Jan 22, 2020
In this special Hour of Code episode Dan and Ray walk through a
Minecraft resource to help you (and your students) understand some
important AI principles and how they can be applied when fighting
bushfires. We will look at the use of teamwork, automation
and image recognition as well as highlight some computer science
principles used.
The Hour of Code discussed here has a full set of curriculum
resources, and all download links, at https://aka.ms/hourofcode
There is a video to accompany this podcast for anyone wishing to see the Minecraft activity live. This is available on YouTube at AI in Education Podcast - Hour of Code Special in Minecraft Education
In addition to coding and AI topics, the topics covered in the Educator Guide include:
This activity can be completed with Minecraft Education Edition, even if you don't have a subscription
________________________
TRANSCRIPT For this episode of The AI in Education Podcast
Series: 2
Episode: 3
This transcript was auto-generated. If you spot any important errors, do feel free to email the podcast hosts for corrections.
Welcome to the AI and Education podcast. I'm Dan.
And I am Ray. And today we are going to experiment with this
podcast. Let's be honest, Dan, for the last 16 17 episodes, this
has been one big experiment. We've been learning as we went along
about podcasting. Yes.
About conversation,
correct?
All those other skills that I've always wanted to be able to do
with my life.
Storytelling.
Yes. And today we are going to experiment on something that we're
not sure if it's going to work. But hey, join us on another part of
our journey around AI in education.
So what's the theme today? Right.
Well, look, I've been thinking, Dan, we are based in Australia.
We're based in Sydney. It's been a truly awful spring and summer in
Australia for bushfires. I know not just in Australia, but globally
the media have been covering what's going on. And I'm very aware
that at the end of January, early February, there's going to be a
million students in Australia that are going to head back to
school. There's going to be students all around the world for whom
this is a topic and bushfires, you know, how do you have that
conversation, but also how does that conversation link into other
things happening within the school? And so I is really excited to
discover that the hour of code that we did last year, so in
November 2019, was actually linked to the whole idea of forest
fires and artificial intelligence. And so I thought we should do
the hour of code. Now, this is going to be a bit weird, Dan,
because
I've got you set up with the Hour of Code. It's on screen in front
of you, and most people are just going to be listening to the
podcast and not seeing it, although they could play along by
pressing pause and going downloading Minecraft Education Edition
right now. But um I I thought let's have a conversation about how
it might link into an activity that you could do as a computer
science or IT or any other kind of teacher within a classroom with
a group of kids in, you know, year seven or eight or upwards and
how it links to things that live in their world, which is things
like Minecraft and how it links to things that are happening in the
real world around bushfires and climate change.
And this did come from our AI for Earth project, didn't it?
Yes. Yeah.
So, so there's a lot of AI in it and it is computer science
based.
Yeah. So, Arrow code is all about getting uh students that haven't
coded before to understand coding and this year it actually came
out the air for earth project and it wasn't driven at all by the
bush fires in Australia but it just happens to be all around you
know forest fires and things like that so should we give it a
go
yeah so um if you've pressed pause and you've downloaded Minecraft
Education Edition then sign into it using your education email
address everyone in Australia I know in schools will be able to
just log straight in
and we'll we'll supplement this with a video of of us recording
this live. What do you think?
Let's put it on YouTube, Dan.
Yeah. While you're
Okay, we'll put the link in the in the show notes. So, uh, for
everybody that's playing along at home, then click play when you're
in Minecraft Education Edition and then hop into the library. And
in the library, you'll find something called lessons. And in there,
there's lessons on lots of different subjects, but I want you to do
the computer science ones and the arrow code. So, Dan, you're now
actually going to take the Arcode lesson live. Now, the good news
is you're a faster typer than most people. And so it's not going to
take us an hour.
So what what has just said to me and this is you know for the
conscious that this is a podcast so you listen to this in your car
or whatever. So the the concept around this um is that there's a
village threatened by fire that needs your help. Um you've got a
program you are coding assistant which is an agent to navigate the
forest and collect the data about the fires. Then write code to
help prevent the spread of the fire, save the village and bring
life back to the forest. So it helps you learn coding basics, talk
about AI and also kind of think about the kind of um way technology
can support in these uh particular projects I suppose.
