LambdaLabTM
Artificial Intelligence · Class 10 · Unit 2 · Advanced Concepts of Modeling
class 10 aineural networks⏱️ 9 min read

How AI Makes a Decision

Say you want to go to the park today. What goes through your head? You weigh a few things up and come to an answer. A perceptron — a single node of a neural network — does exactly that, with numbers.

1Start with the factors

Most of us would rather not get soaked. So the questions you might ask are:

  • Do I have a jacket?
  • Do I have an umbrella?
  • Is it sunny now?
  • What is the weather forecast for later?

These are the inputs. Each answer is yes or no, and we write yes as 1 and no as 0.

2Not all factors are equal

Some of those matter more than others. For most people is it sunny now counts for more than the forecast for later, and having a jacket counts for more than having an umbrella. That ranking is turned into numbers called weights.

Weights
A weight says how much that input matters. A heavy weight on an input means it pulls the decision hard.
Where do the weights come from?
Nobody knows the “real” weights. They come from experience — someone who has found that a sunny morning usually means a bright day will weight it heavily — and from personal preference. There is no right or wrong answer in choosing them.

3The bias

Bias
The bias is an extra input that is always fed a 1, with a weight of its own. It is the person’s general leaning, before any of the day’s facts are considered.

A cautious person sets a bias weight that pulls hard toward staying indoors, so they only go out when they are fairly sure it will not rain. A daring person sets a smaller one and goes out almost regardless. Here we are being cautious, so the bias weight is −4.

4What a minus sign means

That bias weight is negative, and the minus sign is doing real work. A weight can point either way:

  • A positive weight argues for the answer being yes. Sunny now has a weight of 3, so when it is sunny it adds 3 to the total and pushes you towards going.
  • A negative weight argues against it. It subtracts from the total and pushes you towards staying in.
The bias is the starting position
The bias weight of −4 is not about today’s weather at all. Because the bias input is always 1, that −4 is added every single time, before any fact is considered. It means: my default is to stay at home, and the day’s reasons have to add up to more than 4 before they talk me out of it.
Make the bias less negative and watch
A bias weight of −4 means you need a good day to leave the house. A bias weight of −1 would mean almost any excuse is enough. A positive bias would mean you go to the park even in the rain unless something stops you. Same four questions, same four weights, a completely different person.

5Add it all up

Multiply each input by its weight, add the bias term, and compare the total with the threshold, which is 0. Above the threshold, you go.

The formula
y = w₁x₁ + w₂x₂ + w₃x₃ + w₄x₄ + (1 × b)
If y is above the threshold, the perceptron fires and the answer is yes. Otherwise it is no.
Should I go to the maidan to play cricket?

Flip the answers and watch the total move. The decision changes the moment the total crosses zero.

The total is 0.5, which is above the threshold of 0 — so the answer is: go and play.
y = w1x1 + w2x2 + (1 × b)
y = (1.5 × 1) + (0.5 × 0) + (1 × -1) = 0.5
Friends goingw = 1.5
Sunnyw = 0.5
Bias — always fed 1, so it counts every timew = -1

An input that is 0 contributes nothing, however heavy its weight — that is the step most marks are lost on. The bias weight is negative: it pulls towards stay at home, and because the bias input is always 1, that pull is applied every single time before any fact is counted.Originally written as “go if the total is greater than 1.0” with no bias. That is the same machine as the bias of −1 shown here with a threshold of 0 — and the board asks it in the bias form.

You will also see it written without a bias
Plenty of books and slides write a perceptron as “add the weighted inputs up and fire if the total beats 1.0”, with no bias at all. That is not a different machine — a threshold of t is exactly a bias weight of −t with a threshold of 0, because subtracting t from the total and comparing with 0 asks the same question as comparing the total with t.
The cricket example above is written both ways: its bias of −1 is the same rule as “fire above 1.0”. Answer in the bias form, because that is the form the board prints.
The two cases from the handbook
I have a jacket, no umbrella, it is sunny, rain forecast. The total comes to 0.5. That is above 0, so: go to the park.
No jacket, I have an umbrella, not sunny, forecast ok. The total comes to −0.5. That is below 0, so: stay in.
Try both in the widget above.

A pair of scales

Each input that is switched on drops its weight onto one pan. The bias is a lump already sitting on the other pan before you start. Whichever way the scales tip is the decision — and an input that is switched off drops nothing at all, no matter how heavy it would have been.
The mistake that costs the mark
When an input is 0, its whole term is 0 — you do not add the weight in. The bias is the opposite: it is always multiplied by 1, so its weight is always counted.

6Why people disagree

There is no correct set of weights. Two people looking at exactly the same sky can pick different weights and a different bias, and so reach different decisions. That is the whole reason everyone does not do the same thing on the same morning — and it is also why two AI models trained differently can disagree about the same data.

7From the board papers

How this is asked
The 2025–26 sample paper states it plainly: questions related to Perceptron will be asked in objective (MCQ) format. You will be given a small table of inputs and weights, a bias, and the formula — and asked for the value of y. It is arithmetic. Do it slowly.
1
A company screens job applicants with a simple AI model using four attributes, each with a weight. A bias value of 0.3 is added. The formula is y = w₁x₁ + w₂x₂ + w₃x₃ + w₄x₄ + (1 × b). What is the value of y?CBSE sample paper 2025–26 — the paper supplies the table of values
1 mark
▶Show the answer

The method, which is what you are being marked on: multiply each attribute’s value x by its weight w, leaving out any whose x is 0, add those products together, then add the bias — here 1 × 0.3 = 0.3. The sum is y. Add the bias exactly once, and never multiply it by anything but 1.

2
Convert this scenario to a perceptron: a manager is deciding whether to approve a work-from-home request. Factors: does the employee perform well remotely; are there upcoming team meetings; does company policy support remote work; is it beneficial for both sides?Class X handbook, Reflection Time
2 marks
▶Show the answer

Draw four inputs x₁ … x₄, one per factor, each answered 1 for yes and 0 for no. Give each a weight w₁ … w₄ according to how much the manager thinks it matters — company policy would normally be the heaviest, since nothing else counts if the policy forbids it. Add a bias input fixed at 1 with its own weight, standing for how willing this manager is in general. Sum w₁x₁ + w₂x₂ + w₃x₃ + w₄x₄ + (1 × b) and compare with the threshold: above it, the request is approved.
There is no single correct set of weights — a different manager would choose different ones, and that is the point.

8Check yourself

Quick Check

In a perceptron, what does a weight represent?

Quick Check

What value is the bias input always given?

Quick Check

Is there a single correct set of weights for a perceptron?

Quick Check

An input answered 'no' contributes what to the sum?

← Previous
🕸️ What is a Neural Network?
Next →
What is Evaluation? 📏