Definition
How long would it take you to check a million payments for fraud?
AI is needed for problems that are too large, too fast, too repetitive or too complex for people to handle alone, such as sorting millions of images, spotting fraud in real time, or finding patterns in huge amounts of medical or weather data.
In simple words
If checking one payment takes you 10 seconds, a million payments would take about four months without sleep. India makes billions of digital payments every month. Only a machine can check them as they happen, and flag the few that look suspicious for a person to look at.
Why it matters
Understanding why AI is needed helps you judge when it is worth using. AI is most useful where the scale or speed is beyond people, and least useful where a simple rule or a human conversation would do better.
Key terms
- Scale
- The sheer number of cases to handle, such as every payment made in a country.
- Real time
- Happening immediately, as events occur, with no noticeable delay.
- Repetitive task
- A task done the same way again and again, which tires people and leads to mistakes.
- Pattern
- A regular relationship hidden in data, such as the link between rainfall upstream and floods downstream.
- Human judgement
- A person's ability to weigh values, context and consequences when deciding.
Explained in detail
Problems that outgrew people
For most of history, the amount of information a society produced was small enough for people to handle. A shopkeeper knew every customer, and a clerk could read every form. That is no longer true. India's UPI system now handles well over fifteen billion payments every month. Billions of photos and videos are uploaded to the internet every day. A single modern hospital scanner produces hundreds of images for one patient. No number of human workers could look at all of this, and certainly not at the speed required.
The five reasons, one by one
Too large: some jobs involve more items than people could ever process, such as ranking billions of web pages for a search. Too fast: some decisions must be made in a fraction of a second, such as blocking a fraudulent payment before the money moves. Too repetitive: some checks must be repeated millions of times with the same care, and human attention fades after a few hundred. Too complex: some patterns involve so many factors at once, as weather does, that no person can hold them in mind. Too dangerous: some places, such as a collapsed building, a deep mine or a damaged nuclear plant, are unsafe for people to enter.
Reaching people who had no access
AI is also needed where skilled people are too few. Many villages have no eye specialist, but a health worker with a camera and an AI model can find the patients who most need to travel to one. A farmer far from any agriculture officer can photograph a diseased leaf and get advice on a phone. Someone who reads only Odia or Assamese can have an English document translated in seconds. In these cases AI does not replace an expert who was there before. It brings a first level of help to places where there was none.
When AI is not needed
Because AI is fashionable, it is sometimes used where it does not belong. A good engineer first asks whether a simpler solution will do.
| Situation | Better choice | Why |
|---|---|---|
| The rule is simple and never changes, such as adding tax to a bill | Ordinary software | Exact, cheap and easy to check |
| There is very little data to learn from | Human expertise | A model cannot learn patterns from a handful of cases |
| The decision changes someone's life | A person, perhaps assisted by AI | Needs judgement, explanation and responsibility |
| A mistake would be very costly and hard to notice | Careful human process | AI errors can be confident and hidden |
How it works
| Reason | Example |
|---|---|
| Too large | Searching billions of web pages in under a second |
| Too fast | Stopping a fraudulent payment before it goes through |
| Too repetitive | Checking every bottle on a factory line for cracks |
| Too complex | Predicting the weather from millions of readings |
| Too dangerous | Robots inspecting a damaged building or a deep mine |
Try it yourself
A job too big for people alone
India makes a huge number of digital payments each day. Who could check them all?
Too big for people alone?
Decide whether each task really needs AI, or whether a person or a simple rule is enough.
Translating a live speech into ten languages at once
Choosing a birthday present for your best friend
Spotting fraud in every card payment in the country
Finding one face in 10 million photos
Switching on a fan at 30 °C
Common mistakes
- Using AI for everything because it is popular. Sometimes a simple rule is better.
- Thinking AI is needed only for difficult problems. It is often needed for simple tasks repeated millions of times.
- Forgetting the cost. AI needs data, electricity and people to check it.
Check yourself
Summary
- AI is needed when tasks are too large, fast, repetitive, complex or dangerous for people.
- It is not always the right tool: simple rules or people are sometimes better.
- Used well, AI frees people for work that needs care and judgement.
Finished reading?