How to explain AI to middle schoolers?

[sairoof]

Hi, I'm a middle school scince teacher
And would like to give a class about AI.
I have little experience with the subject so I would like to know how to make the concept clear for a middle schooler.
Also, I don't want to just include generative AI text and pictures but rather actually explain to them how AI works on a fundamental level.

 

[Vanadium 50]

"AI" is not a technology. At best, it is a collection of different technologies. and it worst it is marketing: Now, with extra AI!

 

[sairoof]

"AI" is not a technology. At best, it is a collection of different technologies. and it worst it is marketing: Now, with extra AI!

I guess I should have said machine learning instead?

 

[DaveC426913]

:following:
Because, if you figure it out, I'd like to sit in on your class!

 

[berkeman]

Have you already read through the Wikipedia articles? That is a good place to start. If you have questions about that reading, post them here so we can help out.

https://en.wikipedia.org/wiki/Artificial_intelligence

https://en.wikipedia.org/wiki/Large_language_model

 

[Andy Resnick]

Hi, I'm a middle school scince teacher
And would like to give a class about AI.
I have little experience with the subject so I would like to know how to make the concept clear for a middle schooler.
Also, I don't want to just include generative AI text and pictures but rather actually explain to them how AI works on a fundamental level.

What about AI do you want your students to better understand? Can you be more specific about what you mean when you say "how to make the concept clear" (bf added to the operative word)?

 

[sairoof]

What about AI do you want your students to better understand? Can you be more specific about what you mean when you say "how to make the concept clear" (bf added to the operative word)?

The schools here almost demand us to include ai in our classes
So I want to clear some things about it for them. Like what does the word AI actually mean, and how generative ai works. And the most important thing is machine learning.

Well writing this reply already gave me some ideas to what to search for.

 

[DrClaude]

There are good videos on YouTube where people explain how they used machine learning to teach a computer to play video games. I think the ideas are easier to understand in that case than, say, Large Language Models, and should be appropriate for the youths.

For example:

 

[berkeman]

In case you have not seen it, this (long) thread in the General Discussion forum has some good examples of problems with AI/LLMs, mostly in terms of inaccurate results (click on the up-arrow next to anorlunda's name to be taken to the thread):

I've been experimenting with ChatGPT. Some results are good, some very very bad. I think examples can help expose the properties of this AI. Maybe you can post some of your favorite examples and tell us what they reveal about the properties of this AI.

 

[Vanadium 50]

I guess I should have said machine learning instead?

You could have. But where do you draw the line between ML and plain old "statistics". And how do you you explain it to people unfamiliar with plain old statistics?

 

[PeroK]

The schools here almost demand us to include ai in our classes
So I want to clear some things about it for them. Like what does the word AI actually mean, and how generative ai works. And the most important thing is machine learning.

Well writing this reply already gave me some ideas to what to search for.

My approach would be to treat the whole thing as a research project, rather than pretend you know stuff you don't. Get everyone working on it and thinking about it.

You also have to decide whether you really want to teach how it works, rather than focus on its capability and applications. I suggest the former is ambitious. AI is out there already and it's likely to make rapid developments in the next decade. So, you are aiming at a moving target. I suggest the best you can do is give a snapshot of the current status.

 

[Dale]

I guess I should have said machine learning instead?

You probably should focus on deep neural networks. That is the specific class of algorithms that most people are referring to when talking about AI.

 

[Vanadium 50]

The schools here almost demand us to include ai in our classes

Great! Ask them for their proposed curriculum. For extra fun, ask them what to remove to make time for this. They'll scurry away and that will be that.

 

[sairoof]

Great! Ask them for their proposed curriculum. For extra fun, ask them what to remove to make time for this. They'll scurry away and that will be that.

it's not really a requirement but in every workshop or class (I'm currently taking a diploma in education) they remind us that it is the future and we should include it in our teaching.

 

[PhDeezNutz]

it's not really a requirement but in every workshop or class (I'm currently taking a diploma in education) they remind us that it is the future and we should include it in our teaching.

I think it’s a bit self righteous on their part to make this suggestion and you should the follow the advice in post #13.

Since they are ssssooo sure that AI should be included in the curriculum they should be able to tell you how. But somehow I don’t think they know themselves. They just wanted to sound cool.

How in God’s name do you explain Principal Component Analysis to someone who doesn’t know what an eigenvector is?

How do you explain bias variance tradeoff and cost functions to people who haven’t taken statistics?

Even linear regression requires a basic knowledge of statistics.

 

[sairoof]

That is exactly what is happening, but they won't really add anything to the curriculum because all of the material is borrowed and cheaply translated from English

 

[PhDeezNutz]

I guess I should try to be a little bit more constructive as opposed to cynical so here’s a suggestion.


