AI Skeptics: Get Started On Your AI Journey

0:00 [Music]

0:06 hello everybody i'm zeus caraval

0:07 principal analyst with zk research and

0:09 i'm hosting uh this episode of ai

0:12 skeptics

0:13 in our previous conversations of the ai

0:15 skeptific series we've heard about

0:17 perspectives around how

0:19 ai or artificial intelligence will help

0:21 it practitioners

0:23 and technology channel partners

0:25 accomplish more

0:26 in today's conversation i'll be joined

0:28 by uh sharon mandel ceo of juniper

0:31 networks uh jerome say hi

0:34 hi uh thank you for having me i'm

0:37 excited to be here today

0:39 and we're also joined by juniper

0:40 network's ambassador christian scholes

0:42 of axion networks and solutions

0:44 uh and so what you say as well christian

0:48 yeah hi everybody um yeah thank you for

0:50 having me um great to be here

0:52 and it's a it's a it's a great duo to

0:54 have because they both share some

0:55 takeaways on what it's like to actually

0:57 implement the ai and work alongside it

1:00 and uh you know sharon i'll start with

1:02 you um

1:03 i you know you and i have been talking

1:05 about ai for quite some time and it's

1:07 kind of a funny topic because

1:09 i think everybody kind of conceptually

1:11 understands the value although there's a

1:13 lot of skepticism and

1:15 uh trepidation around it so

1:17 just talk a little bit about the process

1:20 you went through rolling out mist inside

1:22 juniper well sure well rolling out mist

1:25 is is

1:26 fairly straightforward um because a lot

1:29 of it comes out of the box so

1:31 um i think for a lot of people

1:33 they hear ai they think oh my god i have

1:36 to hire

1:37 you know an army of data scientists i

1:40 have to do

1:41 all of this work and

1:44 in this case the ai is built in and so

1:49 it

1:50 it it starts with

1:52 models that are trained and then it gets

1:54 to understand your network and

1:57 so it's it's not a lot

1:59 different than um

2:02 installing any other wi-fi network

2:04 except that

2:05 it's easier because we we build in the

2:08 pre-configuration and and make that that

2:11 simpler i think the other challenge is

2:13 you know people start to

2:15 wonder oh my god you know are my is my

2:18 job gonna go away as a network

2:20 administrator

2:21 and i think that um very quickly people

2:24 understand that that's not true right

2:27 because

2:28 um

2:30 the the situation was they always had

2:32 more work to do than there were people

2:35 to do it um they were solving problems

2:39 um that you know were like finding a

2:41 needle in the haystack because the scale

2:43 of the data that had to be analyzed to

2:45 solve them was so large and so really

2:48 what what introducing ai into that mix

2:52 did was it took away

2:54 the mundane problems it made some of the

2:56 very complex ones much easier

2:59 to solve and then they got to go back to

3:01 the work that they really like doing

3:03 which is architecting and designing

3:05 networks not

3:07 um troubleshooting them

3:09 yeah it's it if the analogy you used

3:11 about it sort of being in bed it's

3:12 interesting because it's not unlike an

3:14 automobile today right there's ai in

3:16 your car that helps with lane change

3:18 alert parallel park assist you know

3:20 adaptive cruise control and it's not

3:22 like we need to be data scientists to

3:23 drive today we just need to go in and it

3:26 makes us smarter better drivers

3:27 similarly it sounds like your

3:28 implementation of ai made your engineers

3:30 smarter better engineers

3:32 so uh christian uh now you've been using

3:35 this a couple of years now so why don't

3:37 you talk about your journey

3:38 yeah um i started with uh

3:40 i was around 2020 when i started with

3:43 mist

3:44 i started with adopting it in my home

3:47 so i built this new house for me and my

3:50 wife

3:50 and obviously it

3:52 always was my dream to build a smart

3:54 home a true smart home so i started with

3:56 roughly 200 sensors i'm currently at

3:59 roughly 300 to 350 sensors

4:02 and yeah as you can imagine with that

4:04 amount of sensors

4:06 you don't want to administer all of them

4:08 by hand so you basically want to to

4:11 architect what you want to achieve like

4:13 i want a certain temperature or i want a

4:15 certain air quality or

4:17 yeah whatever you feel comfortable with

4:19 and then you want the sensors to

4:21 interact with each other and the

4:23 orchestrator or the ai part to basically

4:25 control all of that

4:27 to yeah

