AI THROWDOWN: HUMAN vs CHATGPT
OVERVIEW
I challenge AI in a head-to-head throwdown to see how my humanity stands up in a mano-e-machine competition. Read on and let's get started.

PREFACE
Recently, during my morning coffee routine with CNBC playing in the background, an analyst proclaimed that the “rise of the market in 2023 was largely due to the hype with AI”. Hearing this, I nearly spit my coffee out! What?! Not again! I have barely recovered from the crypto hype, preceded by the metaverse hype, preceded by the Kathy Wood disruptor hype, preceded by the Google Glass hype, and so on. In lockstep, corporate messaging began flooding media outlets with proclamations of the inevitable societal transformation AI would bring. Granted, I anticipate that a future of AI omnipresence is inevitable. Yet for me, the current frenzied hype is, once again, overblown and distracting. My science background teaches me that details and evidence can be better guides than exaggerated proclamations. With that in mind, I decided to challenge AI in a head-to-head throwdown to see how my humanity stands up in a mano-e-machine competition. Here we go!

COMPETITION CONTEXT
Using a technical Knowledge Transfer Guide PDF document as source material, create a six-question multiple choice quiz to assess participants' performance upon completion of the knowledge transfer guide. Each question should have four possible choices, where only one of the choices is the correct answer.

Prepare the Source Document
Human: Read the document and understand the concepts necessary to create a meaningful quiz.
ChatGPT: Since ChatGPT cannot access a PDF file, the human converted the PDF file into a text document and removed unnecessary headers and footers.
Note: Slight advantage to the human, considering the human contestant is also the document author.

Competition Round 1
Preparation
Human: Open, read, and understand the concepts in the PDF Knowledge Transfer document. Make a quiz based on the document concepts.
ChatGPT: Create the following prompt for ChatGPT to execute: Context: training assessment. Using the source document, create a multiple-choice quiz with 6 questions, each question must have at least 4 concise answers. Only 1 answer is correct. Source document: (text document copy/paste into ChatGPT window).

Results
Human: 2 hours to read and understand the basic concepts in the document and to create six questions with four multiple-choice answers, with only one of the answers as the correct answer.
ChatGPT: Executed the prompt and generated a quiz in less than 15 seconds.

Score

Human	ChatGPT
0	1

Competition Round 2
Preparation
Human: Create an HTML working copy of the quiz from Round 1.
ChatGPT: Prompt: Create HTML code for the quiz results content, use a radio button for each multiple-choice selection and add a submit button at the end of the quiz, display the score when the submit button is clicked. Quiz results: (copy and paste the previous ChatGPT quiz results into the ChatGPT prompt box).

Results
Human: 2 hours: using Dreamweaver IDE to create the HTML and JavaScript for answers and scoring; evaluating the JavaScript functions. Note: I used previous CSS style sheets and a few modified JavaScript functions from my code repository.
ChatGPT: Executed the prompt and generated the HTML quiz with scoring in less than 15 seconds.

Score

Human	ChatGPT
0	1

You can view the HTML quiz generated by ChatGPT with this link:
https://www.graphcube.com/KTQuizCGPT.html

CONCLUSIONS
ChatGPT outperformed the human exceptionally well in this competition both in terms of its speed, its ability to synthesize meaningful content from the technical document source, and its ability to construct a working HTML quiz with minimal input. Clearly, it may make sense to deploy ChatGPT in cases like this, with the following caveats:

1. Human SMEs (Subject Matter Experts) should review the responses ChatGPT generates. For this competition, the responses were nuanced and on-point. For more complex technical tasks, the involvement of a human SME is critical.
2. Meaningful generative results can be achieved from complex tasks if the SME provides clear and succinct prompts and context to ChatGPT. For example, I first asked ChatGPT to create the quiz based on some simple parameters. I then asked ChatGPT to create an HTML version of the quiz with slightly more complex parameters (scoring, JavaScript) which ChatGPT performed flawlessly and in under 15 seconds.
3. For highly complex or long and detailed prompts, ChatGPT performs better by breaking the tasks into multiple prompts and then letting ChatGPT build on the successive responses.

GOING FURTHER
I further prompted ChatGPT to perform the following everyday developer tasks:

Task 1: Prompt: Context: PHP Code. Write an example of URL variable sanitization in PHP code where the URL variable = productID.
Result: in under 5 seconds ChatGPT wrote the PHP code as well as an explanation of the code (see below).

PHP Code Block Results

// Get the productID from the URL
$productID = $_GET['productID'];
// Sanitize the productID to prevent potential security vulnerabilities
$sanitizedProductID = filter_var($productID, FILTER_SANITIZE_NUMBER_INT);
// Now $sanitizedProductID contains the sanitized product ID that contains only numbers

// Rest of your code here...

Task 2: Prompt: Context: JavaScript Code. Write JavaScript code to calculate the volume of a sphere where the radius variable = rd, and the value of rd = 20 cm.
Result: in under 5 seconds ChatGPT wrote the JavaScript code as well as an explanation of the Code and the formula for calculating the volume of a sphere (see below).

JavaScript Code Block Results

// Given radius in centimeter
var rd = 20;
// Calculate the volume of the sphere
// Formula: 4/3 * π * r^3
var volume = (4/3) * Math.PI * Math.pow(rd, 3);
// Output the calculated volume
console.log("The volume of the sphere with radius " + rd + " cm is: " + volume + " cubic cm");

Of course, these are basic programming tasks. Using ChatGPT to automate these and more complex tasks holds enormous potential. To extrapolate, a developer could prompt ChatGPT a series of code blocks to perform, then construct the blocks into an app, then give ChatGPT with instructions to perform app testing and display the testing results in a detailed summary report.

OUR FUTURE
So, as a developer and technical writer, one of the burning questions is “Will I be replaced by AI?” The answer is somewhat nuanced: “Yes of course, I will, in terms of routine tasks, but my subject matter expertise and ability to effectively use AI are unique human asset...for now”. Perhaps we should think of this new paradigm in terms of “I will be augmented by AI”. This is an age-old phenomenon often repeated when new technology becomes commonplace. I am reminded of an analogy I heard recently on YouTube (paraphrased):

AI is positioned to create new era of specialized skills, requiring a strategic approach to prevent a displaced workforce from being left behind. To illustrate, consider the realm of creative possibilities that open up when routine tasks such as software programming and document writing are delegated to our AI counterparts. By entrusting tasks that once consumed hours, now accomplished within seconds by AI, creative energies can be channeled towards refining and perfecting products and output. I envisage an AI-powered future where humans are untethered to physical office spaces and are allowed to thrive from anywhere, alleviating strains on our environment, while working alongside both human and AI peers, fostering a collaborative and productive synergy.

User Comments

02/02/2024
Mike G  Are you not afraid that AI will soon completely replace you in your role as a developer?

↳ Hi Mike. No, not at all. I have found that AI (ChatGPT) has helped me become a much more efficient developer. It can quickly debug my code, give me code alternatives, and explain code logic. That said, I do see a near-future where I will probably no longer code, but I will give instructions to a more sophisticated AI to code for me.

---

01/11/2024
Thomas  Interesting, relevant, and enjoyable to read. Clever work Keith.

---