AIYINGFrom AI Awareness to Accomplishment
THE AIYING KNOWLEDGE HUB

See Differently. Generate Possibilities. Accomplish What Matters.

Explore practical AI Engineering, Generative AI, Lateral Thinking, PO provocations, case studies and learning experiences.

Explore the SeriesExplore Courses

Explore AIYING Series

The Effective AI EngineerPractical engineering case studies The Generative ViewFrom answers to possibilities Lateral AI & POProvocation, movement, discovery Hands-On LabsBuild, test and accomplish AI Engineering PrinciplesUnderstand and apply AI Fluency & LearningAwareness to accomplishment

Latest Articles

Thursday, November 19, 2026

Data, Examples and Labels in Machine Learning | AI Made Simple, Chapter 2



Ravindra Dastikop with ChatGPT

The college office has decided to try an AI system that sorts student messages. Two staff members agree on how to label the first few enquiries. Then they reach this one.

“I paid the fee yesterday, which was the last date, but the portal still says it is due.”

One staff member marks it Deadline because the message mentions the last date. The other marks it Payment status because the student wants to know whether the payment was recorded.

If the purpose is to route the message to the right office, Payment status is the more useful label. The deadline explains the urgency. The payment record is what someone needs to check.

Before a model can learn from examples, people need to decide what those examples should teach.

2.1 What counts as data?

Data is recorded information that a computer can process. A student message, a photograph, and a row in a spreadsheet are all examples of data.

For the enquiry sorter, one example contains a student’s message and the category a staff member assigns to it. A collection of these examples is a dataset.

Student messageLabel
“What is the last date to pay the examination fee?”Deadline
“The money has left my bank account, but the portal says it is due.”Payment status
“I missed the deadline. Can I still pay?”Late payment

The message is the input. The assigned category is the label. When a model learns from inputs paired with labels, the approach is called supervised learning.

The sorter can work with the wording of an enquiry. Confirming whether a payment succeeded requires the actual payment record. The college can keep those private records out of the sorter’s training dataset.

2.2 Write a label guide

A label guide explains what each category means and helps staff handle unclear messages consistently.

LabelWhat the student wants to know
DeadlineThe scheduled date by which payment is due.
Payment statusWhether a payment they already attempted was recorded.
Late paymentWhether they can pay after the deadline or get an exception.

“The deadline was yesterday. I paid on time, but the portal hasn’t updated.”

The student wants to know whether a payment already made was recorded. Under this guide, the label is Payment status.

“My payment failed. Can I try again tomorrow even though the deadline has passed?”

If the college needs one label for routing, it could tell staff to follow the student’s explicit question. That makes this Late payment. If the college needs to track both issues, it could permit multiple labels.

Ask two staff members to label the same small batch independently. Where they disagree, ask what each understood the student to be requesting. Use those disagreements to clarify the guide before labelling a larger batch.

2.3 Collect examples of how people actually ask

A hundred messages that all say “What is the last date?” give the model a hundred examples of nearly the same wording. Students may also ask:

  • “How much time do we have to pay?”
  • “When does the fee window close?”
  • “Is Friday the final day for exam fee payment?”

All three ask about a deadline. Include varied wording, and if students commonly use several languages or mix them in one message, include that variety and test the system on it.

Include difficult cases too. “I paid after the last date, but the portal shows nothing” could be a payment-status question. More deadline examples may do little to help if the model keeps confusing late-payment requests with deadline enquiries.

2.4 Keep some examples for a fair test

Imagine a student gets every question right on an exam that repeats the practice sheet. You still need to see how the student handles a new question.

A training set contains examples the model learns from. A test set contains separate examples used to check how the trained model handles new inputs. If a student sends the same enquiry by email and through the portal, keep both copies on the same side of the split. Otherwise, the model could get credit for correctly handling a message it effectively saw during training.

Check results by category. Suppose a model gets all 18 deadline questions right and both late-payment questions wrong. Its overall accuracy is 18 out of 20, or 90%. On late-payment questions it scores 0 out of 2. Two examples are too few to estimate its failure rate for the whole category, but those failures tell you where to test further.

2.5 Try it yourself · Label the college enquiries

Use the three labels from Section 2.2. Mark a message Needs review if it gives you too little information to route it. This flag tells a person to look at the message.

  1. “When does the exam fee window close?”
  2. “My bank shows the amount was debited, but I still don’t have a receipt.”
  3. “I missed the final date. Is there another chance to pay?”
  4. “I paid before the last date, but the portal still shows the amount as due.”
  5. “I have a problem with the exam fee. Please help.”
  6. “Is tomorrow the final day to pay?”

Assign a label or review flag to each message. Underline the words that tell you what the student needs. Write one sentence explaining how you labelled message 4. Then set aside messages 4 and 6 as practice test cases. If a model labelled both Deadline, which prediction would you correct?