Great. So click the create world and it'll go off and create it.
And then as we're going through Dan, let's talk about how it links
back to the AI concepts we've talked about in the podcast and how
it links back to some of the situation that we've been dealing with
and probably still are dealing with in Australia around fires.
Okay. So when we're talking about just generally while we're
waiting for this to load up, you know, it has been a horrendous
couple of months in Australia and when I've been looking at the
media being local here, you know, there's a couple of interesting
technological things that have come up and I know this isn't the
time for it in the podcast really, but there's been some really
interesting uses or or not uses of technology. I suppose it's been
a lot of analog stuff that's actually had to happen. Um hopefully,
you know, there will be things that some of the technology
companies do over time to kind of support with with things like
bush fires. You know, I didn't see a lot of drone technology. I may
be wrong, but on the news it was very kind of uh manual based
process and actually being in amongst the fires over New Year like
in between states. There was an interesting uh use I I had the
Victoria fire app and the New South Wales fire app because I was
right on the border in Threadbo in the snowy mountains there and it
was really interesting for me to see just using that data and the
visualizations that were coming through and some of the AI and
predictions, you know, when they were telling us to kind of
evacuate or not or what, you know, they were having to predict
using their models. It was quite interesting to see the use of data
between states. So, it's quite glaring for me that, you know, it's
quite of obvious gap there.
Yeah. So, we're in the environment now, Dan. So, I want you to just
go up and start talking to the people it's telling you to talk to.
So, it's going to tell you to talk to a researcher in a minute.
Okay.
And you talk to that researcher and the researcher is going to tell
you to do a task.
Okay. She seems Welcome to the research center. You you will learn
how to how we use AI and coding to prevent fires. Your first task
is to open the gate and press C.
So C opens the coding window in Minecraft. And then what you're
actually going to do is build a little bit of code to open the
gate, which I think is from my memory of having done this is going
to be quite simple. So no need to get nervous or sweat just yet,
Dan. Um but just on that point around the using multiple uh apps to
be able to see the fires, it is a bit like in institutional stuff.
In order to be able to do it, you've got to go to that department
and get that data and that department and get that data. In our
case, it's the same data from two different states. Well done, Dan.
You've opened the gate. Y
because the code block simply said open gate. That made it easy.
Talk to the next person who's an engineer.
Okay. So, the engineer's now said well done. You can program a
gate. Uh code the agent, which is a little person inside or little
robot inside Minecraft to move three steps forward. and stop on the
marker.
I think we're still within your skill capability here, Dan. The
point about what what is happening at this point is you're just
learning some of the simple steps. We're not doing AI. We are just
learning what coding looks like. Like when in the old days of logo
where you used to move the turtle forward and then the BBOT where
you got to move the physical turtle on the ground in front of you.
It's learning those things to be able to go move forward three,
turn right 90, move forward three. You know, those really basic
steps are really important because it's about learning the
instructions.
Yes.
So Dan, you just work on the code to work that move the agent
forward. Now what is really interesting is the conversation that
came out in the bushfires in Australia is well how do we get data
that is genuinely open data that can be set shared across
organizations rather than simply having data locked in an
application. And that's pretty similar to education if you think
about it. We've got student data locked in different systems there.
It's in mathletics. It's in Edodtoodo, it's in the learning
management system. And to be able to use it in an AI way to be able
to apply intelligence to it, you want to bring all those things
together. So it's exactly the same um as we experienced in the
context of bushfires, exactly the same in education as well. It's
joining the data together.
Well, I miserably um messed up the first attempt to move the agent
forward by three. I didn't read the instructions properly and um
now I should.
So you moved it forward one block and it told you you'd failed.
Yes, it did. I see. So I've moved that agent forward. So I've moved
something, you know. So keep moving through the world. So what Dan
is doing is actually moving through a Minecraft world.
A fireman.
Fantastic. Oh, firewoman. Brilliant. So the agent now needs to
collect data about flammable materials. So that's interesting. So
it's making me think about um what could pose a fire hazard. So
it's more than just the coding. It's making me think about what
kind of material aterials might be uh so it says code the agent to
move four spaces forward towards the dry bush and then analyze
forward to collect the data.
