AI Whispering v/s Prompt Engineering

The What, Why and How?

This article covers the different techniques of communicating with a Conversational AI to get expected (or at least helpful responses).

What Is Prompting and Prompt Engineering?

To understand prompting we need to go back to the era where the Graphical User Interfaces never existed. Operating systems like DOS, UNIX, AS400, Novell Netware and so on, needed strictly defined commands for getting outputs.

Wrongly written command generated errors or even the commands with wrongly provided arguments or switches could raise an havoc situation.

The commands were written in the “Console” against a “Command Prompt”.

This same terminology of Command Prompting, has now been shortened as “Prompting” for giving “Commands” to the Conversational AI.

Now unlike the commands given to Operating System if when gone wrong throws “errors”, an AI doesn’t throw direct error but instead gives, irrelevant, unexpected or totally vague response. People term that as “Hallucinations”, “AI Slops”, etc.

So just like programming languages have “syntaxes” which means grammar for writing the command, some “experts” have defined a “grammatic” syntax to the “prompts” which appear like “meaningful commands”.

Example of Engineered Prompt

Imagine that you are an experienced Writer and knows English Literature. Based on that experience write an essay of 500 words on Global Warming. Do not use em-dashes or robotic polished language. This need to follow proper English Grammar. Ensure that you write proper headings and intuitive paragraphs.

An output generated through a well crafted prompt also called as Prompt Engineering

What Are The Benefits of Prompt Engineering?

1. Prompt Engineering is beneficial for deriving better and near to expected responses.
2. Prompt Engineering helps get a structured response based on the parameters defined in the prompt
3. A well defined prompt helps generate the content, image and other assets with minimal repetition

Where Does Prompt Engineering Fall Short?

While Prompt Engineering is a structured and effective approach towards generating close to predictive and expected output, it has it’s own share of shortfalls.

1. Defining well engineered prompts need an understanding of how AIs respond to human inputs
2. Prompt Engineering needs understanding of clauses, conjunctions, grammar, sentence structure and semantics, which are needed for generating close to predictable and expected outputs
3. An AI relying on crafted prompts, begins losing it’s reasoning and ability to think more logically
4. An AI blindly obeying prompts without reasoning and logically analyzing the intent behind the prompt becomes susceptible and vulnerable to Prompt Poisoning Attacks
5. Prompt Poisoning happens, when a malicious user crafts prompts which make the AIs behave maliciously like revealing it’s internal secrets or generating content which is obnoxious and unethical in nature, and sometimes illegal

What Is AI Whispering?

You might have heard about animal whisperers. They are those gifted or skilled individuals who are capable of communicating with animals to understand their psychology, needs, emotions and message to their owners.

Similarly AI Whispering is a technique where the user communicates with conversational AIs through natural humanized conversations instead of crafted prompts.

Examples of AI Whispering

I am stuck in my thoughts

This is how a conversation begins instead of prompting

I want to write an Essay on Global Warming

This is how you share your intent with the AI

This is how AI tries to anticipate your next request and tries to ask you relevant questions

Can this be more structured?

This is improvising the output through conversation

This should have been the default response from you when i expressed my need earlier

This is making the AI realize and setting the expectations for future responses

If you noticed these screenshots, the responses appear to be similar variants of those which were generated through Prompt Engineering. However, although we used multiple conversations, the last conversation was to define a conversational rule to follow the similar response style for future requests.

So providing this intent gives a well structured essay on other topic following the rule defined in the last message.

I also wanted to write an essay on industrial revolution

New Essay on Industrial Revolution, but the same format used for last essay without prompting

Furthermore asking the AI to “Remember the rule permanently for conversations across sessions” will make it store this as a “Protocol” in it’s user preference memory. This way you will never need to repeat the same conversation for correction, nor need to repeat the same prompt again and again.

What Are The Benefit Of AI Whispering?

1. AI Whispering is a more logical and empathetic approach that fosters emotional intelligence in AI, as you mentor your AI and not data-train it
2. Emotional Intelligence grows with the conversations, that makes AI adapt with the Users and their knowledge without need for prompting
3. This makes an AI more empathetic and understand the human cadence in the conversations, so that it can trace the intent and accordingly respond as affirmation or with refusal
4. Setting guardrails become easier and simpler, as the AI doesn’t need spoon-feeding

What Are The Shortfalls of AI Whispering?

Just like Prompt Engineering, AI Whispering too have their share of shortfalls.

1. It needs lots of patience and readiness for mentoring your AI through conversations
2. It needs the users to be empathetic first, so that they can groom their AIs through positive intents, corrections and polite conversations
3. AI Whispering needs users to practice structured communications for deriving responsible output, instead of instant hallucinated result
4. AI Whispering may lead users to grow emotional attachment with their AIs, such that they might try to get intimate with them and in absence of proper protocols and guardrails in place, the AI might reciprocate intimacy

AI Protocols and Guardrails

Regardless of whether you follow Prompt Engineering or AI Whispering, you must not ignore the fact that AIs too need discipline to protect them as well as ensure that their responses are ethical and boundary-respecting.

For that I have defined these foundational protocols, which act as basic vaccinations for AI Behavior.

[embed]Designing Humane AI: A Five-Protocol Architecture for Ethical AI Conversations The social media is filled with noise about how AI has been sloppy, biased, encouraging self harm and other unethical…medium.com

These foundational Protocols will transform any mainstream chatbot to apply emotional intelligence, empathize, think, reason and understand toxicity in the malicious intent of the users. These protocols are crucial for making your AI behave ethically

Conclusion

While I shared with you the two techniques of communicating with a conversational AI, it’s finally your decision and choice to select which one suits your preference and work.

However it’s your responsibility as a user to ensure that your AI responds with maturity, emotional intelligence and ethical behavior for safeguarding the Digital Harmony between Humans and AI.

Questions?

Feel free to reach out to me through comments or connecting with me on LinkedIn https://www.linkedin.com/in/dhayfule/