24pcoqhe6Cg

so below there and welcome to another
tutorial my name is Tamara Bakshi and
today I'm joined by mark stir event so
thank you very much mark for joining
today
would you like to quickly introduce
yourself sure I work in developer
advocacy in IBM specifically I work on a
lot of code patterns so we're creating
code patterns with Watson and AI and
we're making the public on github so
that you can follow the code pattern and
try them out
that's great thank you very much for
joining in today but today specifically
we're talking about the Watson assistant
and how you can tie it in with the
Google assistant and Amazon the lexo
into example use cases
first of all car rentals and how you can
actually rent out car leasing chatbots
and they're not going to be talking
about checking the weather through
natural language in fact you may
remember I already have a video on the
Amazon Alexa and tiny then would ask 10
Meg and in fact helping me out with that
project was Mark servant that guy wasn't
able to find out how exactly to have
Alexa output its intense or whatever you
say to Watson system because that
requires a little more grounds thank you
very much for that Marc but now let's
get into the actual use cases okay so
tell us a little bit about what kinds of
applications we're gonna be building
today okay so the basic applications and
then they're meant for demo they're
fairly simple but the main thing we
wanted to do with both of these is show
you how you can use an Alexa or a Google
assistant front-end which is fairly
popular and available a lot of devices
but instead of trying to build that
conversation and those platforms take
advantage of Watson wants an assistant
and build a conversation a dialogue
there so what we have is we have ways of
getting basic text from the conversation
when you're speaking to these other
devices it gets into Watson and Watson
controls the reply and the response in
the back and forth so first we'll use
slots to gather information so what kind
of car do you want what color car all
these slots need to be filled before you
satisfy that and later we'll give an
example of digressions because
we did with the with the Google
assistant example the rental cars is as
you're talking to a retail agent and
being prompted maybe what kind of car do
you want a lot of people will ask
questions typical ones are are you a
robot you know guys talk to your manager
or I'm just I'm really upset right now
and you want to be able to handle those
and we don't have examples where we're
actually calling out to another person
but we have example if I could handle
the dialog you'd have to be able to
reply to the digression and still jump
right back to what kind of car do you
want a very powerful feature and it's a
very surprisingly natural conversation
that you can have with a fairly simple
chat bot that's great mark now I mean as
you mentioned a lot of people are gonna
do things like chit chat with a bot
after you Bank in Australia I applied
Watson to their mortgage or loan
applications the second most asked
question on that chat bot was what's
your favorite color and the first most
asked question was will you marry me so
because you can tell a lot of people
will use a job often when it's intended
to work before before but even more
gonna use it the things that it wasn't
intended for I'm trying to stump it
because it's a robot but that's why
Watson exists and now using Watson
assistant with other platforms like the
Google assistant ends on Alexa Sirikit
so anymore that you could use it with
you're able to create one unified
experience in chat BOTS across all
platforms that having to rebuild it for
every individual library or every
individual platform and then you can use
really powerful tools that Watson
assistant provides in order to build
more natural conversations like that
Russians going from one topic to another
and then slots trying to get the correct
information and no matter how many
messages you provide so now that let's
talk a little bit about how exactly this
application is gonna flow so what could
you have to quickly walk us through say
the car rental use case okay so the car
rental use case is using Google
assistant so we start over here with
Google assistant
we care how abbreviate some of these
things on video so it's calling out so
we're gonna have an HTTP endpoint
calling out and it's got JSON in there
with enough information basically all we
want is the text so Google assistant
will handle the voice and we're going to
get a string of text along with some
other information but that's the request
so it's coming into IBM cloud which
we're using to host an ojs server so
that's where we have the endpoint so IBM
cloud takes care of the hosting it gives
us the endpoint we've got code that's
getting this JSON and I've got the text
and I could call out to Watson so Watson
services our service also hosted on the
cloud and so that's where we have
workspaces for different conversations
so the the Weather Channel one was one
convert one workspace the rental car
conversation was another workspace so I
have those hosted and they have the
node.js server setup so that it's going
to communicate with Watson malfini Alexa
one was built a little bit differently
and we switched to another color I guess
so I need another marker more than
another color so if we go to the Alexa
it's the same model we set it up so that
we're basically just getting all the
words that were said in in text and it's
going to also call our HTTP we're going
to have that text in a JSON structure
but there we use cloud functions so IBM
cloud functions
that's hosting our code so that was
JavaScript but again once we get that
request we just call out using an SDK -
call out - what's an assistant and get
