In 2005 — two decades before the current AI revolution — I developed an application in Java, as a hands-on experiment with artificial neural networks. I developed this application in my spare time, driven by curiosity and a desire to understand how neural networks actually worked. I converted the application to JavaScript in 2026 so that the original algorithm can now run directly in a browser.

The application demonstrates how a neural network can learn to locate an object in a two-dimensional space. It was inspired by Tom M. Mitchell's seminal 1997 textbook Machine Learning, in particular Chapter 4, Artificial Neural Networks. Rather than simply studying the theory, I wanted to deepen my understanding by translating the mathematics and algorithms described in the book into a working application.


When the application opens, the neural network first trains locally in your browser. Once "Neural network ready" appears, click Create Object to randomly position the black square, then Locate Object to see the network's estimate, represented by the red circle. Rather than calculating the position using a programmed formula, the network learns from examples and makes its own prediction. Find out more.

The experiment illustrates a fundamental principle of machine learning: the network learns from examples and applies what it has learned to previously unseen situations. Its predictions are approximations — sometimes surprisingly accurate, sometimes with a significant margin of error — illustrating both the potential and the limitations of learning from data.

There is another lesson hidden in this experiment. In 2005, developing the original Java application required considerable time and effort to translate mathematical concepts into working software. In 2026, ChatGPT converted that same application into a working JavaScript version in just 55 seconds. This raises an intriguing question: what does this mean for the future of software development?

The lesson is not that software development no longer matters, but that AI is changing how software is created — and where human expertise adds the most value. The emphasis increasingly shifts from writing code to understanding the problem, designing the solution, providing direction, and critically evaluating the result. AI dramatically shortens the path from understanding a problem to delivering a working solution.

About six months later, I made a deliberate choice about where to invest that same curiosity and learning effort. I became an independent consultant and decided to dedicate my professional development — and much of my free time — to becoming a better manager, rather than continuing to study mathematics and software development. The qualifications I have earned along the way reflect that journey.

The 2005 application is more than an interesting artifact from my technical past. It reflects the technical foundation on which I built my management career. There is also a fitting irony to the story: the neural network I programmed in 2005 was brought to the web 21 years later with the help of a far more capable descendant of the same field of machine learning. In a sense, AI helped bring my original AI experiment back to life.