4:28 fulfill exactly what you asked it to do

4:31 that's interesting so uh i can't wait to

4:33 see that's the next analyst summit we

4:35 should do at your house um

4:41 and uh another question for both you and

4:43 sharon i'll start with you again um how

4:45 long did it start seeing did it take to

4:47 start seeing value from ai

4:51 well you know each use case is a little

4:53 bit different right so um you know you

4:56 asked me specifically about miss that

4:59 happened very very quickly because it

5:01 came with a certain

5:03 um

5:04 set of um

5:07 features and and and problems it could

5:10 solve straight out of the box right so

5:12 so immediately we we could see

5:15 um

5:16 we could see a

5:19 ticket resolution things like that um

5:21 other solutions uh have different

5:24 requirements to get them going so

5:27 um as as we've implemented some chat bot

5:30 technologies right which people consider

5:32 ai because in theory

5:35 you know it's it it it's it's talking to

5:37 a machine and getting answers um in that

5:40 case there there was a lot more work to

5:43 do to get effectiveness and maybe we

5:45 never with with the particular

5:47 technologies we've played with so far

5:49 we've never gotten quite as far as we'd

5:51 like because

5:52 there was so much work we had to do

5:55 which required an army of people we

5:57 didn't have

5:58 to prep the data

6:00 to create a knowledge base to structure

6:02 questions properly

6:05 that

6:06 it

6:07 we never really reached that full

6:09 potential so i think you know people

6:11 have to understand that there's a range

6:14 of of ai type solutions all the way back

6:18 to the the simplest which is just

6:20 um you know predictive analytics right

6:23 where where you're using

6:25 ml models to

6:26 to draw conclusions and take next

6:28 actions and in that case

6:30 um you know you may be building it from

6:33 the ground up and then really preparing

6:36 the data is the largest job and can take

6:39 some time so i think each company has to

6:42 understand

6:43 um the value of the use case that

6:45 they're trying to solve really

6:47 understand the products they're buying

6:50 what needs to go

6:52 into

6:53 um

6:54 training the model there is it is it

6:56 trained out of the box and then is gonna

6:59 adapt and learn more about you and get

7:01 more specific about you

7:03 over time or are you building that model

7:06 from scratch because there's no way that

7:09 that problem can be solved because it's

7:11 truly uniquely about your company and

7:13 then there's going to be a different

7:14 level of effort and you ought to be

7:16 convinced that the problem you're

7:18 solving

7:19 is has that business value that that's

7:21 worth that additional effort

7:23 yeah so it sounds like there's in some

7:25 ways two types of ai there's the

7:27 embedded

7:28 uh kind of invisible ai that people

7:30 don't even know is there that helps the

7:31 network run better

7:33 right and then there's other ai that

7:35 the engineer would interact with that

7:37 does need a bit of tweaking and training

7:39 and things like that like with chat bots

7:41 that actually changes the way they work

7:42 is that a good way to think about it

7:44 yeah and again you know ai when i i'm

7:47 speaking about ai i'm not speaking just

7:49 about ai to serve the network but

7:51 ai to solve a broad range of problems

7:53 right so we use ai to make decisions in

7:56 supply chain

7:58 we use

7:59 you know the chatbot example

8:01 we have chat bots at the help desk we

8:03 have chat bots through the procurement

8:05 tools we have chat bots you know to help

8:08 our employees navigate

8:11 how to find things in in the marketing

8:13 department so

8:14 they get applied to all different kinds

8:16 of problems and each

8:18 implementation um you know watson is

8:21 different than servicenow's um chatbot

8:24 which is oracle's chatbot and what they

8:26 expect you to do and in some cases

8:29 some companies that provide those

8:31 technologies out of the cloud

8:33 they bring as part of their professional

8:35 services offerings the people who do all

8:37 of that data work because they've done

8:39 it a bunch of times before with

8:40 companies and they're not learning on

8:41 the fly so there's there's a whole range

8:44 of level of effort depending on the

8:46 nature of the solution

8:48 yeah the ios caution people do when

8:50 they're using ai to interact with

8:51 customers

8:53 make sure it works well because you can

8:55 actually have a very negative reaction

