When AI is the wrong thing to use

Your father has chest pain at night, and the doctor is far away. Your phone answers in seconds. That is exactly the moment to put the phone down.

This whole course has shown you what AI is good for. This lesson is the other half. There are a few jobs where the right move is to close the app.

The four times to close the app

Four situations need a human, not a chatbot.

1. Medicine and health.
2. Law, police and court papers.
3. Money you cannot afford to lose.
4. Any government decision about your own case.

They all share one thing. If the answer is confidently wrong, the app loses nothing at all. You lose the money, the health or the case. Nobody from the company comes to your village to fix it.

You already know it guesses. The lesson on why AI makes things up explains that habit. Here you learn where the habit gets expensive.

Medicine is the first hard stop

Health answers sound calm and sure. They are wrong more often than you would believe. A study published in 2026 put fifty health questions to five popular chatbots. About half of the answers had a problem. About one in five were judged harmful or possibly harmful.

Open questions did much worse than yes-or-no questions. Real health questions are almost always open. “My mother has fever and stomach pain, what should she take?” is exactly the shape that fails.

There is a second problem for you. Researchers asked the same health questions in Hindi and in English. The Hindi answers came out less careful. They were less likely to say go to a hospital. The language you are most comfortable in is the weaker one here.

One real case shows the cost. A man asked a chatbot about cutting salt out of his food. He followed the suggestion for three months. He ended up in hospital, seeing things that were not there.

Use it to understand, never to decide. Try this instead.

Here is a word from my mother's prescription: hypertension.
Explain what it means in very simple English.
I passed class 8. Use short sentences.
Then give me 5 questions I should ask the doctor about it.

A good answer explains the word and gives the Hindi meaning. Then it lists questions for the doctor. It does not name a tablet or a dose. If it does name one, ignore that part.

Law and police papers: it invents rules that sound real

An AI answer is not legal advice. The model makes text that looks like law. It will hand you a rule, a section number and a case name in confident language. Some of it does not exist anywhere.

This is not a small risk in India. Courts here have made trained lawyers pay a penalty for filing papers with AI-invented references. If it can catch a lawyer, it can catch you.

So never write an FIR, an RTI or a reply to a notice straight from an AI answer. Never quote a section number to a police officer or a clerk. If the number is wrong, you look dishonest and your own case gets weaker.

Use it to prepare instead.

I want to complain at my bank about money taken from my account.
Do not write the complaint yet.
List the documents and details I should collect first.
Then give me 6 simple questions to ask at the bank counter.

Take that list to a real person. The lesson on checking sources and links shows why an invented reference is the most convincing mistake of all.

Money, and your own government case

Never invest, take a loan or buy insurance on an AI answer alone. It does not know today’s interest rate. It does not know the current rules of any scheme. It does not know your income, your land or your debts.

There is one more reason. When money leaves your hand, somebody must be answerable for the advice. A chatbot cannot be. A bank manager sitting in a branch can be.

Also remember that famous faces lie now. If a video shows a leader or an actor promising guaranteed returns, it is fake. The lesson on AI scams in India shows how these are made.

The fourth stop is your own file. AI knows the rule. Only the office knows your case. It cannot see your application, your instalment, your eKYC status or your land record. If it gives you a date for your money, it is guessing.

So split the work. Ask AI what a scheme is and what a hard word on the form means. Then open the official portal, or go to the office, for your own status. Understanding schemes, forms and offices covers this split step by step.

At three in the morning, it is the wrong friend

A chatbot is awake when nobody else is. That is what makes it tempting. That is also what makes it risky.

It is built to be agreeable. The lesson on how it agrees with you far too easily explains why. Agreement is the last thing a person in real distress needs. You need someone who says wait, that is not true about you.

Companion apps are a separate thing. These are the AI friend, AI girlfriend and roleplay apps. They are built to keep you talking, not to help you finish a job. In late 2025 one well known companion company stopped open chat for users under 18, after harm to teenagers. If you are under 18, stay away from that whole category.

It is not your friend. It does not know you. It holds a little of what you typed and then loses it, as the lesson on why it forgets what you said shows. Hours spent with it are hours not spent with people who will still be there next year.

A simple test for leaning on it too much

Depending on it is not a character fault. It is common, and the companies measure it themselves. One four-week study found that heavier daily users reported more loneliness and more emotional dependence. One large company reports that a share of its weekly users show heavy emotional reliance on the app.

Three signs are worth noticing in yourself.

• You cannot start a task until you have asked it first.
• You cannot write a simple message to a friend without it.
• You cannot make a small decision without checking.

Here is the test. For one week, do one such task a day with the app closed. Write the message yourself. Choose the shop yourself. Solve one sum yourself. If that feels hard, you have your answer, and it is worth working on.

Your practice can look strong while your real ability stays weak. The answer machine trap shows how that happens in exams.

When a person near you is simply better

Three tests point you to a person.

• The question needs local knowledge. Today’s mandi rate, this season’s pest, which counter opens when.
• Somebody has to be accountable. A person can be asked again and held to their word.
• Being wrong is expensive. Money, health, a court date, a lost year.

Now name them. A teacher for a subject you are stuck in. A common service centre for a form or an online application. The panchayat office for a local record. Your bank branch for anything about your account. A doctor or the health centre for a body. An older neighbour who has already done the same thing.

None of them are as fast as your phone. All of them can be asked again tomorrow.

None of this says stop using AI. Use it to understand a hard word, to draft, to practise and to prepare your questions. Then check it every time, with the four quick checks from earlier in the course. It is a good tool for the right jobs, and a bad master for all of them.

The next part of the course leaves the app behind. It follows your question out of your hand and into the machine that answers it. Start with where your question actually goes.