their reply now the amount of code you
need to call out is very little and once
you get the reply from once and you have
the answer in both cases you're just
building a little bit of JSON in a
specific format that Google will
understand or Alexa will understand the
Alexa one does have one other bit the
conversation is the same but if it comes
back and says to a weather lookup then
it has some code to call up to the
weather game service I should mention
that weather weather channel data
weather channel data so even though
that's a fairly simple test example that
you can use you give it a location it'll
give you the real weather report from
the Weather Channel okay thank you for
walking us through that now let's take a
look at a quick demo how you can apply
this to the Google assistant and that
let's get into how this actually works
featuring of course service computing
the weather channel of course kind that
really accurate weather day of my VM and
of course using nodejs along with the
Watson assistant in order to host these
natural conversations let's take a look
hey Google talked to Watson
all right getting the test version of
Watson I can reserve a rental car for
you what day do you need a car I need it
tomorrow what time on Thursday do you
want the car to be ready sometime around
noon
what type of car would you like and like
an economy car what color economy car do
you want why don't you make that a car
that's really fast what color sports car
do you want definitely a red sports car
okay please confirm you want a red
sports car on the 16th of August 2018 at
12 p.m. is that correct that's correct
great unfortunately I don't really have
any cars this is a demo chat
let's start over what date do you need a
car I'd like a blue mid-sized car
Thursday at midnight what time on
Thursday do you want the car to be ready
how about 12 a.m. ok please confirm you
want a blue midsize car on the 16th of
August 2018 at 12 a.m. is that correct
ok
unfortunately I don't really have any
cars this is a demo chat okay so we'll
show you where the code patterns live so
you can use them and you'll try them out
on your own so we have the code here and
all the steps you need to run it here at
github.com slash IBM we have Alexis
skill Watson conversation this is the
one word we've hooked up a Watson
workspace with an Alexis skill and if
you look at the readme you'll see what
we provide with code patterns a lot of
information about what we're doing and
the exact steps you need to run it so
there'll always be an architectural
picture that shows you what I tried to
draw on the board you'll see the
components here we're using IBM cloud
functions we're using red is to store
context and we're calling out to Weather
Channel in addition of course to Watson
assistant but if you look at the readme
you'll see that we have an example video
and we have detailed steps so you can
run this one locally or on the cloud
well this one we run it locally but we
also use IBM cloud functions and we have
a very similar thing for the Watson
Google assistant so this one's at
github.com slash IBM slash Watson Google
assistant again with dashes and the
readme here you'll see a very similar
format you'll see the architecture you
talk to Google it talks to no js' and
talks to Watson assistant and the steps
here since this one is running right on
the cloud we just have a deployed IBM
cloud button so you can do that and
it'll deploy your node.js server it'll
spin up your Watson assistant service
with your workspace so very easy to get
this one up and running but it's also
documented so you have all the steps you
need so one that fires up you'll get a
Watson assistant service running an IBM
cloud and it will upload a workspace so
here we uploaded a workspace for the
rent-a-car conversation and if we take a
look at this you'll see that there are a
lot of intense and I didn't have to put
in most of these I simply use the
existing content catalog we have some
industry specific conversations and also
things like vodka
all general and customer care so for
rent renting a car I used a lot of the
bot control well I use a little of the
bot control to say confirm reject for
example and start over customer care for
things like when are your open and
general for things that people ask a lot
that I wouldn't have even thought about
but when I looked in there I saw
questions like are you human are you a
robot and I talked to your manager
you're stupid
all those things that can be thrown in a
chat box and you want to address and
since I have another sort of PowerPoint
of walks an assistant is that you don't
need to build absolutely everything from
scratch of course some very specifics to
your use case you will need to program
in of course or the content catalog
takes a bunch of things they are
repetitive or mundane and you won't need
to code those in because the intents and
entities are there for you you just need
to put it in the relevant dialogue for
them yeah
for example if you look at the intent so
like taking those three categories I
just added those to my workspace and
you'll see the list of intents I got so
a lot of bot control a lot of customer
care a lot of just general you know like
this human or bot question and if you
look at any one of these you'll see
let's see what's a very easy one just
approve response you could expect a yes
or no response in your chat bot and
that's a pretty weak chat bot but just
by using this one that came from the
catalogue it already had 21 examples you
know affirmative all right correct for
sure got it great plus it's being
trained so you don't have to
specifically say any one of these these
aren't keywords so you've gone from
expecting and yes or no if you're hard
coding it to 21 examples plus machine
