Chapter 5 — Data Literacy

Every day you meet numbers and facts: the cricket score, a weather app saying 70% chance of rain, a shop claiming '4.5 stars from 10,000 buyers', a news headline about exam results. Some of these are useful, some are misleading, and some mean nothing without more context. Being able to tell the difference is a skill, and it has a name: data literacy.

Two words. Data means raw facts and figures: numbers, words, images, sounds or measurements that have been collected. Literacy means the ability to read, understand and communicate in a language. Put together, data literacy is the ability to read, understand, work with and communicate using data. A data literate person can collect data, analyse it, show it in a sensible way and explain what it means, and can also spot when data is being used badly.

The data pyramid. Data becomes useful as it climbs four levels, often drawn as a pyramid (Data, Information, Knowledge, Wisdom):

A traffic-light example. A sensor records Red, Light_7. That is data: by itself it tells you little. Processed with a map, it becomes information: 'the north-facing signal on MG Road has turned red'. A driver approaching that signal turns it into knowledge: 'the light in my direction is red'. Wisdom is the decision: 'I should stop the car now'. Notice that only the top level leads to action.

Why it matters for AI. An AI system learns from data. If the data is wrong, incomplete or one-sided, the AI will learn the wrong lessons. People who build AI need data literacy to choose and clean data; people who use AI need it to judge whether an answer or a chart makes sense. In a world where apps, advertisements and social media constantly show us numbers, data literacy is as basic as reading and writing.

Data literacy is not only about maths. It includes asking good questions (Where did this number come from? How many people were asked?), choosing the right chart, and explaining results honestly to others. It is a thinking skill as much as a number skill.

Imagine two newspapers report the same event. One says 'Thousands attend rally'; the other says 'Only a few hundred turn up'. Which is right? A data literate reader does not simply believe the first headline seen. She checks who wrote each report, which source the numbers came from, how the situation was described and whether the key figures agree with other sources. The CBSE activity Impact of News Articles asks you to do exactly this: compare several articles on one trending story and rank the sources from most to least reliable, giving reasons.

Every data tells a story, but be careful before believing the story. Data can be incomplete, out of date, taken from a tiny or unusual group, or shown with a misleading chart. Data literacy protects you from being fooled and helps you make choices based on evidence.

Impact of data literacy. It helps people and organisations to:

• make informed decisions: a family compares electricity bills month by month to see whether a new appliance raised costs;
• think critically: a student asks how many people were surveyed before believing that '9 out of 10 dentists' recommend a toothpaste;
• solve problems: a school studies attendance data to find which days and classes have the most absences;
• innovate: businesses notice patterns in what customers want and design new products.

Example: buying a video game online. A shopping website lists hundreds of games. A data literate buyer uses the data the site provides instead of picking the first item:

• filters by price (low to high) to stay within budget;
• checks user ratings, and also the number of ratings, since 4.8 stars from 6 reviews is weaker evidence than 4.4 stars from 6,000;
• reads the specifications to check that the game runs on the device at home;
• looks at recent reviews, because problems may appear after an update.

How to become data literate. A data literate person is someone who can interact with data to understand the world. You can build this habit step by step: ask where every number comes from; look for the sample size and the date; read the labels and scale of every chart; compare at least two sources; separate facts from opinions; and practise turning data into simple tables and graphs yourself, using a spreadsheet. Talking about data with others, and explaining what it shows in plain words, is part of the skill too.

In AI projects, these same questions decide whether a dataset is good enough to train a model. A model trained on unreliable data, however clever its algorithm, gives unreliable answers.

Becoming data literate does not happen by accident. When a school, a company or a class sets out to build data skills, it needs a plan. The CBSE course uses a six-step Data Literacy Process Framework, which gives guidance on using data efficiently and at every level of awareness. Like the AI Project Cycle, it is iterative: after the last step the team looks back and starts again with an improved plan.

An example in a school. Suppose your school wants every Class 9 student to read graphs confidently.

• Plan: the goal is that 90% of students can read and draw bar, line and pie charts by the end of the term; teachers of AI, Maths and Social Science take part.
• Communicate: the plan is explained in assembly and to parents, so everyone knows why it matters.
• Assess: a short quiz shows that many students can draw bar charts but misread pie charts.
• Develop culture: every subject starts using one graph a week from real data, such as weather, sports or school results.
• Prescriptive learning: students who struggle with percentages get practice sheets; others get spreadsheet projects.
• Evaluate: the same quiz is repeated every month and scores are compared.

If the evaluation shows little progress, the school goes back to Plan and changes its approach. This loop is what makes the framework iterative.

Why a framework? Without one, data skills training becomes a one-time lecture that people forget. The framework makes sure the goal is clear, people know why it matters, starting levels are measured, learning fits each person, and progress is checked with numbers. Notice that the framework itself uses data at the Assess and Evaluate steps: you use data to improve data literacy.

