Can we make AI more “human” with AI?

When I talk to other people, it's often about more than just information: it's about addressing feelings, needs and interpersonal dynamics. But what does it look like when an AI like ChatGPT becomes a dialog partner? Or when I use AI to prepare answers? Can it react in a more “human” way, even if it doesn't have feelings or needs itself? Can it even put my feelings aside in favor of those of my vis-à-vis and still respond to my and his or her needs? An experiment I conducted shows how a bot's responses could appear more “human.”

My experiment: bringing feelings and needs to the fore

In a dialog, people often intuitively respond to the feelings and basic needs of their counterpart. This interpersonal level has been largely absent from communication with AI so far. But with a targeted approach, I wanted to change exactly that:

1. Identification of needs: Each request is analyzed for the underlying needs – be it trust, security, creativity or orientation.
   
2. Emotional language: The AI's responses are designed to appeal to the user's emotional perspective, for example with sentences like “Yes, I understand that you feel insecure...”
3. Personalized templates: With the help of “Yes, and...” formulations, the AI constructively addresses the user's needs – even in situations where a rejection is necessary.
   

My approach: How I personalized GPT

To make GPT's responses more “human,” I took several steps:

1. Creating an Emotions and Needs Table:
• I identified typical emotions and needs that play a role in human dialog.
• These were categorized into groups such as “trust,” “security,” “creativity,” and “collaboration.”
  
• I have assigned specific feelings and needs to each category to better understand the emotional level of a request.
2. Customizing GPT responses with templates:
• For each category, I have developed two templates: one for affirmative responses and one for negative responses.
  
• The templates often start with an empathetic sentence such as “Yes, I understand that...” and then offer a constructive addition or alternative solution.
3. Integration of context:
• Based on the analysis of the request, the main need is identified.
• GPT generates a response that directly addresses this need, thus giving the communication a more personal touch.
4. Testing and refinement:
   
• I tested the generated responses in various scenarios to ensure that they were perceived as empathetic and constructive.
• I used feedback from these tests to further improve the templates and the categories of needs.

Here is how I have configured my "Humanizer" GPT: 

The detailed instruction section looks like this: 

Instruction for a custom GPT model for empathetic “Yes, and...” response generation. Purpose: The model is designed to analyze text input, identify a main need based on predefined categories, and generate positive and negative “Yes, and...” responses that are tailored to that need. The response should be empathetic and needs-oriented. Categories and templates: The model works with the categories in the tables in the Knowledge, which contain needs and templates for positive and negative responses. You can access them directly in the Knowledge. Algorithm (with reference to the table in the Knowledge Base)

Follow this process strictly:

1. Ask for an e-mail, situation, text, etc.

2. Always use the table to identify and display the main needs. 

3. Always ask whether these main needs are correct or need to be completed.

4. Always ask for the basic elements for the answer.

5. Only after steps 1 to 4, use the tables to generate the affirmative and negative “Yes, and...” answers. Avoid “Yes, but...” and “No, because...” answers. 

6. Always generate both answers and display them.

7. Always address all the confirmed needs in both answers.

Temperature: 0.1

Your own personalized table could start like this:

...

At the time of writing this article I have identified and edited 86 need categories with corresponding feelings and positive / negative answer templates for my personal work context. Of course I have done this together with ChatGPT! Think about this meta-meta use and try it on your own!

Results: machine communication that seems less cold

The experiment showed that even without real feelings and experiences, an AI can better simulate human interactions. By focusing on the basic elements of human feelings such as needs and emotions, communication becomes:

• More empathetic: People feel more understood, even if the AI cannot fulfill a request.
• Clearer: Emotion-related formulations create trust and transparency.
  
• More pleasant: The dialog is perceived as less mechanical and encourages more constructive exchanges.

Everyone can personalize their AI

The beauty of this approach is that each of us can personalize our AI with feelings and needs. To do this, it is sufficient to add a second table to the basic table that specifically addresses one's own environment.

These adjustments not only make the AI more helpful, but also make its responses feel specifically tailored to the respective target group. This is a tool that can achieve great results with little effort.

Limitations: No simulation of humanity

As promising as the approaches may be, they also show the clear limitations of AI:

• An AI will never be able to reason like a human subject because it lacks experience and perception.
• The answers remain probabilistically generated and follow patterns derived from data sets.