Notice that there are now little vehicles powered by AI to deliver fast food on college campuses. This AI has to recognize objects (people, pets, crosswalks, stop signs etc).

So in order to “train” the AI you start with a bunch of pictures of things that are and aren’t said objects. The pictures you give it already have classifications attached (yes person, no person) to them and you partition the data to only train on parts of the data (randomly selected parts). The reason you train on parts of the data (build a model) is so you can test on the remaining data where tags are removed.
This is to avoid “overfitting”…..the more specific your training is to a given data set the less it will be able to adapt to other circumstances/situations. This is called “bias variance tradeoff”.

 

[PhDeezNutz]

Continuing on from my last post (because you’ve already seen it and I want to make sure you see the addition so I’m making a new post).

I remember being taught the “distance formula” in middle school. In particular the “Euclidean distance formula” but little do middle schoolers know that there are other valid ways to define “distance” (another popular way is called the taxi cab or Manhattan metric).

The notion of defining distance is very useful in classification problems (street lights, stop signs, people, etc). The idea is as follows

1) again you partition the data into training(tagged) vs testing sets (tags removed for now)

2) You chose a “distance formula” and you calculate the average distance from each classification.

- Given unknown object A
- Given known cluster of things classified as B
- Given known cluster of things classified as C
- Calculate the average distance from A to all the members of B
- Calculate the average distance from A to all members of C
- whichever distance is smaller A gets classified as that group

3) Object A was in the testing data, now you add back the tag and check the accuracy

 

[PeroK]

Continuing on from my last post (because you’ve already seen it and I want to make sure you see the addition so I’m making a new post).

I remember being taught the “distance formula” in middle school. In particular the “Euclidean distance formula” but little do middle schoolers know that there are other valid ways to define “distance” (another popular way is called the taxi cab or Manhattan metric).

The notion of defining distance is very useful in classification problems (street lights, stop signs, people, etc). The idea is as follows

1) again you partition the data into training(tagged) vs testing sets (tags removed for now)

2) You chose a “distance formula” and you calculate the average distance from each classification.

- Given unknown object A
- Given known cluster of things classified as B
- Given known cluster of things classified as C
- Calculate the average distance from A to all the members of B
- Calculate the average distance from A to all members of C
- whichever distance is smaller A gets classified as that group

3) Object A was in the testing data, now you add back the tag and check the accuracy

What has this to do with teaching AI?

 

[PhDeezNutz]

What has this to do with teaching AI?

I learned it in my ML class. Maybe not AI but it is ML or at least a statistical method.

 

[PeroK]

I learned it in my ML class. Maybe not AI but it is ML or at least a statistical method.

It seems rather specific to me.

 

[PhDeezNutz]

It seems rather specific to me.

It definitely is but I think it’s a good way to tie in to the curriculum of a middle schooler with the “distance formula”.

 

[Vanadium 50]

How do you explain bias variance tradeoff and cost functions to people who haven’t taken statistics?

These are middle schoolers. They are still working on division and fractions.

"AI is the future" is meaningless twaddle. Pouring that into kids heads is educational malpractice. If they want to develop an actual curriculum, fine, but this looks like a session of Buzzword Bingo.

 

[PhDeezNutz]

These are middle schoolers. They are still working on division and fractions.

"AI is the future" is meaningless twaddle. Pouring that into kids heads is educational malpractice. If they want to develop an actual curriculum, fine, but this looks like a session of Buzzword Bingo.

You’d be surprised to learn that (or maybe you wouldn’t) that many data science bootcamps are not much better. Expensive ones too.

 

[Vanadium 50]

Not at all surprised. There is money to be made in peddling twaddle.

I am not sating, BTW, that elements of what we now call AI will play an increasing role. They will. But the story will be more complicated than "One word. Plastics".

 

[BillTre]

I would think teachers would want to know about AI at some level, in defense of the Dark AI Arts on the part of students doing (not doing) their assignments.

 

[Greg Bernhardt]

An overview of the technology at a high level is worthwhile along with more focus on learning logic, critical thinking, and skepticism in terms of how to navigate living in a gen ai world.

 

[Swamp Thing]

CGP Grey has a couple of videos that the students can watch. Each student will take away more or less, depending on their prior priming and aptitude, but this will help you check the "did some AI in class" box.



I think he has another one more specifically on neural networks, or maybe this one is that one - I don't quite remember.


... And as someone commented below the video, you can observe a minute of silence for the student bots that didn't make it.

 

[ohwilleke]

Using a concrete example, like predictive text on a cell phone or when you are entering a search into Google, and then explaining how that is done with huge training databases and statistical weights of a likely next word is a good starting point. The key is to use something familiar to bridge to new concepts.