Video 1

this video will introduce you to machine

learning for kids this is a front-end to

IBM Watson assistant and it’s I find it

pretty easy to use and I think you will

enjoy using it when you go to the

website and which is machine learning

for kids dot co uk and by the way this

is this is a British website and it’s

from a British IBM employee who just set

upon himself to make something easy and

useful for us when you go to the website

you’ll see a prompt for getting started

learning more and you’re free to explore

that but what you are eventually going

to do is log in I’ve given each of you a

password a user ID and a password and if

it prompted you to log in which it’s not

going to here because I’ve already

logged in that’s what you will use so

once you log in I’m gonna go over here

this is my account and I have a project

you can you can look at things like

project ideas well those are my projects

I’ll come back to this one in a minute

you can look at worksheets for all

different kinds of projects we won’t be

doing the image projects here that costs

money and so we won’t be doing that

we’ll be focusing on the text base

projects here’s the chatbot project and

we will have a worksheet for that one

but you can experiment with the numbers

worksheet I think there’s a lot of these

are images here’s one for numbers and

you can certainly experiment with that

on on a free code day let’s go back to

projects and I have a project here that

I want to show you and I did the make me

happy project so what you will do is

you’ll you’ll say add a new project and

then it will create this if I go here

this is where we make our own intense so

you remember in those tone

Eliezer there was joy and sadness and I

know what was it analytical and some of

the others there were seven different

ones here I’ve only made two when you

make your chatbot you can make as few or

as many intense as you want here I put

two happy and sad so we’ll see if Watson

can detect the difference between

something happy and something sacked so

you can read these I put some happy

things that I hear students talking

about and some sad things so that’s good

so you will put different you have to

have at least five in each group then

you will train the model this is called

transfer learning this is an AI

technique where you take your data and

you train an existing AI model in this

case Watson to be a little bit smarter

about what you are interested in so here

I have already trained it and you can

see that it says available if I had not

trained it it would say train train the

model again I’m not going to do that

it’ll it takes anywhere between 30 30

seconds to as much as five minutes if I

had a lot of data but it says that it’s

trained it’s available so I can enter

text here and it’ll respond very

similarly to the way tone analyzer

responded so let me just type something

in here that I think is sad my dad took

my phone that’s possibly the height of

middle-school sadness and I’m going to

test it and I’m expecting Watson to put

it in one of those two categories and

also give me a confidence level well

very quickly so this goes this is

actually running in Watson this is going

off to a supercomputer in Dallas and it

has responded and it says it thinks

that’s a sad thing and it gives it about

83 percent confidence so that’s pretty

impressive especially given that there

wasn’t anything in the training data

about

dad’s or taking or phones so Watson has

some sense that that’s a sad thing

so once you’ve trained your data you

feel like it’s understanding now what if

it didn’t understand what I type then

you might have to go back and add some

additional training statements so you’re

gonna do that a lot in this model all

right so once we have that then I can

click on make and you’ll be making a

Python model and there will be a

worksheet in blend that you can look at

for more details on how to do that I

want to show you just a super simple

example of what I can do with the make

me happy model when I do this I’ve added

some more code to it so here’s the most

basic version of what you can do with

this a trained model here this is a

scratch extension and there’s some

commands here that very specifically go

off and talk to Watson I think that’s

pretty cool to be able to do that in

scratch and so I have a couple of

variables here the confidence number and

here it comes back zero to a hundred

it’s not it’s not the percentages like

in tone analyzer and then in which label

so let’s run this and let’s see you can

look at this scratch code it should be

pretty pretty self-explanatory so let’s

give it the exact same example we just

did my dad took my phone and sure enough

it we get the same response it thinks

it’s a sad thing

eighty-three percent confidence and I

switched the costume on the cat to be

sad what you’re gonna do in Python is

something a little more elaborate

because you’re gonna respond also to

that confidence rating so here’s an

example in scratch where I’m doing that

also it’s kind of easy to to get this

and scratch so basically the same thing

if it’s a hat if the label is happy then

we’re gonna get a happy face here’s my

costumes there’s that one that’s that

one

those are really great drawings of cats

but the if it’s happy then it switches

to the happy face and then it’s going to

go through a series of if statements on

the confidence and if it’s you can see

what’s happening here there are three

levels of confidence so let’s try this

again

now we’ll try this one again and we’ll

try some other variations helps if I

could time yeah so it saw that it was

83% confident that it was a sad thing so

it said you sound really sad because I

put my my level at seventy-five you can

put whatever you want I have a little

reset button here alright let’s try

another one let’s say oh I don’t know we

could do it’s my birthday and that might

have been in actually in the training

data that’s what I’m guessing it it

really nailed that one I think that was

probably actually in the training data

let’s try something else I’m not sure

why but let’s go ahead and say that and

it knows that that’s a pretty pretty

happy thing if you said something like

feel blah today we don’t know what blah

is exactly for this person but if we say

thinks that sounds sad I would have put

it more in the middle but anyway you get

the idea

this code both sets of code are posted

on blend you’re gonna get a chance to

get started in Python and I think that

you will have fun with this it’s not

often you get to to train an AI and

that’s it