Check your work

Messages 1 and 6 are Deadline. Messages 2 and 4 are Payment status. Message 3 is Late payment. Message 5 Needs review. A useful follow-up would ask whether the student needs the due date, help with a payment already attempted, or permission to pay late.

Correct message 4. The student mentions the last date but wants to know why an earlier payment still shows as due. You have a small labelled dataset and a test case that exposes the weakness of relying on a keyword. A working sorter would need a broader sample.

Chapter summary

  • In supervised learning, the examples pair inputs with targets such as category labels.
  • Labels reflect the task the system is meant to perform. A written guide helps people apply them consistently.
  • Include varied wording and difficult cases. Leave out personal details the task does not require.
  • Test on separate examples and examine errors by category. An overall score can hide weaknesses that matter to users.

Review questions

  1. What is the input and what is the label in a training example for the enquiry sorter?
  2. Why might two staff members label the same message differently? How could a label guide help?
  3. Why should copies of the same enquiry stay on the same side of the training and test split?
  4. A sorter works well on deadline questions and often misses late-payment requests. What examples would you collect next?
  5. Write a message that mentions a deadline but should be labelled Payment status. Explain why.

Next chapter · How a Model Learns
How does a model turn labelled examples into a prediction about a message it has never seen?

What Makes an AI Answer Useful in Your Language? | AI Commentary #006



By Ravindra Dastikop

With assistance from ChatGPT

Imagine a parent asking a college assistant in Kannada whether their child can apply before receiving a marks card. The assistant replies in fluent Kannada and lists the documents required at admission. The family still has to decide whether to submit the application today or wait. The answer has missed the distinction between applying and completing admission.

What language benchmarks can tell us

The 2024 IndicGenBench study evaluates tasks such as translation, summarisation, and question answering across 29 Indic languages, including Kannada. AI4Bharat’s IndicVoices project collects read, spontaneous, and conversational speech across 22 Indian languages. The first gives teams ways to assess generated text. The second illustrates how much variety a voice interface may encounter in the way people ask for help. IndicGenBench paper · IndicVoices paper

Can the reader take the right next step?

My view is that an AI answer in any language should be judged partly by what the reader can do with it. In the admissions example, the assistant should check the official notice, explain which documents are needed at each stage, and link to the relevant section. If the notice leaves the question open, the answer should say so and point the parent to the admissions office.

When a translation may be enough

A checked translation of a clear notice may work well for routine questions. I’d stick with it if readers could reliably identify the correct next step. I’d change the wording or add an example if readers kept confusing application requirements with admission requirements.

Try it with one real question

Choose a question people often ask and an official source that answers it. Invite a few speakers to phrase the question as they normally would, including any English words they naturally use. After they read the AI reply, ask them what they would do next. Check their answers against the source and note where people hesitate or choose the wrong action.

What would you ask a reader to do to find out whether the answer actually helped?

AI assistance. ChatGPT helped locate and check the research sources, draft the commentary, and generate the cover illustration.

Wednesday, November 18, 2026

AI Made Simple, Chapter 1 | What Is Artificial Intelligence?

What Is Artificial Intelligence? | AI Made Simple, Chapter 1



What Is Artificial Intelligence?

Ravindra Dastikop with ChatGPT

At nine on Monday morning, three messages arrive in the college office.

“When is the last date to pay the examination fee?”

“I paid the fee yesterday. Why does the portal say it is pending?”

“Can I pay the fee after the last date?”

All three messages contain the word fee. A computer could send every message containing that word to the accounts office. That might save someone from sorting the inbox by hand. But the students need different kinds of help.

The first student needs a date from the current examination notice. The second needs someone to check a payment record. The third needs to know whether the college allows late payment and under what conditions.

Could a computer learn to tell these enquiries apart from examples that staff have already sorted? And once it recognizes a deadline question, could it give the student the correct date? Those are two different tasks.

1.1 What do we mean by artificial intelligence?

Artificial intelligence, or AI, is the field concerned with building computer systems that perform tasks such as recognizing patterns, interpreting language, making predictions, and generating content.

Your phone uses AI when it turns speech into text. An email service may use AI to estimate whether a message is spam. A writing assistant uses AI to produce a draft from your instructions. When you encounter a new application, ask what task the AI is doing here.

System Input Output
Speech recognitionA recordingA transcript
Spam filterAn emailA spam score or label
Writing assistantA request and supporting materialA draft
Student enquiry sorterA student’s messageA label such as deadline or payment status

The system receives ____. It produces ____. I can check the result by ____.

For speech recognition, compare the transcript with what you hear. For an examination deadline, check the official notice.

1.2 Following rules and learning from examples

Suppose the college asks you to write rules for sorting fee enquiries. You might label a message containing “after the last date” as late payment, one containing “last date” as deadline, and one containing “pending” as payment status. Apply the more specific “after the last date” rule first, since it also contains “last date.”