8:56 if it doesn't work well i think we've

8:58 all experienced that chat bots that

9:00 you know don't really understand what

9:01 you're saying but they've come a long

9:02 way you know in the last five years and

9:04 so i think for companies that had a bad

9:06 experience with let's say five six years

9:08 ago it's really good again i'm gonna

9:10 bounce back to you what was your you

9:12 know you set up in your own house which

9:14 is kind of cool can you just kind of

9:16 double click on the learning process

9:18 that you went through for your iot setup

9:21 yeah well at first it was just adding

9:23 all the sensors and that's basically

9:25 where all the magic starts because at

9:27 the beginning i just had a bunch of

9:28 sensors and yeah i had a pre-trained

9:31 model for this whole ai setup but it

9:34 wasn't really aware of what i wanted in

9:36 my house so it was in this constant

9:38 learning phase for about i would say six

9:41 to seven months

9:42 um

9:43 and at some point i even had to tell it

9:44 okay if there is now a snowstorm in june

9:48 for example because sometimes the

9:50 weather is just crazy for one day um

9:52 feel free to ignore that and stuff like

9:54 this because uh yeah if i'm if i

9:56 wouldn't have told it that um snow and

9:59 summer is something that it has never

10:00 experienced before so even for me that

10:02 was very weird so uh

10:04 yeah

10:06 and it was very interesting because at

10:08 at the beginning it's it almost feels

10:10 like a very

10:11 dumb piece of software

10:13 um and then the more you interact with

10:15 it the more you train it the more you

10:18 work with it every day almost like you

10:20 would work with a new co-worker the more

10:23 it learns about your let's call them my

10:26 needs like what temperature does he like

10:27 or what humidity does he like and

10:30 when do i need to open the window and

10:31 close the window when do i need to shut

10:33 the blinds because it's getting too

10:35 bright and stuff like that um

10:37 yeah now it's just working it's like

10:39 magic i mean i'm walking through my

10:41 house wherever i come into the room the

10:44 light gets turned on when i leave the

10:46 room the light gets turned off and yeah

10:48 it basically learned all of that from

10:49 the behavior that i showed it

10:52 so uh yeah yeah

10:54 that's interesting

10:55 i find sometimes those systems can

10:56 always get too smart i've got some smart

10:59 stuff

11:01 oh yeah

11:02 it shuts things off and you don't want

11:04 it to and things like that but i guess

11:05 that's that's all part of the learning

11:07 process what does it take to sort of vet

11:09 an ai solution or vendor like how do you

11:12 evaluate them

11:13 yeah so

11:14 i think first of all

11:16 you have to really understand the

11:18 problem you're trying to solve

11:21 and

11:22 then

11:22 understand

11:24 what went into the vendors thinking

11:27 about the problem they were solving and

11:31 the first thing is are they

11:33 closely enough aligned that

11:36 the models that got baked into theirs or

11:39 that they're going to apply to your

11:40 problem are actually applicable and

11:44 going to solve the problem after that

11:46 you then have to especially in the case

11:48 of sort of the prebaked

11:51 ai you need to understand the types of

11:53 data that it was trained on and whether

11:56 again that data closely enough

12:00 reflects or is broad enough

12:02 to at least start creating that value

12:05 out of the box so that it can learn and

12:07 continue from there you have to

12:09 understand

12:10 how does the vendor think about keeping

12:13 its models up to date so you know you

12:15 saw

12:16 during the pandemic certain ai solutions

12:19 that

12:20 were built based on

12:22 well-understood behaviors about how the

12:24 world worked

12:26 so you know for retail there were you

12:28 know years and years of data about

12:30 people going in and out of stores all of

12:32 a sudden it stopped well since the data

12:34 is the fuel for the model suddenly that

12:37 model didn't work anymore in answering

12:40 some of the questions that retails one

12:42 example you saw

12:43 you saw a number of others obviously

12:46 certain things in the supply chain

12:47 changed pretty radically so

12:49 you have to understand where data

12:52 discontinuity might break that

12:54 technology

12:55 and what's your likelihood for that so

12:57 it's it's it's really all about

13:00 primarily that many of the algorithms

13:02 that are the basis for ai

