Get AI on Par with Your Network Domain Experts

0:01 [Music]

0:07 artificial intelligence is improving

0:09 user experiences across the spectrum

0:11 from it teams to consumers the ability

0:13 of our networks to access large amounts

0:15 of data quickly analyze it and make

0:18 decisions is changing the way we work

0:20 and live i'm rick rutter from juniper

0:23 i've assembled a couple of my colleagues

0:25 to help me in a discussion around ai and

0:28 how it's improving our world in

0:30 delivering positive change gentlemen

0:32 thanks for coming if you could quickly

0:34 introduce yourselves yeah rick bob

0:35 friday cto of juniper's enterprise

0:37 business

0:39 and thank you rick uh sudhir mata uh vp

0:41 of products here at the juniper's ai

0:43 driven enterprise

0:44 right on well to kick things off what

0:46 are some of the challenges that

0:47 companies are facing today and how do

0:49 you think ai ops can help

0:51 one of the things that's happening in

0:53 customers around the world big and small

0:56 is more users more devices more

0:59 applications more bandwidth every single

1:02 day right but the scale of the it team

1:05 isn't growing at the size of the

1:06 exponential growth we're seeing in these

1:08 other vectors so how does the same says

1:12 i.t team deal with more and more and

1:14 more and more of this you can't throw

1:16 people at the problem it is time for ai

1:20 to bring that cost of operations down so

1:23 the same size it team can handle more of

1:27 the scope on the same network right

1:29 that's the power of ai makes sense yeah

1:31 and for me personally i think we talked

1:32 about you know when i was at cisco it

1:34 was clear what i heard from these

1:36 customers was hey they were tired of

1:38 controllers crashing you know they were

1:40 tired waiting for a year to actually get

1:42 new code and they really wanted to know

1:44 visibility it wasn't good enough to know

1:46 whether or not the ap was up and running

1:47 they wanted to know if the user was

1:48 having a great experience and so that

1:50 was really for me what i saw the need

1:52 for mist and relate this cloud ai

1:54 architecture

1:56 that makes sense i mean like so let's

1:57 dive into the virtual assistant because

1:59 i think that's amazing like why

2:02 help help me understand a little bit

2:04 like why a virtual assistant and how it

2:06 can really help my business

2:08 you want to start yeah so so

2:11 one of the things that we talked about

2:13 at the very beginning right why do we

2:14 need ai there's more more users more

2:16 data more applications more trouble

2:18 tickets every day

2:20 the

2:21 simplest reason for a virtual assistant

2:24 is we want problem solving to be pushed

2:27 down to the lowest tier level tier one

2:29 when a call comes into university and on

2:32 the other side is a zoology major taking

2:33 a trouble ticket in the help desk they

2:36 could just ask marvis hey what's wrong

2:38 with this user's experience right and

2:40 marvis comes back with a curated answer

2:42 with a factual actual answer right and

2:44 so pushing the the amount of the

2:48 knowledge graph needed to solve these

2:50 problems we went from ccies and jncies

2:54 to level one to be able to solve that

2:56 problem that's one of the reasons we

2:57 went down this marvelous path yeah i

2:59 mean i think the fundamentally the other

3:00 thing is you look at the poor i.t teams

3:02 in the future these networks are getting

3:04 much more complex than 20 years ago 20

3:07 years ago everything lived inside their

3:09 little data center and their firewalls

3:10 nowadays we have everything distributed

3:12 to the cloud you know zoom microsoft all

3:15 this stuff is outside of their control

3:17 you know so we're moving to a much more

3:19 mobile world you know if you look where

3:21 we are with mobility you know we're at

3:23 60 percent of everyone in the world is

3:25 connected to the internet through some

3:26 wireless mobile link that by itself has

3:29 another layer of complexity and they're

3:31 going everywhere on the internet and so

3:33 that's where ai ops is he is helping

3:35 these deal with complexity right and

3:38 that's what aiml is really designed to

