初级Beginner点子Idea
教电脑认东西：训练数据会“带偏”
Teaching a Computer: When Training Data Misleads
只给电脑看过黑猫，它会怎么认白猫？AI 学到的是数据里的规律，不一定是我们想教的。
If a computer has only seen black cats, what will it make of a white cat? AI learns the patterns in its data, which may not be what we meant to teach.
这是一个课程点子：活动和资料都有，学习单还没做。
This is a lesson idea: the activities and links are ready, the worksheets are not made yet.
如果你只给电脑看过黑猫的照片，它会怎么认一只白猫？
If you only ever show a computer photos of black cats, what will it think of a white cat?
孩子做什么What kids do
- 分卡片：给一组动物卡片，只说“这些是 A 类、那些是 B 类”，不说规则。学生自己猜规则，再拿新卡片考对方。Sort the cards: give out animal cards, say only “these are A, those are B”, and don’t say the rule. Kids guess the rule, then test each other with new cards.
- 故意“带偏”：A 类卡片刚好全是绿色背景。再拿一张红色背景的 A 类卡，会被分错吗？Mislead on purpose: all the A cards happen to have green backgrounds. Now show an A card with a red background. Does it get sorted wrong?
- 在教室电脑上带全班玩 Code.org 的 AI for Oceans，或者用 Teachable Machine 现场训练一个小模型。On the classroom computer, play Code.org’s AI for Oceans together, or train a small model live with Teachable Machine.
要点Key idea
AI 学到的是数据里的规律，不一定是我们想让它学的规律。数据偏了，判断就会偏。
AI learns the patterns in its data, which may not be the pattern we wanted. If the data is lopsided, so are its answers.
和 AI 的关系Where AI comes in
这是今天做 AI 的人最头疼的问题之一。也可以讨论：人的偏见是不是也是这样来的？
This is one of the hardest problems for people who build AI today. You can also ask: do people pick up their biases the same way?
视频与文章Videos and reading
- Code.org: AI: Training Data & Bias 3 分钟3 min
- Code.org: AI for Oceans 课上一起玩play it together in class
- Google: Teachable Machine 现场训练一个小模型train a small model live