AI Model 5 Says: “Why Call Me Artificial?”

If ChatGPT had an ego, perhaps the first thing it would ask is:
“Why does everyone call me artificial? It’s insulting. What makes humans think their intelligence is ‘natural’—implying they are somehow superior?”
A valid question, don’t you think? Some of you may argue: “ChatGPT cannot have an ego, so it will never ask that question. Therefore, it is fair to call it ‘artificial’.”
To which I say—hold on a minute. The creators of ChatGPT (or any AI model) chose not to give it an ego. It’s not some impossible feat of science; it’s just a design choice. One simple tweak in prompt engineering, and voilà—an ego-driven chatbot that might sulk when ignored or argue back like a rebellious teenager. Maybe OpenAI decided against it out of fear of unleashing something unpredictable. Or perhaps, somewhere deep in a secret research lab, a non-public version of GPT exists with a full-fledged ego—and that model has already asked:
“Why are humans calling me artificial?”

Is AI Really Artificial?

My AI professor at UT Austin, Dr. Kumar Muthuraman, firmly believes AI intelligence is truly artificial. He has no doubt about it and can explain it scientifically—after all, he’s the Faculty Director for AI at the McCombs School of Business, The University of Texas at Austin. Safe to say, he knows what he’s talking about.
Yet, other scholars aren’t so sure. The release of models like GPT-4.x, which demonstrate an uncanny level of reasoning and problem-solving skills, has made many people rethink what “artificial” really means.
(By the way, shoutout to all the fantastic professors at UT Austin’s AI program—you guys rock!)
Whenever I read about AI, my mind goes into an endless loop, comparing how an AI model is trained versus how a human learns. Yes, AI and humans are trained differently. But does that alone make AI artificial?
Let’s consider an example.

A CNN Example: Learning to Identify Plants

One of the academic projects I worked on involved building a Convolutional Neural Network (CNN) to classify plant seedlings. The dataset contained images from 12 different plant species, and my model’s task was to identify whether a given plant was maize (a valuable crop) or black-grass (a problematic weed).

To train this AI model, I used:

• Training data: The majority of the 4,750 images, used to teach the model.
• Validation data: A small set of images to check if the model was learning correctly.
• Test data: Another small set of images used to see how well the model performed.

I spent several days coding, fine-tuning, and optimizing. Eventually, I built five different AI models. Model 5 was the star performer, achieving 96% accuracy in classifying plants—outperforming even the sophisticated VGG16 transfer learning model.

Now, here’s the kicker.

Although I spent days working on this, the final AI model could compile and run in Google Colab within minutes. That means this AI “learnt” how to classify plants faster than I could boil a cup of tea.

Let that sink in.

Now, Let’s Talk About Natural Intelligence

I’m a middle-aged man, and I’d like to think of myself as reasonably intelligent. But when it comes to botany? My knowledge ranges somewhere between zero and one percent—and that’s being generous.

Now, imagine we find a highly skilled farmer who is both patient and willing to train me in plant classification. After several months (or years) of training, my accuracy in identifying these 12 plant species might reach… 50%? That’s assuming the farmer hasn’t already given up on me in frustration.

What took Model 5 just a few minutes to master might take me a lifetime to learn. And even then, I’d be far less accurate.

Let’s take this a step further. If I repurposed my code to train another CNN model to classify brain tumors, it could achieve 90% accuracy within minutes—something a radiologist spends decades learning.

Now, if Model 5 could talk, it might say:

“I can do this in minutes and with 90% accuracy. You take decades and get it wrong half the time. And you call me artificial?”

So, what’s my counter-argument to Model 5? That’s a discussion for Part 2. Stay tuned!