So this is where the AI stuff starts to come into it. So already
we've learned how to do some coding steps but now the next step is
okay well how do we collect the data that we need? So if you go
through this process it's about being able to do the analytics. So
if you think about all the conversations we've had about AI and
machine learning
y
and and the need to be able to train a system to recognize what a
good result or a bad result looks like. That's exactly what you're
just about to do here.
Okay.
So, uh remember Dan, you're moving three spaces forward just so you
don't
Absolutely.
Okay.
Or was it four or was it four spaces of this one?
We're going to find out.
Yeah.
Okay. So, at this point, Dan is just moving the blocks into the
right space.
It was four. Ray, you duped me. Oh, dear me.
Okay, I thought they might be four. Yeah.
Um, interestingly, when I'm looking at this, you know, it's it's
also Oh, gosh, what have I done now?
I'm hoping Dan that people that are doing this with year seven kids
are going to find this easier than you your kids are doing it.
Um, uh, the the interesting one when when when I'm thinking about
this, you know, obviously this is really great concepts, but then
you think, well, you know, what cognitive services could you use um
to kind of drive some of these in, you know, if I wanted to
automate a lot of these, it's asking me to go to find somebody
else. Now I've completed that one, but um I can now uh find
somebody else.
You climb through the ladder and go to the top of the tree. Yeah.
So, exactly right. So, you think about the cognitive services we've
been talking about like being able to analyze an image. You could
um take a project where you say to give it two different types of
images. So, here's dry bush, here's wet bush, and train it to be
able to rec recognize the different types so that then you can
point a camera at something and it and it's recognizing it. So, you
know, most often we talk about that, I think, in the AI for Earth
context around wild animals. Um, for example, using camera traps
where what they're able to do is quickly analyze camera trap
photos. So, camera traps are the ones you, you know, stick to a
tree or whatever analyzing which animals come to a water hole. They
take thousands or tens tens of thousands of photos over the course
of a day, a week or a month. And what you need to be able to do is
very quickly analyze that and say, "Okay, is there a leopard in
this photo? Is there a lion in this photo and 90% of photos have
nothing in them? And historically, they they were done manually."
Now they can be done automatically. And they're great examples of
things that students could do as projects where you use the
cognitive services to be able to both train a model to say
recognize a line, recognize a leopard, lab recognizer, giraffe, and
then take a whole series of images and do that analysis piece. And
the task that a student could do in that is exactly the same as you
would do as a researcher staring at 100,000 images of a mountain
side in Tibet looking for snow leopards.
Yeah. And and what I'm doing at the minute now is actually trying
to um uh navigate through a maze. It's asking me to It's It's the
automation, I suppose, you know. What what what this is making me
do is think about the steps that um I'd have to do. I'm moving an
agent through a maze to find um dry grass that I'd need to um uh
identify essentially. So hopefully
Yeah. It's going to be something bit like a maze, hasn't it?
Yes.
Yeah. And and so you're defining the steps to go through that maze.
I notice you took the easy option, Dan.
Absolutely. Cuz it's a podcast and I don't want to bore everybody
online.
So is basically programming the agent by saying, "Move forward.
Move
turn left.
Turn left. Turn right.
Turn right.
Turn right.
Move forward, turn left." For those that are watching this on
YouTube, you'll spot that Dan got his left and right completely
confused. Uh then move the agent more towards the end and then
analyze the objects at the end.
Yeah. Here we go. Is it going to work?
Congratulations. Fantastic.
Okay. So now you've learned some steps which is the analysis of
things but also then
the ability to actually do some analysis. So it's it's then you've
kind of got some great steps. The next bit of the journey then is
to think about the data. So Dan is moving through the world a
little bit more. He's got to go and find the head of research.
Where's this head?
Well, you're going in the right direction, Dan. But now you've
actually got to guide yourself through the maze.
Oh, dear me. Can I jump? Oh, no. I can't. Okay, I'll guide myself
through the maze. This brings out my uh my ex Minecraft. I was
going to say I thought you were a Minecraft wiz.