13:05 they're well understood statistical

13:07 analysis and so first of all you know

13:10 are they applying the right one to the

13:11 problem most likely if they've had any

13:14 success in the market they are but then

13:16 it's really about the fuel and the data

13:18 that that feeds that model

13:20 how do they deal with the drift that can

13:23 happen so those are some of the

13:25 considerations i think you know you need

13:27 to look at but

13:28 it's always first

13:31 does does the solution match my problem

13:33 and do i really deeply understand my

13:35 problem that i'm trying to solve and i

13:37 think that underscores the importance of

13:38 pre-built uh ai systems uh obviously

13:42 you know the the more

13:44 turnkey it is the

13:46 the easier it is to implement i think

13:48 you know from an engineer's standpoint

13:50 where i've seen

13:51 some hesitation uh from it pros

13:54 is that uh there's a fear that you need

13:56 to be a programmer and a data scientist

13:58 and frankly um as part of the media

14:01 right i read a lot from media the we

14:04 that was a big theme many years ago that

14:06 everybody in the world would turn into a

14:08 programmer and everybody would be have

14:10 to become a data scientist

14:11 and i think that scared some people and

14:13 that's not true i do think from a skill

14:15 set perspective though um

14:18 engineers do need to think more about

14:19 becoming a software power user right but

14:22 that's with or without ai you look at

14:24 the sophisticated network systems today

14:26 not just from yourself but even from

14:28 your competitors everything's you know

14:30 tapped into with apis you know every you

14:32 know everything's available via sdks and

14:34 things and so um you know software is

14:37 the way forward and that lets you access

14:39 a lot of these great features so like i

14:41 said you don't need to be that the data

14:42 scientist and things like that but you

14:44 do need to be comfortable with software

14:45 so

14:46 um you know and christian is that would

14:48 you like to you know maybe expand your

14:49 own insurance i do

14:51 yeah i mean it gives you a lot more

14:52 flexibility with this new skill set that

14:55 you build your like you said you're

14:56 becoming more of a designer you don't

14:58 need to be the one

15:00 um coming from a network implementation

15:02 point of view where you go to every

15:04 switch and configure every vlan on every

15:06 port you simply design okay this is the

15:08 vlan that i want here are my devices go

15:11 for it um so yeah you have the weekends

15:14 back you have all those night shifts

15:15 back so uh

15:17 i think it's a it's a great thing so

15:19 it's not threatening the jobs at least

15:21 from the way that i see it and what i've

15:23 seen with customers

15:25 is it's not threatening the job it's

15:27 giving them more air to breathe so they

15:29 can focus on the important business

15:31 things and not need to waste any time on

15:34 those yeah little

15:36 times the reality is i'm not going to

15:38 sit here and say that

15:40 the infusion of ai in technology isn't

15:43 going to get rid of some jobs because it

15:45 is but

15:46 this this is just part of technology

15:48 maturation we don't have mainframe

15:50 engineers today we don't have people

15:51 working on token ring you know things

15:53 change over time and everything right

15:56 and every engineer i talk to the one

15:58 piece of advice i give them is if you're

16:01 doing things today forget about your the

16:03 company you work for think about your

16:04 own selfishness your own needs and your

16:06 own job if you're doing things today

16:08 that aren't

16:09 additive to your resume right don't do

16:12 them find a way to automate them out of

16:14 your job and that's really the power of

16:17 what you were talking about christian

16:18 that ai can do it's like you don't have

16:19 to come in on weekends you don't have to

16:21 spend hours and hours updating vlans and

16:23 frankly nobody's going to hire you for

16:25 your next job yeah because you can do

16:27 those things right so

16:28 exactly

16:30 and when i think about um you know ai

16:32 and the network i do think one of the

16:33 things to look for though is a solution

16:36 that works across the network it is

16:39 valid because the experience that a user

16:42 or customer gets is based on how

16:44 applications work across the network not

16:47 just within specific silos and

16:49 historically we've treated networks and

16:51 styles and that's why even today you'll

