Artificial Intelligence — I Want to Understand It, Not Just Use It

AI is probably my biggest interest, but what interests me about it goes way beyond using ChatGPT or learning how to code a model.

I want to understand how AI got here.

I’m interested in how artificial intelligence evolved from basic rule-based systems and search algorithms into machine learning, neural networks, CNNs, LSTMs, transformers, and the much larger models we have today. Every time a new type of model appears, I want to know what changed. What could the previous model not do? Why was a new architecture needed? What problem was researchers trying to solve?

Then there is the part that interests me even more: the data.

I used to think the model was basically the most important part of machine learning. The more I learned, the more I realized that the model is only one part of the story. Where did the dataset come from? How was it cleaned? What was included or left out? How much data was used for training versus testing? What happens when the model sees something completely different from what it learned from?

My own research started giving me a chance to explore these questions instead of just reading about them. I used CNNs with satellite imagery for poverty prediction, worked with training and testing datasets, and achieved about 90% accuracy with a custom CNN. I later worked with LSTM networks for market forecasting. Through different AI programs, I also learned about neural networks, CNNs, deep learning, NLP, computer vision, Python, PyTorch, pandas, NumPy, and scikit-learn.

But I don’t want to stop at:

“The model got 98% accuracy.”

My next question is:

Why?

Why did this model perform better than another one? Was it the architecture? Was it the dataset? Was it preprocessing? Was it the parameters? What did the model get wrong? And would it still work on completely new data?

Then comes the question that matters even more to me:

What can we actually do with it?

That’s where my interest in AI connects with people.

I’ve already explored AI for poverty prediction, and my other projects have made me think about how technology could help communities.

I’m interested in AI applications involving poverty, mental health, education, communities, the environment, public services, and other problems that actually affect people’s lives.

At the same time, I don’t think every problem needs an AI model just because AI is exciting. I want to be able to ask both questions:

“Can we build this?”

and

“Should we build this, and who will it actually help?”

That’s really why I’m interested in AI.

I don’t just want to become someone who knows how to use AI. I want to understand how it evolved, how the models actually work, how they learn from data, why they sometimes fail, and eventually how we can use them responsibly to make people’s lives better.