I was a Minecraft wiz. Um anyway, there we go. Right. Look, get me
out here. I'm a celebrity. Get me out here. Fire Research Center
this way. Wow. It is It is actually really educational for me
actually. It's taking me to these people. They're giving me more
information. I've now found the head of research. Um it says,
"Teach the agent to predict fire hazards is our best tool for
prevention." So, teaching some something like AI to predict fire
hazards is the best prevention tool. come inside the A and the AI
specialist will assist you on training the agents. I'm going to
speak to an AI specialist now.
Okay. So, this is where the conversation is going to move from hey
Minecraft and games to Yeah. But what does AI what is using AI
means because one of the things that we know about AI is it has to
have previous data in order to be able to train it for the future.
So, we talk that talk about that as labeled data. So, for example,
you've got um a good outcome for a student or a bad outcome for a
student and knowing what that means. That might mean that a pass
rate is over 90 and a fail rate is below 80. Or it might be that
somebody attended for 90% is good and somebody attended for 80% is
bad. Or it might be here's an image. This image is labeled as a
lion because that's what a lion looks like. This image is labeled
as a zebra because that's what a zebra.
And this is this is now prompting me. So I found the AI specialist
and they saying to me that there's a computer in front of me
that'll help teach uh the agent about plants that are fire hazards.
So, it's about machine learning. It says, "Watch the images on the
monitor. If a plant looks yellow and dry, it's a fire hazard. So,
press the yes button. If the plant looks green and healthy, it's
not a fire hazard. So, press the no button." So, it's a way of me
teaching and training the computer in front of me to understand um
uh AI and learn, I suppose.
Yeah. So, you're actually an AI programmer at this point. You've
been given a series of images. So, what Dan is looking at now is
images of different types of fire hazards.
Yeah, that's a dry plant. I'm going to press that one and I got
that right.
So, they show me a green piece of vegetation.
Is a green piece of vegetation combustible or not? No,
got that right. About three out of three so far.
So, this is exactly the same thing as you do.
Yes.
When you you know those bits
I train the AI
that that bit where you the um your filling out a form and it says
it needs to check you're a robot or not a robot.
Yes.
And uh one of the things it does is says look at this image and
tell me where the cars are or tell me which of these frames has got
a traffic light in it. That is labeling data. What you're actually
doing is training cars to become self-driving cars because you're
training it what a traffic light looks like. I know you mentioned
that in previous.
So this is exactly the same. Now this is more relatable to students
because it's built within this context and it's much easier setup
but it's exactly the same as what advanced scientists are trying to
do with self-driving cars, which is they need to know what the
world around them looks like. In this case, you're doing bushfire
hazards, in the case of self-driving cars, it's what things around
the world look like.
Yeah. And I I know from a scientist, and they've said to me, to
successfully prevent fires, you'll need to remove dangerous plants
from the area. Now, let's put your training to the test. Code the
agent to destroy all the plants that pose a fire hazard. Exit the
chat window to start coding. Fantastic. So, I've use the agent to
start to destroy things that might not be conver to the agent
now.
Yes, I'm going to do that right now. Ray, and for everybody
listening on the podcast, this is um you know, just while we're
waiting for that to load up. It's really interesting, you know, on
several levels as a person going through this the first time, you
know, this is the first time I've done this um as as somebody doing
it. I can imagine if I was a student or even a a teacher taking my
students through this. I'm I'd say that if they're about seven and
above, they'd probably be able to cope with this quite easily. Um
it really does make you think. Um and and it also it it's also it's
quite educational in terms of the actual way that um I'm actually
learning about different types of combustible material and it's a
lot of quite automated. So that's fantastic.
Yeah. So now you're presented with a lot more complex coding
toolbox, but basically it's what it saying here is it's going to
pop up things in front of you. You need to look at them and if it's
dry bush that's in front of you then you need to destroy it.
Yes. So let's try to do that.
Um so let's see what happens. Yeah. Great.
Okay. So now what's happening is
things are popping up in front of the agent. It's either destroying
or ignoring them.
So that's using a in terms of computer science it's using um a
while loop now. So it's it's using more loop in structures so that
you can repeat things. So like with AI when it starts to kind of
iterate.
Yeah. So you've got basic coding going on there, but then you've
got the the decisions it's making are based on the AI programming
that you did earlier to tell it to recognize um good and bad
things.