3:39 do is help deal with complexity and deal

3:42 with these more complex networks it

3:44 makes a lot of sense and sometimes i i

3:46 feel like people get confused the

3:47 difference between like ai and ml and

3:50 some of these you know other topics like

3:52 can you help me understand the

3:54 difference between ai and ml and and how

3:56 businesses can use those you know so you

3:58 know this is a great question what i

4:00 tell people right we have been working

4:02 on automation ever since networking

4:04 started ai is this ultimate taking

4:07 automation to the ultimate level right

4:09 ultimately ai is a concept of building

4:11 something on par with a human you know

4:14 when i started in this i was really

4:16 inspired by jeopardy right when i saw

4:18 watson playing jeopardy i was like if

4:20 they can build something that can play

4:21 jeopardy

4:22 we should be able to build something

4:24 that can answer questions on par with

4:25 network domain experts right and so that

4:28 is ai and most people confuse ai with ml

4:31 aim is doing something on par of the

4:33 human and it takes a lot of different

4:35 algorithms ml is actually math and

4:38 algorithms that actually helps us get

4:39 that implemented now that's important

4:41 because

4:42 of late in 2014 is when ml really took

4:45 off that's when we saw

4:47 google and

4:49 all these neural network stuff we got to

4:51 enough computing power and data to

4:53 actually train these models so there was

4:55 an inflection point in 2014 that really

4:57 saw the birth of ai and ml across all

4:59 types of industries

5:01 that's amazing right on this is very

5:03 cool by the way bob's phd is in this

5:05 field so i didn't know that

5:07 but it makes sense yeah that's awesome

5:10 one other thing that seems to be coming

5:11 up more and more is like sustainability

5:13 and i think we were talking a little bit

5:15 like how can we help like a lot of our

5:17 customers like hit their sustainability

5:19 goals i mean climate is something that's

5:21 on everybody's mind can something like

5:23 this help

5:24 with that actually it can and let me

5:26 step back and say we as juniper we have

5:28 pledged that we are going to we're going

5:30 we're going green and we're where we'll

5:32 be at 100 carbon carbon neutral by 2025

5:36 it's is as a company great goal for us

5:39 but let's talk about the impact of ai on

5:41 sustainability not too many in it think

5:44 of it this way

5:45 so if i can eliminate a truck roll to a

5:48 site by virtue of having ai to solve a

5:51 problem either i've self-remediated it

5:54 or we've we've created the

5:56 the data needed in the cloud so that

5:58 somebody doesn't have to go

5:59 and real example the gap eliminated 85

6:03 percent of the truck rolls going to gap

6:05 retail stores around the world 85 truck

6:08 rolls are down

6:10 that's sustainability ai can have a very

6:13 real impact in terms of operations and

6:15 efficiencies we drive including

6:18 the green initiative that's amazing

6:20 really really cool um so let me look i

6:23 think we've gone through a lot of these

6:24 but i mean one of the things uh

6:27 that that i keep like struggling with is

6:29 i i hear that like people are still

6:32 skeptical of ai like why is that

6:36 so so so let me tell you i would say you

6:38 know the hardest thing for for people to

6:41 walk away from is if i'm a cli uh you

6:45 know expert on on cisco if i've mastered

6:48 and carefully hand chiseled a network of

6:51 aruba it's hard for me to walk away from

6:53 that and that's that's the challenge and

6:55 so there's a lot of skepticism in ai in

6:57 terms of mostly

6:59 knowing what they know and the

7:00 incumbency and the legacy architectures

7:03 but

7:03 if you're going to prove a concept with

7:05 us if you see a proof of value there is

7:08 no more skepticism once you try it once

7:11 there is no ai skeptics on the juniper

7:13 missed customer side yeah i mean i think

7:16 you know if you look at people it's like

7:17 you know you're driving your tesla

7:19 right i don't think most people are

7:21 ready to take their hands off the wheel

7:23 yet you know so gaining that trust

7:25 inside your virtual ai system is