When you sign up for an app, you hand over your name, phone number, photos and perhaps your location. Two questions follow. Who is allowed to use this data, and for what? And how is it kept safe from people who want to steal it? The first question is about data privacy; the second is about data security. The two are related but not the same.

Data privacy is about the proper handling of sensitive data, such as personal details, financial records and medical history, so that it is used only with permission and within the law. Two everyday actions put privacy at risk: downloading an unverified app, which may read your contacts or photos, and clicking 'I agree' to terms of service without reading what you are allowing. Good privacy practice means knowing what data you hold and where it is stored, collecting only the data a project really needs, and always taking the user's consent.

Data security is the practice of protecting digital information from unauthorised access, corruption or theft throughout its life. As more data moves to the cloud, cyber attacks grow, and they can affect anyone.

Risks of a data breach. A breach happens when data reaches people who should not have it. A breach at a government office may expose citizens' identity details; one at a company may hand its secrets to a competitor; one at a hospital may expose patients' medical records to people who could misuse them. The harm includes identity theft, financial fraud, blackmail and loss of trust.

How this links to AI. AI systems are trained on huge amounts of data, often about people. Developers must protect that data (security) and use it only in fair, agreed ways (privacy), or the AI itself becomes a source of harm.

Best practices for cyber security.

If you face cyberbullying or something online makes you uncomfortable, tell a trusted adult at once.

Look at a school report card. It has numbers (marks, attendance), words (remarks such as 'very attentive'), and perhaps a photograph. Each of these is data, but they are different kinds of data, and each kind is handled differently by people and by AI.

Qualitative and quantitative data.

Be careful with labels written as numbers. A PIN code or a roll number is made of digits, but adding two roll numbers makes no sense, so it behaves like a category. Ask: does arithmetic on it mean something?

Discrete and continuous data. Quantitative data has two kinds:

• Discrete data can take only separate, countable values, usually whole numbers: number of students in a class, goals scored, shoe size.
• Continuous data can take any value in a range, including fractions: height, weight, temperature, time, voltage. Its precision depends only on the measuring instrument: 152 cm, 152.4 cm, 152.37 cm.

Data in the three domains of AI. Each domain of AI works mainly with one form of data:

Structured and unstructured. Data neatly arranged in rows and columns, like a spreadsheet of marks, is called structured. Photos, voice notes and free text have no fixed table format and are called unstructured. AI is especially useful for unstructured data, which is hard to analyse with ordinary formulas.

Pick and choose. Classify these: temperature (quantitative, continuous); gender (qualitative); shoe size (quantitative, discrete, because sizes come in fixed steps); favourite colour (qualitative); weight of a person (quantitative, continuous). Practising such classification helps you choose the right chart and the right analysis later.

An AI that recognises potholes must first see thousands of road photographs. Where do they come from? Data acquisition (acquiring data) is the process of gathering the data needed to train and test an AI model. There are three main ways:

Primary and secondary sources. Primary data is collected by you, first-hand, for your own purpose: surveys, interviews, experiments, observations and sensor readings. It fits your question well but takes time and effort. Secondary data has already been collected by someone else and is reused: government open-data portals (many countries, including India, publish data on .gov sites), Kaggle (an online community where people share datasets), Google Dataset Search (a tool for finding datasets by name), and the UCI Machine Learning Repository, a long-running collection from the University of California, Irvine. Secondary data is quick to get, but you must check that it suits your problem and is reliable.

Good data and bad data.

Data from websites. Collecting data from web pages with software is called web scraping, and there are tools that do it. Web scraping itself is not illegal, but using data without permission is. Always check that the website's terms allow the data to be collected and used.

Ethical concerns in data acquisition. Five principles help you avoid harm before it happens:

• Bias: try to avoid any preference or one-sidedness in the data, for example photos of faces from only one region.
• Consent: take permission before collecting or using anyone's data.
• Transparency: explain how you intend to use the data and do not hide intentions.
• Anonymity: protect the identity of the people who are the source of the data, for example by removing names.
• Accountability: take responsibility for your actions if the data is misused.

Keep in mind that more data is not automatically better. A small, clean, relevant, fairly collected dataset is worth more than a huge, messy, biased one.

Two friends collect the same information about some people. One types it into a spreadsheet: one row per person, with columns for first name, last name, date of birth and country. The other writes paragraphs: 'Michael Davison lives in the United States and was born on 04/03/1986...'. Both hold the same facts, but only the first can be sorted, filtered or fed to an AI model directly. This is the idea of usability of data.

Three factors decide usability.

• Structure: how the data is stored. A spreadsheet in which each record has the same set of fields is well structured; free text with no rules is poorly structured.
• Cleanliness: clean data is free of duplicates, missing values, outliers and other errors. A list of phone numbers in which the same number appears twice needs cleaning.
• Accuracy: how closely the data matches real-world values. If the true length of a box is 12.0 cm and three measurements are 12.1, 11.9 and 12.0 cm, the data is accurate; readings of 15 and 9 cm are not.