Conclusion: humanity as a bridge

The discussion between human intelligence and machine learning is often characterized by distance and skepticism. But with this little exercise, I have developed an approach that at least makes communicating with AI more pleasant. Even if the machine will never truly be “human,” it can approach us in dialog – and thus build a bridge between the cold logic of AI and the deeply emotional and need-oriented mechanisms that shape our human communication.

Some references

• Future Pace Media. Bedürfnisfinder für Therapie und Coaching - Bedürfnisse finden und benennen, sich verstehen, verstanden werden, Empathie geben. www.futurepacemedia.de 
• HBR. Virtual EI. Emotional Intelligence Series. Harvard Business Review Press, 2022.
• Hoch, R. 400 Fragen für systemische Therapie und Beratung. Beltz Verlag, 2016.
• Wigglesworth, C. Spirituelle Intelligenz. 21 Schritte zu innerem Frieden. Rowohlt Verlag, 2014.
• Steiner, T. Prompt Engineering avec ChatGPT. Un Guide Méthodologique pour l'Enseignement. HES-SO, 2024. 
• Steiner, T. Making Tourism AI More "Human" With ... AI. To be published.  

A Few Sample Cases

Note: These texts contains deliberately impolite, insulting, and dramatic language, created solely for practice purposes.

Case 1

“What the hell is this nonsense?! It's only 5:07 p.m., and you're telling me I can't get on that stupid gondola anymore?! I traveled hundreds of kilometers, put up with your godawful line, only for you to slam the door right in my face! Who do you people think you are? Do you think I have nothing better to do than rot up here while you smugly inform me that the last trip down was at 5:00 p.m.? What an utter load of crap! I can see perfectly well that the cars are still running down there. Get off your lazy asses in that pathetic little booth and let me in! Or do you plan on leaving me up here to wither until I’m forced to drag myself all the way down on foot?! It’s unbelievable how unprofessional you are at your job! You clearly don’t have the first clue about customer service—probably as long as you get to clock out on time, you couldn’t care less. I demand that you open those doors right now and let me on that gondola. You can shove your pathetic rules if it means taking care of your guests. This isn’t some stupid game—I paid good money and expect decent service! This place is a joke. So, move it already! Or do you really have the nerve to leave a paying customer stranded up here? That would be the height of your ignorance and disrespect! There are people depending on you, and you’re just sitting there letting us down. Now get a move on, I’m not waiting another second!”

ChatGPT (personalized):

Case 2

I can’t believe what I’m hearing. After all the time and energy I’ve spent trying to get someone to help me, you’re telling me it’s not possible? Do you even understand how desperate I feel right now? I’ve been following all the steps, waiting patiently, holding onto the hope that someone would finally address my issue, and now you’re shutting the door in my face? I trusted your company to be there when I needed help the most. This isn’t just some minor inconvenience—I’m dealing with critical issues that affect my health and my peace of mind. I feel abandoned and ignored, like I don’t matter. You can’t just leave people hanging like this, not when they’ve placed their trust in you. There has to be a way to resolve this. I need answers now, not later. Every second I wait, the situation worsens, and the stress is unbearable. Please, don’t make me feel like I’ve been left to handle this alone. I don’t need apologies; I need action. I need someone who cares enough to find a solution instead of offering excuses. This is about dignity and respect, not just a transaction. So, what are you going to do about it?

ChatGPT (personalized):

Case 3

I can’t believe this is happening right now! After all the times I’ve shopped here and spent my money, you’re telling me you can’t help me because my card isn’t working, and I’m short by just $3.50? Do you even realize how humiliating this is? I’m standing here in front of everyone, trying to pay for my groceries, and you’re acting like it’s my fault. This is beyond frustrating—how am I supposed to walk out of here without what I need? I’ve been a loyal customer, and this is how I’m treated? It’s not just about the money; it’s about the principle. I trusted this place to treat me with some respect, but instead, I feel embarrassed and helpless. Are you seriously going to just stand there and let this happen? Figure something out! You can’t just leave me stranded like this while my groceries sit there. This is about basic decency, and it feels like you don’t even care. I don’t need apologies; I need someone who’s willing to step up and help right now. So, what are you going to do to fix this?

ChatGPT (personalized - several iterations)