a chatbot that would bring new leads to the studio.

We automate this process!

We need to choose a platform where we will develop our bot (so far it’s just theoretical, but I’ll try to explain why it is). From the simplest and most obvious – Landbot. She is popular and has a clear bot designer, where you can make both ordinary fools and AI monsters (if you are interested in something cooler, you can turn to VoiceFlow, there is plenty to do).

To decide on the next step, we will turn to the interface of this platform:

And here we see the “feed the bot” button. The key word for us is “feed”.

A reasonable question arises – what to feed him? Of course, by talking. In order for the bot to delight customers with useful information, and not ask about the weather, it needs to be created for it Conversational flow.

Conversational flow is built from several things that will be clear to any person who has at least once talked to people (a head nod is considered communication, honestly):

Our Flow is built mainly on if/else blocks. If a certain question was asked, this answer should be given. If it is different, then the answer, accordingly, is different.

If it is very simplified, then it turns out like this:

In addition to the obviously unbranched structure of this Conversational flow, it is simple for several reasons:

  1. This Flow is suitable for a robot that knows how to answer only specific questions that we prescribe to it. That is, choose a question from the available list;

  2. In this structure, basic “anyway” fallbacks are written, which do not explain to the bot what to do if the user clicked “other”. He simply ends the conversation;

  3. The scheme does not collect information from the user: his number, address or social. network, and this is important to get a contact from the client for further communication with him.

And then what is the point of this scheme? Thanks to her, we understood their main components:

  1. Everything always starts with an entry point, which prompts the robot to the first stage – greeting;

  2. From each answer option there is a branch and an answer option for the robot;

  3. Each variant of the question is written as an intent (literally intention), which is perceived as a question of the user.

Let’s analyze another Flow with a minimal amount of filling, but with the main stages

And yes, Conversational flow is not necessarily a scheme, it can be just an example of dialogue.

Goal: generation of leads by identifying the potential needs of the client. And the next generation of ideas taking into account the available services of the studio. The bot should work when no one has answered the client, that is, everything is offline for one reason or another. Plus, he should be able to answer questions in a free form.

1. Greetings:

  • Bot: Hello! I work in a web studio. While no one else is there, I will be able to answer your question

  • Bot: I can tell you more about us or help you decide on the development of a specific product. What do you say?

2. Qualification (clarification of the client’s wishes):

  • User: I would like to make an application

  • Bot: Sounds cool! Tell us more about your business. What tasks do you want to solve with the help of the program?

3. Gathering Information:

  • User: I have a small online jewelry store

  • Bot: Ok, do you already have a website or app?

4. Identifying Pain Points (find out the problems of the existing business):

  • User: We have a website, although it performs the main function, but it is morally outdated, we would like something more modern

  • Bot: Got it. Modern design and functionality really attract more users. Tell us more about why you are not happy with the way things are now

5. Offering Solutions (we offer solutions):

  • Bot: Your business would benefit from the app as most of your orders come from there.

  • Bot: Would you like to call to discuss further strategy?

6. Call to Action:

  • User: Yes, of course

  • Bot: Great! Write your email so that we can schedule a meeting at a time convenient for you

7. Follow-up (completion of dialogue):

  • Bot: Thank you for your interest in us! We will contact you shortly

  • Bot: In the meantime, you can explore our site to get acquainted with the best cases and reviews of other companies

8. Additional Assistance:

  • Bot: Can I help in any way?

  • User: No, that’s all. thank you

  • Bot: Please! It was nice to meet you, if you have any questions – I’m happy to answer them!

Although this scheme is simplified, it gives an idea of ​​the main points of the dialogue. Of course, you can handle all the borderline cases and paint the bot in as much detail as possible, but you shouldn’t do it. Follow the 80/20 rule. After all, 80% of conversations will take place in 20% of all possible branches.

The last step consists in integrating the work with the company’s knowledge base. This is a bit beyond the scope of dialog design. However, it’s worth going over briefly, because your conversational flow should be based on the bot’s real knowledge. In tabular form, the database looks something like this:

Case studies

Case 1

Case description, strengths

Case 2

Case description, strengths

“Pains” of business

Pain 1: Outdated design

The design of the existing solution does not meet the requirements of the market

Pain 2: <назва>

Typical problem description

Possible solutions

Update the design

A complete review of the design concept, taking into account the current trends in the market

Develop an application

Description of the solution

Our stack

Frontend

React, React Native, Flutter

A description of the strengths of development on a particular technology, combined with the best cases

Backend

Laravel, Node.js

A description of the strengths of development on a particular technology, combined with the best cases

The knowledge base should cover everything that may be needed for the ultimate goal of lead generation. And if your conversational flow is written as correctly as possible, then the knowledge base will have only useful information for the user.

Okay, we have figured out the main steps.

How to make dialogue so that it is not annoying?

And here there are difficulties, because the design of dialogues involves a lot of practice and knowledge of psychology.

But we can highlight several basic principles without which it will not be possible to build a high-quality dialogue. They are based on the main feature – naturalness. Real, human consciousness requires empathy, respect, and encouragement. I borrowed the principles themselves from Conversation design institute:

  1. A thoughtful personality of the bot is the key to consistent dialogue

    Personality matters. One of the best strategies in dialog design is to make sure your AI assistant is consistent in its conversation. This will make him more attractive and trustworthy.

  2. Use your bot’s personality to build relationships with your users

    Now that your bot has a “personality” that sets up a coherent dialogue, you can start fleshing out that personality. There is a misconception that the personality of the bot is all chatter. That it in no way affects the result. However, this is fundamentally wrong.

  3. Correctly prescribe the dialogue framework and limitations of your robot:

    Hi my name is Johnny and I know everything. What do you want to know?

    You invite your users to ask any question they can think of, and there’s a good chance your assistant doesn’t know the answer. But how to do it right? Specify the scope of your bot’s knowledge:

    Hello, my name is Johnny. I can tell you anything about coffee. What would you like to know?

    By using the word “coffee”, we increase the likelihood that our user will ask us a question about coffee. Assuming the bot is properly trained to answer questions about coffee, we can increase the chance that the AI ​​assistant will understand the user and provide useful information.

  4. Clearly express concern for the customer to gain trust

    We often talk about empathy in design. This is really important if you want users to interact with you. But what does it really consist of? Many believe that empathy is related to the manifestation of feelings. It does not. It’s about showing that you understand the user’s situation. By demonstrating that you really understand what is going on with the user, you gain trust and increase the completion rate of your dialogues.

  5. To speed up the dialogue, use question/answer

    A question signals to the user that it is his turn to speak. Why is this important? Because the concept of step-by-step dialogue encourages the user to answer questions in detail. And also with pleasure, because the obvious interest first makes him believe that you are guiding him to the right decision.

To summarize, a good bot should cosplay a pleasant interlocutor who understands you well. Or pretends to understand. But at the same time, he does not talk too much and gives only the necessary information, constantly maintaining a dialogue and not being distracted by details.

If you decide to create a chatbot for your company, think carefully about not only the logic of outputting messages, but also the empathy of your assistant. Do not try to cover all options of questions, concentrate on the main options. And do not forget to warn the user that he is communicating with a bot and not a real person, so you will avoid many ethical problems.

***

Who thinks what about the new profession? Is it interesting enough to turn it into a hobby and then a real job?

Related posts