Data-sharing sites such as Kaggle show a usability rating for each dataset, which reflects how completely it is described and documented. A large dataset that is dirty is not good for AI: the model will learn the errors along with the patterns.

Data preprocessing means getting raw data ready for use. Typical steps are:

• removing duplicate records;
• handling missing values, for example by filling them in from another source or leaving out incomplete rows;
• correcting errors and odd values, such as an age typed as 150;
• making formats consistent, for example writing all dates the same way and all heights in centimetres;
• removing data that is not relevant to the problem.

Features of data. The characteristics or properties that describe each item in a dataset are its features. In a table of student records, the features might be name, age, attendance and marks; in a photo dataset, the colours and shapes in each image. In a table, each column is usually one feature.

Independent and dependent features. An AI model uses two kinds:

The output 'depends' on the inputs, which is where the names come from. Choosing the right independent features is one of the most important decisions in an AI project: if a feature that really affects the result is missing, no algorithm can make up for it.

Niki has 7 candies and Ruchi has 4. How many do they have altogether? Adding gives 11: that is data processing, working on data to get a result. Now a second question: who should get more candies so that both have the same, and how many? Since 11 is odd, they cannot have equal whole shares of 11; if Ruchi gets 3 more, both have 7. Deciding what the numbers mean for a real situation is data interpretation.

Key terms.

Two ways of interpreting data. Because there are two kinds of data, there are two kinds of interpretation:

Collecting qualitative data. Common methods are record keeping (using existing reliable documents, like research in a library), observation of people's behaviour, case studies, focus groups (a guided group discussion), longitudinal studies (studying the same source repeatedly over a long period) and one-to-one interviews. Qualitative analysis usually follows five steps: collect the data, organise it, give each response a code or category, analyse the coded data, and report.

Collecting quantitative data. Methods include surveys of many people, polls (usually one simple question), structured interviews, observations counted over a fixed time, and longitudinal studies. Quantitative analysis then links each variable to a measurement scale, applies descriptive statistics such as averages, decides the scale, and shows the results in a suitable format.

Trend analysis. A longitudinal look at data helps spot trends. The CBSE activity uses Google Trends to list the most searched sports and films of a year. Seeing how a quantity changes over time lets us predict and prepare.

Why interpretation matters.

• Informed decision making: a decision is only as good as the knowledge behind it. If a school knows the average height of its students, it can order desks of the right size.
• Reduced cost: a restaurant that sees some dishes rarely ordered or badly reviewed can drop them and save money.
• Identifying needs: if data shows a particular pizza is most popular with 8 to 10 year olds, the menu and advertising can be planned for them.

The same result can be shown in three ways, and choosing well makes the difference between a report people understand and one they skip. These are the three types of data interpretation (presentation): textual, tabular and graphical.

Textual. Data is written in sentences, as in 'Of 50 students, 6 scored below 20 in Maths and 5 scored 45 or above'. This suits small amounts of data that can be understood by reading, but becomes hard to follow when there are many numbers.

Tabular. Data is arranged in rows and columns, with a title describing the table and column headings describing each column. Here is the CBSE activity table showing how many of 50 students fall in each marks range in five subjects:

Graphical. Graphs reveal patterns at a glance:

• Bar graph: bars (vertical or horizontal) whose lengths show values; best for comparing categories, such as the number of Maths students in each marks range (see the figure).
• Pie chart: a circle cut into slices, each slice proportional to its share of the whole; best for parts of one total, such as how Physics students are spread across ranges. Each slice's angle is its fraction of 360°: 20 of 50 students is 20/50 × 360° = 144°.
• Line graph: points joined by lines; best for showing change over time, such as monthly rainfall or a student's scores across tests.

Reading the table with a graph: in Maths, 6 + 14 = 20 students (40%) scored below 30, while in Biology only 8 did. A bar graph of the two subjects makes this difference obvious immediately.

Data visualisation tools and dashboards. Spreadsheets such as Excel or Google Sheets draw charts from a table in a few clicks. Tools such as Tableau Public and Datawrapper go further and build interactive dashboards: pages with several linked charts that update when you filter the data. In the CBSE project, students list 5 to 10 favourite songs with album, artist, genre, year and song length, count the songs in each genre, draw a bar chart by hand, and then rebuild it in Tableau by dragging the Genre field to Columns and the count to Rows; Tableau's Show Me menu can then turn the same data into a packed-bubble chart, where the biggest bubble is the most common genre.

Why visualise? Visualisation makes it easier to extract useful information from a dataset, spot outliers and trends, and communicate findings to people who will not read a table. It is used at the Data Exploration stage of the AI Project Cycle.

Honest charts. Always label axes and give a title, start bar-graph axes at zero, and make sure pie slices add up to the whole. A chart that breaks these rules can mislead even when the numbers are correct.