16:53 see

16:54 ai solutions for wi-fi ai solutions for

16:56 sd-wan ai solutions to the data center

16:59 things like that and and while those

17:01 may

17:02 work

17:03 well they don't really give you a view

17:06 of the end-to-end network and so you

17:08 know in data science says there's an

17:10 axiom right that states that good data

17:11 leads to good insights well

17:13 silo data will lead to solid in size or

17:15 fragmented insights and i think uh you

17:18 need to have that kind of end-to-end

17:19 view so here i'm just i don't know

17:22 yeah so so zeus

17:25 i i

17:26 as kind of a cio who has to not only

17:29 integrate the network but

17:31 a broad series of applications and

17:34 experiences both for our customers and

17:37 and our employees

17:38 um

17:39 across a broad spectrum of you know

17:42 functionality

17:44 because

17:45 you know the pre-built ai solutions

17:49 they tend to be fairly narrow in focus

17:51 again because you have to curate

17:54 uh a data set

17:57 um that's

17:59 unique to that problem and so

18:02 we haven't yet seen the company

18:04 you know sort of like oracle and sap

18:07 came in and said well i'm going to

18:08 integrate everything from financials

18:11 through supply chain and then you know

18:14 work their way back to you know order

18:16 management the sales front office right

18:18 you've seen these companies come in and

18:20 and provide

18:22 application suites that that you know

18:25 cover very broad spectrums of

18:26 functionality and it took them years to

18:28 build it you haven't really seen that

18:30 use it you know across

18:33 that sort of cross-functional ai view so

18:36 you know i i joke sometimes that we're

18:38 going to have the war of the bots right

18:39 where

18:40 the bot from vendor a is gonna give you

18:42 a different answer and start fighting

18:44 with the bot from vendor b

18:46 um and i what i what i see starting to

18:49 happen is um some of the

18:52 sort of incident management solutions

18:55 that people tend to interact with right

18:57 so a person submits a problem

18:59 and then a person's on the receiving end

19:02 of that you know it might go through

19:03 some escalation path but ultimately you

19:05 get to a subject matter expert

19:07 who

19:08 you know will solve your problem

19:10 well

19:11 you know as you've seen in this world of

19:13 sre and devops it's rarely one subject

19:16 matter expert if the problem is complex

19:18 enough well now some of these solutions

19:20 are starting to treat the bots as

19:23 um

19:24 a part of the conversation um and so you

19:27 might have one or more bots combined

19:29 with five people

19:31 offering insight into

19:33 what might be the solution to the

19:35 problem

19:36 but that's still a step away

19:38 from

19:39 you know some kind of integrated

19:41 perspective or

19:43 integrated world view that i think will

19:45 will ultimately need to have

19:48 in certain areas right um

19:51 but it is because of the nature of how

19:53 these solutions are built um and how

19:55 important curated data sets are

19:58 yeah i think that's uh

20:00 the the war that both already happens in

20:02 social media

20:07 we just we don't want to come to our

20:08 corporate systems though so

20:10 um

20:11 anyways i just have one more question uh

20:13 for both of you uh and we'll do this um

20:16 you know i'd like to ask this question

20:17 because i think it's helpful for people

20:19 watching this uh what's one piece of

20:22 advice that you would give to it pros as

20:26 the 19 leaders as they embark down their

20:29 i.t journeys and that or their ai

20:31 journey they'll actually go first and i

20:32 think

20:33 it kind of parlies off what christian

20:35 said

20:36 um i think you need to be patient with

20:38 ai and you need to understand um and

20:40 this can actually help you evaluate real

20:42 ai solutions versus fake i solutions ai

20:45 washington is rampant out there and

20:46 every system used says they use ai

20:49 and a good way to understand if

20:51 something's real ai or not

20:53 is to see how it progresses and how it

20:55 learns over time as christian said at

20:57 first it may be kind of dumb you may

20:59 look at it and go well this isn't very

21:01 helpful at all but over time

21:03 it does get better and it gets smarter

21:06 and it understands your environment

21:07 better in fact i think juniper does a

21:09 nice job of highlighting the efficacy of

21:12 it and you know being actually

21:14 demonstrating you know the

