The Fruit Farmer Activity

Part 1: Before You Play

In this activity you become a quality inspector at a produce warehouse. Your job is to grade tomatoes, and then teach an artificial intelligence to do the grading for you. Before you start, here are the big ideas you will be using. You do not need any coding experience; you just need to understand how machines learn from examples.

How does a machine learn?

Think about how a young child learns what a dog is. Nobody hands them a rulebook. Instead, an adult points at animals and says “dog... dog... not a dog... dog.” After enough labeled examples, the child spots the pattern and can recognize a dog they have never seen before.

Machine learning works the same way. Instead of being given strict rules, the computer is shown many labeled examples and figures out the pattern on its own. The branch of AI you will use today, learning from labeled examples, is called supervised learning.

Training data and labels

The examples you show the AI are called the training data. Each example comes with a label, the correct answer. When you grade a tomato “Grade A,” you are creating one piece of labeled training data: the tomato is the example, and “Grade A” is the label. The AI cannot see what you see or taste anything. It only knows what you tell it through your labels. This leads to one of the most important rules in all of AI:

Features: the clues the AI looks at

To make a decision, the AI breaks each tomato down into features, the measurable clues that might matter. In this activity the features are:

Choosing good features is a real skill. A feature like “day of the week the tomato was picked” probably would not help at all, while “amount of visible damage” clearly matters. The AI can only learn from the features it is given.

Classification: sorting into categories

When an AI sorts things into a fixed set of categories, that task is called classification. Your tomatoes go into three classes, following a real grading standard:

• Grade A

  Ripe, no blemishes, perfect shape. Sold at premium price.

• Grade B

  Minor flaws. Sold at a discount.

• Reject

  Unripe or badly damaged. Sent to compost.

Spam filters (spam vs. not spam), medical scans (healthy vs. concerning), and photo apps that tag your friends are all doing classification, just with different categories.

Training vs. testing

It is easy for an AI to memorize examples it has already seen. The real test is whether it can handle something new. That is why machine learning splits the data into two parts:

• Training set (12 tomatoes)

  The examples the AI learns from. You grade these.

• Test set (6 tomatoes)

  Fresh examples the AI has never seen, used to measure how well it really learned. The AI grades these.

If the AI does well on the test set, it has learned a genuine pattern rather than memorized answers. This is exactly how engineers check whether an AI is ready for the real world.

Where this happens in real life

This is not a made-up scenario. Companies like Driscoll's and Amazon Fresh use computer vision AI on conveyor belts to grade fruit at high speed. Human inspectors first label thousands of produce photos; the AI learns from those labels and then sorts produce far faster than a person could. The same label-train-test process powers AI that reads medical scans, inspects car parts for defects, and helps self-driving cars recognize stop signs.

Key vocabulary

Part 2: Fresh Sort: The Harvest QC Challenge

You're a quality inspector at a tomato warehouse. First, you'll grade 12 tomatoes yourself as Grade A, Grade B, or Reject. This teaches the AI your standards. Then the AI will try to grade 6 brand-new tomatoes on its own, and you'll see how well it learned from you.

Grade carefully and consistently: the AI only knows what you teach it!

Part 3: Real-Life Applications

Now that you have trained your own produce-grading AI, use what you saw to think about how these ideas show up in the real world. Answer in complete sentences and full paragraphs. There is often no single “right” answer; your reasoning is what matters.

Real-life scenarios to consider

• Example A

  The hospital scanner

  A hospital uses an AI trained on thousands of chest X-rays that radiologists labeled as "normal" or "needs review." The AI flags scans for a doctor to double-check. It was trained mostly on images from adult patients.

• Example B

  The resume sorter

  A large company builds an AI to sort job applications into "interview" and "pass." It is trained on 10 years of the company's past hiring decisions, which were made by human managers.

• Example C

  The self-checkout produce camera

  A grocery store installs a camera at self-checkout that identifies fruits and vegetables so shoppers don't have to type codes. It was trained on photos of produce taken in a bright studio.

Check your understanding

For teachers

Teacher resources

• Teacher Guide and Answer Key (opens in a new tab)
• Student Handout (opens in a new tab)

Google Docs. Use File > Make a copy to edit, or File > Download to save as PDF or Word.