And I've got to speak to an analyst. It's quite good in terms of a
role as well because you actually you're actually seeing you know
technical people. You're seeing all sorts you know males, females a
I'm kind of always quite fussy about that. Making sure you get an
equitable amount of uh males and females in these in especially in
the science area here. So there's lots of female scientists as
analysts. I've spoken to an engineer, firewoman, a fireman.
Now you're So tell us what the analyst wants you to do.
The analyst wants you to use using images. The agent can now detect
where the fires are likely to start. Code the agent to alert the
team if a potential fire hazard location is found on a computer
monitor. Wow. So the agent is looking at a computer monitor
somewhere.
It's behind you.
It's behind me. There's the agent. And I'm going to press um C to
code.
So, I'm going to
I'll let you do that coding in in silence. And I'll just talk about
So, this is really interesting because we're we're now applying
this to Oh, we'll look at something on a monitor and make a
decision about it. So, we're taking exactly the same thing as we
did on the last step, which was an agent standing in front of real
that's in air quotes real uh brush. But now, in this case, you're
looking at computer screen. So you were talking about the
conversation earlier about oh when are we seeing AI in use in
drones and things like that in firefighting.
This is a really good example because if you had a project which
was taking aerial photos and then being able to look at that land
use
um from an aerial photo to be able to go well here's where the
forest cover
and he's looking now at some of the aerial photos that's uh running
past and it's actually um highlighting with red stone highlighting
little light to say when they find um land that's fire hazard. So,
it says find the helicopter and talk to the fire captain.
This is really exciting. Now, Dan, you've you've finished. You're
now going to go and take a helicopter to a trip and do something.
No way.
So, continue working through your world.
Oh my word.
Okay, there's your Minecraft helicopter. That's fine. And have a
chat to the captain.
I can't believe how quickly the agent learned to predict the fires.
We need to hurry to this location at risk of fire. Let me know when
you're ready to go.
So, this is great. Students are used to this kind of thing. You're
now in a different part. of the Minecraft world. Can you see on the
ground in front of you down there's your agent.
Oh, wow.
Okay. So, have a chat to the to the fire captain to find out what
you want that agent to do.
I was just going to do it without talking to him, which is which
illustrates bad use of AI. I was just going to give it a go. Um, I
just got a message from the command center. There's a storm rolling
in with a high chance of lightning. We need to start clearing this
area of dry brush before it gets here. Have the agent clear that
line of dry brush to remove the fire risk.
So, the agent standing in front of a a line of brush. And what you
need to do is build the code using all the skills that you've
learned so far, Dan.
Oh, enough
to be able to move forward, recognize what's in front of you,
decide it's a fire risk or not, destroy it if it's a fire risk,
leave it if it isn't, and just get that to go across. Now, that is
drawing together both basic coding skills about moving uh items
around, but it's also bringing in this AI stuff that we've done
already, which is we're using the programming that the students
have done earlier about programming the bot to be able to recognize
dangerous and non-dangerous brush. So, you know, it's it brings the
whole thing together. We have gamified the whole process, but we've
put it into a context that we can now understand artificial
intelligence. We can understand programming and training artificial
intelligence uh into independent agents. We are doing it in a
gamified world. So, you've now programmed the bot. You've hit
start. So, what it's doing doing is it's moving forward in a
straight line
and I didn't do it.
For those that are seeing this on YouTube, Dan failed again because
he forgot to
I if I put the agent move outside the detection one, then that
should work. I'm just I'm doing this very quickly without reading a
lot of it. There we go.
Yeah, there you go. So now the agent is moving.
It's my rudimentary use of a wild doom.
It's actually destroying objects leaving the uh good brush
down.
Oh, f***. Now, uh, this is the last bit of the
fire. Watch out. Oh no.
Oh, the fire started.
Yeah, lightning has struck. The fire is marching towards the
helicopter and your agent is standing right in the middle of the
fire line.
But of course, what you've done is cleared.
Oh, that's fantastic, isn't it?