7:27 combination trying to understand what's

7:29 going on and that's probably one issue

7:31 with aiml right because the math is a

7:32 little bit more complicated than most

7:34 networking it guys used to right they're

7:36 used to automating things with scripts

7:37 or something they can logically

7:38 understand this is a model where you're

7:40 basically training it so there is a

7:42 level of trust you have to learn to

7:43 trust it and that's where we worked at

7:45 marvis is around this conversational

7:47 interface right you know how do you make

7:49 a virtual assistant really a part of

7:50 your team you know so that is a key part

7:53 that you know we are working on right

7:54 now is making it easy for people to

7:56 interact with marvis and actually get

7:59 them to see that you know

8:00 when you actually start using it you

8:02 know it's poc right when you get this

8:03 thing up and running you know when

8:05 marvis says there's a bad cable

8:07 well it's like well let me see you know

8:09 i don't know if i trust you the first 10

8:11 times but after the 10th time okay i'll

8:13 trust you so now customers are starting

8:15 to actually let marvis issue those

8:17 support tickets right okay i trust you

8:19 that you found a bad cable i'm not going

8:21 to ask you anymore this saves an

8:23 incredible amount of time resources

8:25 money yeah i'll give you the biggest

8:28 skeptic that uh that i've actually

8:30 worked with uh most recently is someone

8:33 is a about a 4000 ap higher education

8:36 university a top 10 university in the in

8:38 the country and the it manager there a

8:40 long time friend and he said hey

8:44 i built this network sort of hand

8:46 chiseled the entire network for high

8:48 performance

8:49 i don't believe that

8:52 marvis can actually really just out of

8:55 the box work in a high density

8:57 environment like

8:58 300 people a person auditorium

9:01 previously

9:02 people didn't trust ai for that they

9:04 used to manually channel plan and hand

9:06 plan this but now with marvis we see

9:09 data better than humans can see it's

9:11 really that simple

9:13 for the first time he said sudhir

9:15 i called on your powerpoint

9:17 right but i went into the proof of

9:19 concept i let the system figure it out

9:22 and it came up with a better answer than

9:23 i could on that network and i would say

9:25 i think you find you know people are not

9:27 so much skeptic about ai i think

9:29 everyone realizes that ai you know we

9:31 are going to see a self-driving car in

9:33 our lifetime right you know we are

9:35 seeing doctors being able to diagnose

9:36 cancer better with ai so i don't think

9:39 anyone is skeptical about ai working i

9:41 think it's more skeptical when

9:43 you know when are you going to be in the

9:44 back seat of your tesla pressing it to

9:47 drive well there was a guy the other day

9:49 you told me that's what i told you there

9:50 was a guy you know and the scary thing

9:52 it is happening right there was a guy in

9:54 oakland who got arrested for being in

9:56 the backseat of his tesla so it's not

9:58 like as possible it's like when is it

9:59 going to become mainstream and if he can

10:02 do that we can definitely put on the

10:03 network and help businesses improve

10:05 their lives if someone can ride in the

10:06 back of their tesla across the bay

10:08 bridge and someone in jeopardy watson

10:09 can play jeopardy yes we can build

10:11 something that can answer questions i'm

10:12 part of network domain expert the good

10:14 news rick though is we've actually

10:16 proven it now right every single network

10:19 we're going into has fewer tickets

10:22 faster deployments fast resolution it's

10:24 happening now you don't have to wait

10:26 with some of our like with some of our

10:28 competitors who are saying oh yeah i'm

10:30 building this it's coming please wait no

10:34 go into a poc with them right now and

10:36 see the difference

10:38 well listen sadeer thank you so much bob

10:40 thank you as well uh if anybody is still

10:43 skeptical about a i challenge you let's

10:45 give these gentlemen a call get that poc

10:48 working

10:49 thank you all very much and see you soon

10:52 [Music]