“I transferred the money two days ago, but the dashboard still shows that I owe it.”

A person would probably treat this as a payment-status enquiry. The message never says “pending.” You could add another rule, but the next student might write “the amount is still due even though I paid.”

Another approach is to collect past enquiries and ask staff to assign the correct category to each. A computer can use those examples to learn patterns that help it classify new messages. This approach is called machine learning. Fitting the model to the examples is called training. Once trained, the model can estimate which category fits a new message.

If nearly all the payment-status examples contain “pending,” the model may struggle with other phrasing. To find out, test it on messages it did not see during training.

Rules can work alongside a model. The college might use the model to identify a payment-status enquiry and then apply a rule that directs the student to a sign-in page.

1.3 Sorting a question and answering it

When a model labels a message deadline, the label tells the college what kind of information to look for. The system still needs the notice for the right examination session. If it selects an outdated notice, it can give a wrong answer even after classifying the question correctly.

A generative AI model can compose a reply from information supplied to it. Consider this fictional notice.

Examination fee notice · November session

The deadline for paying the examination fee is 24 November. Use the student portal to make the payment.

The model might reply, “For the November examination session, pay the examination fee through the student portal by 24 November.” You can verify the date, the session, and the payment method against the notice.

What if a student asks about late payment? The notice leaves that question open. A useful reply would say that the notice gives no late-payment policy and direct the student to the examination office for confirmation.

Check each claim against the evidence, and identify any part of the question the evidence leaves open.

1.4 A model is one part of a working system

A college website assistant could handle a general deadline question in five steps.

  1. Receive the student’s question.
  2. Find the current examination notice.
  3. Draft a reply.
  4. Check the reply against the notice.
  5. Show the reply and its source.

A student asking about their own payment needs an authorized way to check their record. A student asking for an exception may need someone at the college to make a decision. Someone also needs to keep the published notices current.

A model performs a task within a system. The system determines what information the model receives and what happens with its output.

1.5 Try it yourself · Design the first version

Set aside ten minutes. Use the fictional notice in Section 1.3 and consider these four messages.

  1. “When do I have to pay the November examination fee?”
  2. “Can I pay in cash at the college office?”
  3. “My bank shows the payment went through, but the portal says it is due.”
  4. “I missed the deadline. Can you approve my payment today?”

For each message, write down what the student wants, what information you need, and what you would tell them now. Underline each claim in your reply that you can verify from the notice.

Check your work

  1. State 24 November and specify the November session.
  2. Say that the notice directs students to the portal. It gives no information about cash payment.
  3. Identify the need for a private payment record and an authorized way to consult it.
  4. Identify the need for the late-payment policy or a decision from the responsible office.

You now have four test cases. Give the same messages to a prototype and check whether it answers, redirects, or asks for more information when appropriate.

Chapter summary

  • Identify the task, its input, its output, and how you would check the result.
  • Machine learning fits a model to examples so it can make predictions on new inputs. Rules can work alongside a model.
  • Classification assigns a category. Generation produces content.
  • A useful answer requires relevant evidence. Some questions also require authorized access or a human decision.
  • A working AI system brings models, information sources, rules, and people together.

Review questions

  1. Think of an AI application you use. What does it receive, what does it produce, and how would you check the result?
  2. Why might a rule that looks for “pending” miss a payment-status enquiry?
  3. The model correctly labels a question as deadline, but the assistant gives last semester’s date. Where would you look for the cause?
  4. The notice says nothing about late fees. What should the assistant say if a student asks how much extra they must pay?
  5. Write a new test question for the college assistant. What would a satisfactory response do?

Next chapter · Data, Examples and Labels
Which old enquiries should the college use as examples, and how should staff label them?

Who Can Stop an AI System After It Goes Live? | AI Commentary #005



By Ravindra Dastikop

With assistance from ChatGPT

When an AI system starts making harmful decisions, who can stop the next one? The answer may depend on who sees the problem, who has permission to intervene, and what service can continue once the system is paused.

Three dates in the Cruise case

  • 2 October 2023. A human-driven car struck a pedestrian in San Francisco, sending her into the path of a driverless Cruise vehicle. The Cruise vehicle made contact, stopped, and then attempted to pull over, dragging the pedestrian. Cruise began investigating. Cruise recall filing
  • 24 October 2023. California’s DMV immediately suspended Cruise’s autonomous deployment and driverless testing permits. Its permit for testing with a safety driver remained in effect. California DMV statement
  • 26 October 2023. Cruise paused operations of its driverless fleet while it assessed the underlying risk. Cruise recall filing

Who can pause what?

In this case, a vehicle needed to respond within seconds, a regulator could suspend specific activities in California, and the operator could pause its fleet. Each intervention had a different owner and scope. For any organisation deploying AI, my view is that oversight becomes meaningful when someone on duty can spot a credible problem, stop the affected activity in time, and put a workable fallback into effect.