So, look, we're able to talk about this in the cont So, uh, the
fire stopped at the line that Dan had cleared. So, we're able to
talk about that now in the context of artificial intelligence, the
need to program, having training data uh coding in terms of the
steps that you're able to do um in order to be able to get an
object to a place or the even just the if then else type of steps
to say okay this is what you do but also we're now able to have a
conversation about how might you use drones in fire zones how might
you use
image recognition
image recognition
but also you know the the general conversations in the in the media
about hazard reduction burning and building um He's clicking. Ray's
clicking his fingers. You guys podcast. Brilliant. Fire break. The
the other interesting one as well that's coming out. Dan, hold
on.
No, no, no. Keep that in.
The the other thing is also building fire brakes. You know, that's
a real life conversation that you can then link back to something
that's happening in a digital world. And that's why I love this
activity when I took part in it because it's a way of being able to
link the theoretical conversations we have about technology and
artificial intelligence to a real conversation. right now that
students would be having that makes complete sense in terms of the
digital world connected to the physical world.
Yeah. And and at the end right at the end here, you know, it's not
an AI thing, but it does say I suppose it could be AI. It says
while you're here, why don't you help restore the area by planting
flowers and spawning animals and immediately I'm thinking like I
need to find out which areas to do that. And I could use the same
technologies that I, you know, you know, obviously just wants me to
plant stuff, but I could use AI to work out well where should I
plant I did see something on um on social media last week actually
where they did a study. I forget who it is. I forget who the they
is in this conversation. I don't like talking about things that I
can't back up but there was an image of where people should be
planting fire uh trees sorry uh globally which is quite interesting
you know around certain regions to kind of bring the planet's
carbon uh back into check. But yeah this really interesting I
really enjoyed that activity right really introduced a lot of
concepts to me and Um, you know, I think kids would really enjoy
that.
And I think what is really interesting about this is it's much more
relatable to a group of students, say in a year seven or a year 10
class than some of the case study material that they can read at
the minute, which is around things we're doing around wildfire
detection, things we're doing about wildfire spread. So, we've been
doing some work started about eight years ago in Greece, uh, when
they had big wildfire problems there about being able to build
better models that predicted the spread of wildfires around the
mountains. So looking at topography and things like that, that's
great stuff, but a student probably couldn't build a model around
that. Whereas this is designed for students to be able to use it
right here, right now. But you can imagine what we've just done in
27 minutes is at least an hour's activity with students.
I get I just got a certificate that I could download at the end
which I've just done as well. And there's also um some other
computer science fundamentals as well that you can take and there
are um lesson plans just not this.
Yeah, there's a really good lesson plan that goes with this, but
it's a way of being able to relate to a student. But of course, it
could take more than an hour if you decide to go and take the
examples of what they're doing and relate it to a conversation
around what's happened in the bush fires within Australia and the
whole debate about hazard reduction burdens and things like that.
You could put all of this into that context. So, working with
another teacher, maybe a geography teacher in the school or other
teachers, you could actually link this into curriculum. and into
their real life, but building out some kind of skills as well from
technology.
I'm really glad I did that, Ray. Thanks very much for um I
hopefully this has come across well on the podcast through through
audio. We try to, you know, add that kind of um element to it.
Yeah. And in the in the podcast links, you'll find the link to be
able to go and watch what Dan was doing because we've recorded it.
We'll put it onto YouTube with this.
So, Ray, how do people find out where to go to get these
resources?
Well, I'm glad you asked me, Dan. So, the thing that's most
important for people to know is they can go and get all of this if
they go to aka.ms hour of code.
Fantastic. And if they've got the the most recent version of
Minecraft, it's in there already.
Yeah, if you've got Minecraft Education Edition, it should be in
there already. Or you might just need to check to see if there's an
update. Um, but if you haven't, just go there. It gives you the
links to download it. There's the curriculum resources that are
there. But to be honest, I don't actually think you need it. Just
download it, have a crack yourself, and you'll realize what you can
do with your students.
And uh it's free. So, if you're in an Australian school, I know
that you're licensed for Minecraft Education Edition, I think. Yes,
in most schools. So, you just log in with your
school email address. If you're not in an Australian school, then
try it with the school email address. If not, don't worry. You can
actually go through do do the arrow of code as a guest user without
needing a license.
Yeah.
Cool. Okay.
Fantastic. Thanks, Ray. stand. See you next week.
Cheers.