learning to build on that and get
various variations of that it'll still
understand your attempt yeah so if you
scroll down like for example let's just
say the word absolutely isn't here if I
were to still say the word absolutely it
would understand
this is the intent I mean even though
the word absolutely wasn't in the
examples yes the reason it does that is
because of machine learning it
understands my Intendant understands
what I listen man ticks not the sin
tactics of what I'm trying to say right
and if I found out that it doesn't
understand
you betcha I'm for sure you can just add
the intent it'll train it immediately
perfect and you'll you'll very quickly
be able to handle that mmm-hmm so also
if we look down here so the general ones
are where it was interesting to see what
people might ask like the human or bot
and again you have a bunch of examples
so in my dialogue I didn't care to
address all of these all you really need
is someone's going to type something in
that more or less matches this intent
you understand what they're thinking you
don't need to know the details about
exactly what they asked but you're going
to reply with your answer for what you
do when someone's intent is to find out
whether you're a human or not mm-hmm so
this takes a lot of the work out of
trying to build that and a lot of the
the machine learning the training is
done now in addition to the content
catalog I did a simple reserve a car cop
you know intent that just says I need a
car rent a car but this dialogue is
fairly straightforward look its sole
purpose is to rent a car and deal with
these other digressions while it's
trying to rent you a car exactly so it's
going to go into reserve a car mode with
pretty much everything you do and that's
the dialogue so let's take a look at the
dialogue well those are the intents now
let's take a look at a dialogue yeah I
mean the dialogue is the part that does
this rental car thing I was talking
about and so if we look here I kept it
fairly simple there's a starting point
where I wanted to be able to clear the
context to make sure that it doesn't
remember what you reserved last time
there is a way to start over and start
fresh but it always jumps right to
reserve a car after it says hello I can
rent you a car so that first time that
intro you want is you want to say
something what it's going to do so it's
got the intro and it jumps to reserve a
car quit is important
be able to kind of end this especially
with Google assistant there's some
special checking in the code because if
you don't handle that that end responds
correctly you're gonna get an error in
Google assistant it it will be unhappy
if it says I want to quit and you reply
with something that's asking for more
information yes it's an error is which
is kind of annoying so that's handled
but all those intents that I consider
digressions I put these in a folder so I
can deal with them but I have the
ability to jump in and out of those and
let's take a look at the reserve at cart
one the specific one for observing car
it has slots so this is where you see
what it's checking for and all these are
considered slots when it wants to
reserve a car if it doesn't know what
day is going to ask you what day you
mean it if it doesn't know what time it
will ask you what time okay but if I say
I need a car tomorrow at noon it's gonna
fill in both those slots mm-hmm another
slot is car type car color and then at
the very end when this last ones in
focus that's going to do the confirm but
with these first four now if I say
tomorrow at noon I need a red midsize
car they're all filled in it'll go
straight yeah so yeah you're right
that's actually a very very powerful
feature of these slots it's like for
example last time I created a actual
Facebook live series on the Watson
system back on these we called Watson
conversation and there was someone who
asked during the Facebook slides well if
I'm trying to order a pizza and I want a
thin crust pizza the medium size with
these toppings 22 specify them one by
one and listen to everything watch
enough to say no you don't now using
slots in literally one conversation
they're ready to put everything to get
to and in this simplified rental car
dialogue has two perfect examples for
that the date and time if I say what day
do you want a car and you say tomorrow
at noon yeah a lot of chat bus might
catch the tomorrow mm-hmm but and then
they're gonna say what time because
they're not looking for that time and
that's really annoying if you said
tomorrow at noon and in the bottom
as what time - exact and the other great
example is the car what I think what
type of car would you want a person
might say I want a red sports car yeah
some BOTS would then say what color red
that's right
which it's okay I guess
but it's really annoying and and people
want to have BOTS to interact like a
human and the entire point here is for
the chat BOTS to a Cuban or else I mean
why would you have a forum why would you
want to use a chat bot - for it to be
natural
yeah our point is like for example when
you call up one of those for example
helplines there and the robot is answer
anyway say tell me what your problem is
and they don't understand yes it's so
frustrating to keep telling them yeah
we're trying to get to the point that
you just want human on the line ASAP
what this is meant to do is connect with
humans be natural so you're not
frustrated talking to it and absolutely
repetitive questions you're going to get
frustrated yeah I know people are a
little concerned when chat BOTS start to
sound too human-like
like are these are they being tricky or
is it sneaky is there something wrong
with that I'm not sure but people are
definitely a little uncomfortable with
that but the great thing about talking
