In short
Machine learning is a way for computers to learn patterns from examples and use them to make predictions, rather than following rules a developer writes by hand. Precious Kolawole, a data scientist and machine learning engineer at Lendsqr, makes the idea concrete with a child who learns from a candle flame and a program that learns to spot cats. In her Q&A she adds a practical route for beginners: Python first, then data analysis and data science, then machine learning.
Key takeaways
- Explain machine learning to a non-technical person using a simple everyday analogy.
- Distinguish machine learning, which learns patterns from labelled examples, from explicit programming, which follows written instructions.
- Describe how training on labelled data produces a model that predicts on new inputs, using the Y equals f(X) idea.
- Recognise the kinds of hard, data-heavy decisions where machine learning is most useful.
- Plan a beginner learning path from Python to data analysis, data science and then machine learning.
Cutting through the buzzwords
Kolawole opens with a confession many practitioners will recognise. After writing code, cleaning data and building models, she wondered whether she could explain any of it to a complete lay person, and decided she had to learn how. Her starting definition is deliberately plain:
"Machine learning is a large subset of artificial intelligence that allows computers to learn from data." (0:32)
It observes patterns, learns from them and uses what it learns to make predictions. But definitions only go so far, so she moves straight to an example.
The candle and the stovetop
A young child sees a candle flame for the first time, reaches out and gets burned. The lesson lodges in memory: something red and bright is painful. Two days later the same child spots a glowing stovetop in the kitchen, remembers the candle and leaves it alone. Kolawole draws the line clearly:
"To be clear, this is just machine learning because this young child learned pattern from the candle flame to make predictions on the stovetop." (2:19)
Then she changes one detail to show what machine learning is not:
"Let's now say this young child, he ignored the stovetop because his parents warned him against it. Then that would be explicit programming and not machine learning." (2:19)
In traditional programming a developer writes instructions and the computer produces an output from them. Machine learning reverses that: you give the computer inputs and outputs, such as images and their labels, and it works out the rule, the model, on its own.
What a model is for
To explain why models matter, Kolawole sorts everyday decisions into three groups. Some are easy, like whether to brush your teeth or pay the rent. Some are harder, like asking out a crush or planning a productive day. And some are close to impossible without code and data, such as which stock to invest in or whether a borrower will repay or default on a loan. That third group is where machine learning shines.
"And the ultimate goal of machine learning is to simplify large amount of data into a single decision or a prediction." (3:56)
The model, she says, is a representation of how the real world works on your data, and it is the final product of the process.
Teaching a computer to spot cats
Her second example is a classic. Suppose you want a program that returns true when an image contains a cat and false when it does not. Instead of writing rules for whiskers and ears, you give the computer, say, 1,000 cat images labelled true and 1,000 non-cat images labelled false. She frames the relationship as Y equals f(X): X is the image, Y is its label, and f is the function the model creates automatically. Working out that function from the data is training. After training, a new image passes through the learned function and the model returns its prediction, true or false.
The example is simple on purpose. It shows the full cycle, labelled data in, a learned mapping, predictions out, without any mathematics beyond a single line.
Where beginners should start
The Q&A turns to newcomers who want to get into machine learning or data science. Kolawole says you cannot dive straight into machine learning: there are resources to work through and a programming language to learn first. People usually learn R or Python, and Python is the most common choice. A typical sequence is Python, then data analysis and visualisation with tools such as Power BI, Seaborn, Tableau and Matplotlib, then data science, cleaning data and generating insights, and only then machine learning itself, with model building, deployment and eventually deep learning.
Asked whether you can start with no code, she points to ready-made platforms, mentioning tools on Microsoft Azure, where you can build and use models through a configured interface. Her bottom line is still clear: most of the time, doing machine learning means knowing how to write code.
Watch the talk
Explain Machine Learning as to a kid, Precious Kolawole, about 11 minutes: https://www.womentech.net/video/explain-machine-learning-kid
About the speaker
Precious Kolawole is a Data Scientist/ Machine Learning Engineer at Lendsqr, a financial company in Nigeria, with skills in data preprocessing and analysis, data visualisation, supervised machine learning, deep learning, computer vision and web development. She learned to code through online and offline courses, Stack Overflow and trial and error, and her interests lie at the intersection of artificial intelligence and healthcare.
Questions this talk answers
- What is machine learning in simple terms?
- How is machine learning different from traditional programming?
- What does training a model look like in practice?
- How should a beginner get started in machine learning?
- Can you do machine learning without code?
Drafted with AI from the talk transcript. Editor review pending.