← Back to course
AI Explorers 6-8 / Week 01 / Wednesday
3/6
Week 01 · What Counts as AI?

Wednesday

Explorer Lab: The autonomy dial
// Objectives, outputs and autonomy
⏱ about 20 min

Wednesday: Explorer Lab: The Autonomy Dial

Mission control sends a short message: "How much should Sky Sorter do on its own?"

Rocket wants it to file every card instantly. "Speed is the whole point," he argues.

Raven shakes her head. "If it files a comet as a star, nobody will notice until much later."

Nova draws a dial on the whiteboard with three settings: Suggest, Decide and Check, and Decide Alone.

"Instead of arguing," she says, "let's test all three settings with real cards and count what happens."

Rocket grabs a stack of index cards. Raven grabs a timer. The lab is open.

Your mission

You will play Sky Sorter three times, once at each autonomy setting.

A partner plays the person who works with the sorter. You will count mistakes and checks at each setting.

Then you will compare the trade-off between speed and catching mistakes.

  • 12 index cards
  • A pencil and colored pencils
  • 4 sticky notes or paper labels: star, planet, comet, satellite streak
  • A clock or timer that shows seconds
  • A partner (a family member or classmate)
Safety first
Use only paper, pencils and cards. No device is needed except a simple timer.
Carry pencils point down and clean up the cards when you finish.
  1. Your partner secretly draws 12 snapshot cards: 3 stars, 3 planets, 3 comets and 3 satellite streaks.
  2. Your partner writes the true label on the back of each card, then shuffles them.
  3. Make two or three cards tricky on purpose, such as a very faint comet or a short streak.
  4. Setting 1, Suggest: for each card, say your prediction aloud. Your partner decides where it goes.
  5. Setting 2, Decide and Check: file each card yourself. Your partner checks only every third card.
  6. Setting 3, Decide Alone: file all the cards yourself with no checks at all.
  7. Time each setting. After each one, flip the cards and count mistakes left in the bins.
SettingWho makes the final call?Time (seconds)Mistakes left in bins
1. SuggestThe person
2. Decide and CheckThe sorter, with some checks
3. Decide AloneThe sorter

Copy this table onto paper and fill in your own times and mistake counts.

PREDICT BEFORE YOU PLAY
  • Read the question.
  • Tap your answer.
Which setting do you predict will finish fastest?
Which setting do you predict will leave the fewest mistakes in the bins?

Rocket and Raven's results

Here are the crew's own numbers from the same lab. Use them to answer the questions below.

SettingTime (seconds)Mistakes left in bins
1. Suggest960
2. Decide and Check581
3. Decide Alone313
How many seconds faster was Decide Alone than Suggest?
How many more mistakes were left at Decide Alone than at Suggest?
Which setting was in the middle for both time and mistakes? Type its number.
What the crew's results showTrue or false?
More autonomy made the sorting faster.?
More autonomy let more mistakes slip through.?
The Suggest setting had no trade-offs at all.?
A person in the loop can catch mistakes before they matter.?
WHY THIS EXERCISEEvery autonomy setting trades something away, so designers choose based on the objective.

Great lab work. Tomorrow you will tell AI systems apart from fixed programs.

← Tuesday