like a human is it changes the whole
interaction exactly the way people talk
to a very robot a form-based prompting
thing is very different than the way
someone naturally conversely like with
this time and date thing I don't need
just I don't want to just have prompts
where I have to say the word I can say
oh I guess I probably need it by 3
o'clock yeah um and it's should be good
enough to understand the important part
of that exactly without forcing me to
clean up the way I talk exactly - magic
form yeah a few years they're going to
use a search engine you try and take
whatever you're trying to search for and
move it in this weird weird syntactical
sort of template right we're for Google
or Bing or to find what you're looking
for because it's just doing simple
keyword matching nowadays because of
machine learning for example Watson
right if you take a look at how Watson
played Jeopardy seven years ago he was
able to understand the puns a riddle
we're playing jeopardy questions and now
you can take it back soon New York
technology implement your own chat BOTS
right now the thing I thought was fun
about this particular one we didn't demo
digressions did we not yet let's take a
look at that I forgot about that
so this chat BOTS going to be very
persistent about trying to fill these
slots and get your rental car but as we
mentioned with all the bot control and
the other things I put in this digress
and return folder I don't have the
ability to in the middle of this
collecting these slots digress and
there's a little thing here with the cut
under customize that allows you to turn
on and off digression whether you're
allowed to jump to a digression and
whether you only want to go to
digressions that return to the note so
here I'm going into these this
digression folder and they're all going
to return to my node so I can digress
but it's going to come back and say what
color mm-hmm and most of my digression
is that I that I was interested in are
in this folder there was one bot control
one here this starter start over that I
left separate and I made sure I could
not digress to that because at one point
I did get another loop oh so the easiest
fix for that was let's not allow that
one because of this little jump to yeah
but there would be a variety of ways of
fixing that but if I go into the folder
and I look into a digression so in the
middle of this date and time size color
if someone says are you a human that is
this digression so I just have a note in
the dialogue I put in one response that
says oh I clicked on the wrong one this
is the prefer me to a human agent asking
for your manager things like that would
be refer to a human and in this case I
just had my my robot say I'm sorry the
humans did not come actually watch so I
just have some simple examples like I
put one in there but of course you can
have and should have many responses so
that it doesn't sound so recorded that
the same response at any time the better
example for that
is the the the joke I did one bad joke
but if I really cared to do it I would
put multiple responses in there you get
a different one every time or at least
feel like it's different every time why
did the chicken rent a car to cross the
road seemed relevant I didn't have any
rental car jokes but in the middle of
conversation these questions can be
asked it will say this but since it's
jumping right back to the chatbot you
will get the joke
the hahaha in a robot voice to some
extent of this and then you'll get what
color car did you want because it's
still persistent about filling in those
slots and getting you to the end of that
part of the conversation the like refer
to your manager refer to agent those
types of things I think typically you
might have more code in here where you
would actually have some routing maybe
from your chat bot to human engines but
we certainly didn't get into that with
our example mm-hmm so I don't did you
want to see the digression example let's
take a look because I was going to show
you an example of Alexa but since we
missed that we can go with this first
yeah okay so I have the two different
developer environments here the console
for actions on Google and the Alexus
skills developer kit so I happen to be
here on the Google one and since we
didn't cover that earlier let's just try
this you can type or speak say talk to
Watson unfortunately I don't really have
any cost so this is a demo chat this was
completing what I did last time let me
see if I can get it to start over start
over what Dave you need a car
do you need do you know any jokes I know
one joke why did the chicken rent a car
to cross the road
ha ha ha what do you need a car there go
so that's a digression and it's still
going to say what day do you need a car
and if I say that was a stupid joke
there's an intent to D it was still
listen I know one joke sorry it caught
on the joke thing yes let me try this
again that was stupid
I'm sorry you feel that way I'm still
learning what there's an intent for
negative feedback yeah and there's an
intent for positive feedback this is a
very simple chat bot knows how to say
I'm sorry you feel that way but all that
knows how to do is go back to prompting
but at least it knows how to say I'm
sorry exactly so what it's doing is
you're saying you're providing it
negative reinforcement and saying all
right you know it it's responding to you
phone's going right back to one is a day
at the car and so you don't need to
manually say hey you know based off of
context where were you jump backs that
no you do all of that now with this new
directions future it can go over to
notes and come right back to gather in a
slot information and another very simple
one is that was great well thank you
it's my pleasure what day do you need a