A pause can cause harm too

Consider a college admissions office where a staff member finds that an AI assistant is giving applicants the wrong deadline. Taking the entire service offline would also cut off access to accurate answers. A narrower pause could disable generated deadline answers and direct applicants to the official calendar while staff check the source of the error. If the team cannot isolate the affected answers, a broader pause may be justified.

Run a five-minute drill

Use a mock incident and ask the person on duty to show what they would do. Who has authority to pause the affected feature? What happens to requests already in progress? What will users see? Who decides when to restart, and what evidence will they need? Record how long it takes to move to the fallback and whether a new request still reaches the paused feature.

If your AI system started giving the wrong answer at 6 p.m., who could stop the next one from reaching a user?

AI assistance. ChatGPT helped locate the source documents, develop and draft the commentary, and generate the cover illustration.

How to Assess Student Learning After AI Use | AI Commentary #004



By Ravindra Dastikop

Imagine an engineering student submitting a bracket design. An AI tool helped compare two layouts. The report recommends the lighter bracket, and the student can explain why it uses less material. Then the instructor moves the load farther from the wall on the sketch and asks, “Would you still choose it?”

That small change gives the student a chance to show how they reason. They might point to the mounting point, explain that the longer distance increases the bending moment for the same force, and identify the calculation they would redo before approving the design.

Ask students to defend one decision

When an assignment permits AI, my view is that students should be able to explain one consequential choice, show what they checked, and say what would make them reconsider. For the bracket, that means naming the assumed load and showing how they checked the original design. A polished report can document that work. A changed condition gives the instructor another way to see whether the student can apply it.

PO: The assignment begins after the student submits it.

Treat the submission as the starting point for one follow-up. Change a condition that matters to the decision and ask the student to show where the change affects their work. A sketch, an annotated calculation, or a short written response can make the reasoning visible.

Make the follow-up fair

A surprise oral question can reward quick speaking. Tell students in advance what you’ll assess and let them consult their own submission. In a large class, use a short written follow-up with an equivalent change for everyone. Give credit for a sound explanation and a sensible check, including when the student says they need more information before deciding.

Try it on one assignment

Ask each student to highlight one decision they made with AI assistance and attach the evidence they used to check it. After submission, change one relevant condition. Look for whether they can identify which part of the work needs revisiting and explain why.

What one condition could you change in your next assignment to see whether students can apply what they submitted?

AI Commentary #001 — If AI Does the First Draft, Where Do Beginners Learn?

AIYING · AI Commentary #001



By Ravindra Dastikop

Imagine a junior analyst reviewing an AI-generated sales report. The draft says a new onboarding programme drove an 18% increase in renewals. The analyst checks the spreadsheet. Renewals did rise, but the spreadsheet doesn’t show which customers used the programme. She changes one sentence before the report goes out.

That edit is easy to miss in a finished report. It’s also the kind of judgment the team needs her to develop. If AI writes the first draft every week, where will she get to practise it?

What the first draft used to teach

Writing a report from scratch forces you to decide what the evidence supports, what deserves the reader’s attention, and what you still need to find out. An AI draft arrives with those choices already made. A beginner who is asked only to polish the wording may never have to revisit them.

When a team automates drafting, it should deliberately give beginners practice making the judgments they’ll eventually be expected to defend.

A provocation for the team

PO: The first draft is a test written for the reviewer.

Take that idea seriously for a moment. What would the test ask? In the renewal report, it might ask the analyst to identify the claim that would most affect a manager’s decision, trace it to the source, and explain what further evidence would justify it. The published report could then say, “Renewals rose 18%. The available data does not establish why.”

When the AI draft helps

A blank page can leave a beginner guessing what a useful report looks like. A draft gives them something concrete to challenge, and a mentor can show why an apparently reasonable sentence needs revision. If a beginner can later spot the same kind of overclaim in a different report and explain it without a hint, that’s evidence that reviewing the draft taught something.

One thing to try next week

Choose one recurring AI-drafted report. Ask a new team member to trace its most consequential claim to the source and record one decision to keep, qualify, or remove it. Have an experienced colleague discuss that decision with them. On the next report, see whether they can find and explain a similar problem independently.

What did doing the first pass yourself teach you, and where would a newcomer learn that skill on your team today?

AIYING

Accomplish with AI

Powered by Blogger.

Featured post

Data, Examples and Labels in Machine Learning | AI Made Simple, Chapter 2

Ravindra Dastikop with ChatGPT The college office has decided to try an AI system that sorts student messages. Two staff members ...

Search This Blog

List of blogs

Search This Blog

About

Translate

Instant Stories

Read in Your Language

Test

share this

ShareThis

Popular Posts