21:16 the quad and quantifying the how it's

21:18 changed over time and so real ai isn't

21:21 just fancy rules based systems and if

21:23 you go with them one of those those

21:24 rules based systems um you're going to

21:26 be constantly updating it and constantly

21:29 changing the rules

21:30 um and so you have to understand that ai

21:32 is not perfect out of the gate but it

21:34 does get better over time and it will

21:35 help you more and more and so you need

21:37 to be patient with it and so sharon

21:40 what's some advice from you

21:42 the hardest part of adopting technology

21:44 is fear of change right and

21:47 and

21:48 to your point that you made earlier is

21:50 this which was

21:52 you know yeah some things are going to

21:54 change and some types of roles and work

21:56 will go away

21:58 um

22:00 it's part of the natural evolution of

22:02 technology so as you know i i say

22:05 experiment right try try different

22:08 things try different flavors see what

22:10 works for you

22:12 understand the nature of your company

22:14 and your business

22:16 and experiment um

22:18 and and some things are going to be you

22:20 know as we've discussed much more useful

22:23 straight out of the gate and some are

22:25 going to be um acquired taste because

22:27 the more the more unique

22:30 you know where

22:31 you when you apply

22:33 ai technology to something that makes

22:36 you really special and unique you're not

22:38 likely to find it out of the box right

22:40 again because of that

22:41 curated data so

22:43 you know i say kind of play with a lot

22:45 of different things and and and try

22:48 um and learn yourself from them um where

22:52 you're going to have to change and adapt

22:55 and

22:56 um

22:57 like i said just don't be afraid and get

22:59 people comfortable

23:00 um slowly and that that may go back to

23:03 your patience right i mean as is always

23:05 the case

23:06 um

23:07 technology

23:08 at some level it creates that that's

23:11 disruptive creates change that people

23:13 are uncomfortable with and i think part

23:15 of being patient with ai is being

23:17 patient with the people you're trying to

23:19 get

23:20 to use it so um a lot of experimentation

23:25 experiment and uh be patient experiment

23:27 christian uh what advice would you give

23:29 here yeah i can also echo that don't be

23:31 afraid of it um when i when i started

23:33 with this it almost felt like i'm

23:35 teaching a child or a new colleague all

23:38 of the stuff and at first it felt like

23:40 i'm doing more work

23:42 and i thought like okay if i now have to

23:44 explain everything and basically teach

23:46 everything will this really benefit me

23:48 but after some time you will see the

23:50 benefit and

23:51 the model learns and everything comes

23:53 together and yeah like i said you will

23:56 get

23:57 more

23:58 more free time to do the stuff that you

24:00 really want to do and less of the

24:02 hideous little tasks that yeah just

24:05 waste your time so be patient work with

24:07 it

24:08 um yeah make it your colleague basically

24:11 explain everything at first it's it's

24:13 yeah you need to explain everything of

24:15 course because for the ai it's brand new

24:17 what you're just telling it

24:18 depending on the type of model that you

24:20 use but uh yeah it will be

24:23 so

24:24 worth it so don't be scared

24:27 all right well uh that was a great

24:28 discussion i'm just gonna i think wrap

24:31 up by saying uh

24:33 alekko said don't be afraid of it ai has

24:36 already proven in many industries it

24:38 makes us better drivers it makes contact

24:40 center agents smarter right and i think

24:42 from an engineer's perspective it's

24:44 going to make you a better smarter

24:46 engineer it's going to let you do a lot

24:48 more fun things right and as a former it

24:51 pro i can tell you nobody here in it pro

24:54 to sit and look through router logs and

24:56 update vlans all day long to just that

24:58 just boy that's just if you get

24:59 enjoyment out of that maybe

25:01 you know um you know

25:03 you should you know seek out some other

25:04 forms of enjoyment or something so

25:06 anyways uh ai's coming uh there's really

25:09 you know it's gonna be infused

25:10 eventually in everything that we do it

25:12 already is on the consumer side so uh

25:15 you know christian sharon thanks for

25:17 joining me today i'm zeus caraval from

25:18 ck research and thanks for watching this

25:20 episode of ai skeptics

25:24 [Music]

25:29 you