car but you'll see what this is actually
showing here only by mistake you turn on
the microphone there's some gibberish in
there but it picked up on the bat was
great exactly so what's happening to use
a natural ended processing is
compensating for the fact that the
speech recognition that it was and and I
really feel like the interaction I
didn't spend a lot of time on these
digressions
but just a little bit like this thank
you is my pleasure
what did you need a car and the I'm
sorry you feel that way yeah so much
more natural than when it originally
would just keep prompting the actually
the Alexa one has that example
doesn't have digressions it has weather
yes and it will keep prompting for
location over and over and over yes and
it seems a bit rude this is much more
easy to work with yes I see so we might
as well jump to that I think we demo
what we needed to demo your heads and
that and the digressions but here the
the Alexus skills kit very similar
console if you go through the steps
you'll set it up and you can test it
right here I usually find this as the
best place to test it but you can also
you know you can release it there are
other web forms you can test it with or
if you have an Alexa device you can test
it you see if this is ready to go this
one this one I hold the mic the other
one I just click it Oh
Alexa talked to Weather Channel sorry
I'm not sure I did this feels cold what
is this one called Watson yes
Alexa asked Watson what the weather is
weather for what location remember if
you enter your full address in the Alexa
app I'll be able to automatically get
the weather for where you are
this is using the Alexa built in this is
using the Alexa built in let's try again
ask weather channel
hi I can help you to find current
weather conditions okay so it was a
little tricky getting in and out while
it was resetting but now Alexa's talking
to our Watson Service I recognize that
response yes says it can help with the
current weather what's the forecast
please tell me your location this is our
one slot and the no content is because I
keep clicking the microphone you haven't
supposed to hold it so the one slot is
the location
so like the rental car one if I say do
you know any jokes in what location it's
gonna keep from anything okay yes yes so
not only is it not funny not fun but
it's in a lot of cases is kind of
annoying yet because this agent that
you're working with trying to accomplish
something with is really just ignoring
you exactly so this is what happens and
you don't use directions so that's why
you license it so that's digressions and
we should say Honolulu because we want
to show you that this actually songs to
the weather forecasts okay Oh in what
location oh it only got Luther Toronto
it's a nice day here in Toronto yes it
is okay I'll look up the weather for
Toronto looking up weather information
for Toronto partly cloudy
lows overnight in the low 70s so the
coat pattern for Alexa it it does have
the wiring there to have that
conversation that simple slot a simple
example of how to work with Watson
assistant but then when it's when you
get to the point where you have a
location want to look up the weather
that's your intent it will call up to
the Weather Channel and it will get back
the the weather forecast where you
actually have a lot of details about
whether you could get much more
information but when you're talking to a
voice device yes fortunately there's a
fairly brief one then you can be spoken
okay so
what didn't we cover we covered
everything so we talked about the code
pattern we talked about Watson system
the attempts to use in the dialogue and
how you can use a Content catalog before
we talk about the slots that we can use
specifically like our mental example of
course the digressions and you've also
shown a quick demo of the Alexa and now
I think that's only half of code I think
I think it is I think we covered it oh
if you did want to look at the code
though of course it's on github and
remember there's it these are both
written in JavaScript one has a node.js
server one has IBM cloud functions so
the that most of these call-outs are
very simple not a lot of code you need
to me but you have to code me alright so
thank you very much everyone for joining
in today that's what we had to show you
the code the demo of a like site Google
assistant and how you can use Watson
assistant as the one sort of
conversational fabric for all these
different platforms so again thank you
very much Mark for joining in today of
course bad to have you on but apart from
that good luck if you do have any
suggestions questions or feedback please
do reach out to either me or mark in
fact marks the mark's contact will be
down in the description below but would
you like to keep matilda here's how
people can contact you sure you can
contact me mark stir at gmail it's ma RK
s tu r as in short for mark's turbine
but yeah you can use my gmail or look
for my youtube channel and we'll talk to
you there sounds good perfect again this
contact will be down in the description
below in fact mark actually has his own
YouTube channel so there'll be a link to
his YouTube channel down in the
description below and if you like that
for code patents that he creates make
sure you subscribe to his channel too so
again thank you very much mark you'd
like to contact me my death my contact
will be down in the description below
and apart from that if you do like the
content on this channel and you do want
to see more of it please do make sure to
subscribe and if you'd like to be
notified whenever I release new videos
make sure you turn on a little about my
comments by the subscribe button too
emailing annual notification yeah thank
you very much and thank you for watching
Sammy thank you bye-bye