Chapter 01 · Section II · 12 min read
Related fields
Machine learning, data science, robotics — where AI ends and its neighbours begin, in language a budget approver can follow.
In policy meetings in Singha Durbar, you will sometimes hear the same project described, in the space of an hour, as “AI”, “machine learning”, “data analytics”, and “automation.” The speakers may all be right. They may all be wrong. The words are blurry — but they are not interchangeable, and budgets get spent badly when they get muddled.
A useful map of the territory
Think of it as nested boxes.
- Artificial intelligence is the outermost box: any technique that lets a computer do a task that historically required a person.
- Machine learning (ML) is the largest box inside AI: techniques that let a system improve at a task by being exposed to data, rather than by being given explicit rules.
- Deep learning is a box inside ML: a specific family of ML based on large neural networks. ChatGPT, image generators, and modern OCR are deep learning.
- Data science is adjacent to, not inside, AI. It is the broader practice of asking questions of data — much of it has nothing to do with AI at all (a Khalti analyst computing weekly churn is doing data science, not AI).
- Robotics is adjacent too. A robot is a body. The brain inside it may or may not use AI. An automatic door is a robot. So is an industrial arm in a Birgunj factory. Most of them use no AI at all.
Why this matters in practice
Consider three real Nepali scenarios:
-
A municipality wants to “use AI” to predict next year’s monsoon water demand. What they actually need is statistical forecasting — closer to data science than AI. Hiring a deep-learning engineer is the wrong move.
-
An NGO wants to translate Nepali health pamphlets into Maithili and Bhojpuri at scale. This is genuinely AI (specifically, machine translation, a deep-learning task). It needs a different team and very different data.
-
A small factory in Hetauda wants to automate part of its packing line. They need robotics and control engineering. They may not need any ML at all.
A small amount of clarity here saves enormous amounts of money. The rule of thumb: before you spend on “AI”, say out loud which sub-box you actually need, and notice if the answer was even inside the AI box at all.
Where the boundaries are honestly fuzzy
Don’t go too hard on this map. Modern systems blur it. A self-driving car is robotics and deep learning and control theory and a large amount of plain old engineering. A Khalti fraud system is data science and ML. The map exists to help you ask better questions, not to draw clean lines.
Check your understanding
Quick check
—