Generative Ai And Digital Twins In 2024 Xmpro Webinar

Join XMPro's latest webinar on Generative AI and Digital Twins in 2024, where we dive deep into the transformative powers of Generative AI and the evolution of Digital Twins.

Hosted by XMPro...

Transcript

Join XMPro's latest webinar on Generative AI and Digital Twins in 2024, where we dive deep into the transformative powers of Generative AI and the evolution of Digital Twins.

Hosted by XMPro... welcome everybody um to today's webinar

um thank you for joining I'm Peter F

Scar and we'll talk about gen Ai and

digital twins in

2024 so with J AI uh

Steve hoft um Co actually reminded me of

this when you look at something like

geni and the impact of that um the the

what comes to mind is that it's almost

like magic so if you look at what's

what's happened over the last um year or

so in the space significant changes and

for me it's not so much um around the

the um the the the magic side of it it

is more the impact that it's got so and

it reminded me of the book uh who who

Moved My cheese where um it's all about

change and how we we adapt to new things

that are that are coming in so in that

story um when the cheese was moved the

the target the goal the the the

environment some like the the mice

adapted very well and some people didn't

um and yeah so this is kind of bringing

that um that thinking around what the

impact of digit of of generative AI is

and if you combine that with digital

Twins and how it's changing the

landscape the the changes that we see

especially in 2024 revolve around three

main areas the first one is we're seeing

a a a ground swell of challenges around

skills so lack of skills aging Workforce

in maturity and I'll drill down a little

bit on that but there's some significant

ific challenges ahead in terms of

getting people to work in manufacturing

likewise with the acceleration of of AI

and we've seen a lot of things like

co-pilots and and that happening last

year this year we'll see a lot of

experimentation moving to actual

implementations and adding value and

it's not just the novelty Factor anymore

but um the the the applications that

come out of it and lastly we also see

that the industry for. digitization

digitalization is picking up a lot of

momentum but is it really adding a lot

of value um and what we need to do in

terms of to boost productivity which is

a key challenge in the industry um we've

seen many of the developed countries are

either um stagnant or declining

productivity so how can we get more with

less people which is really what that

what that statement is about so those

three things is creating this perfect

storm for for how I believe um

generative AI will disrupt how we work

going forward if I look at the uh change

in Workforce in the next 10 years this

is from Gardner and I'll just summarize

it into those three into those three

things the workforce um is actually

growing slower than what the GDP is the

manufacturing Workforce is shrinking in

terms of real numbers so getting people

to come in and work in in manufacturing

and Associated Industries in the in the

industrial side is actually shrinking

and and one of the other statistics is

that one in four people in the next um

in the next 10 years in the next decade

will actually be over 65 that is still

in the workplace uh because of this lack

of skills now again looking at some of

the things that g is saying around what

are the type of jobs or tasks that we

can automate and the ones that are

ideally suited um to completely

digitalized you can see the bottom Green

in each of those production schedu and

quality so these are the type of tasks

that lend themselves to being um

automated to some degree and I'll touch

a little bit on how we at X and pro are

looking at at at that part of the

problem so I think generative AI um you

know people are saying well will this

replace jobs will that put people out of

work our real problem is we don't have

enough people to do the work that we

have there's a decline as I showed

previously but what was interesting Sam

Alman um CEO of open AI which also um

owns uh jgpt which is kind of

skyrocketed this whole thing uh made the

statement that AI tools are very good at

doing tasks but terrible at doing jobs

for now so the the focus of the AI tools

is not to replace jobs not to replace

people but to look at what of those

tasks that we can actually

automate um and and use the uh um

generative AI to be able to support us

in doing those tasks so we can get

people to do some of the other more

value added tasks so just on the if you

look at CH GPT and how quickly this is

on us and this is fascinating if you

look at how long it took Netflix to get

to to a million users and Facebook and

Instagram and and Chad GPT this is

phenomenal 180 million users and they

reckon there's about 100 million people

signing in every week um to use the

service so phenomenal with that if you

look at someone like Michael Dell who's

been around um the industry for a really

long time

um and he's seen many many waves what

he's saying is so far um there's nothing

that competes with what he's seeing on

this um on on what's happening with j

really interesting stat that I saw from

my Carol who's with um Georgia Pacific

um is you know how the productivity

which is really the output per hour how

that has changed over the last couple of

years you can see every time we had new

technology internet of things all of

that we get more for the labor that we

put in the challenge that we have is

this is gone completely flat um and his

point is that the more we put just

dashboards and and Analysis and

putting bi over um the information we've

not made any significant progress since

we've done that so what do we need to do

to get a step changed and it remind me

reminded me something or of the of the

great crisis in London New York and

places like that where they said in in

1894 that in 50 years London will be

buried under 9th fet of manure um and

there were conferences and a lot of

debate and what are we going to do

because we looked at the problem from

the horse perspective by 1912 Henry Ford

um got um the way that we make cars into

a repeatable process and since then that

problem has gone away so not within the

50 years but it completely changed and I

think gen AI is that complete change

that we seeing so this combination of

the three

trends that creates this this perfect

storm and if you combine this with

digital twins to shift towards digital

twins so Dr Michael Greaves um explained

you know what he sees as a intelligent

digital twin which compared to

traditional ways which is passive

offline goal given and predictive he

sees a digital twin which is active

always on agents real time goal seeking

versus and goal given and doing things

like front running simulations being

able to do that so it's not so to get

back to kind of the cheese and

analogy um it is not a better Mouse

straap it is a completely different way

that we have to think um so it's not

about the HSE manure problem it is about

how do we how do we change that whole

dynamic and address um the creating the

the vehicle change that we need so it's

not uh smarter repositories with more

data if you ask anyone they just want

more data it's not about um measuring

things like more kbis or building better

AI models or analytical models it's

about a whole shift and that's what we

are focusing on so our Focus um and how

I think geni will impact digital twins

is really in three area so how do we

generate twins so I call that generative

twins so how do I create a twin right

from getting the requirements designing

it building it testing it and putting it

in place the second part is how can I

make it smarter how can I put

intelligence into it because there's a

massive amount of information available

in large language models and you know

the the way that the that that g CH GPT

type applications are developing and the

information that sits in there billions

and billions of data points that it's

been trained on much better than any any

users how can we use that to provide

better co-pilots help with Advanced

analytics and decision making and things

like that so how do we not just build

but also create the intelligence and

then lastly how can we get them to do

some of these tasks and work for us so

we see this and um the term multi-agent

generative systems are being raised at

the moment by organizations like Gart

and other research it is very new um but

this is actually when you have multiple

autonomous agents that work together in

this goal-seeking

approach um but they have to work with

inside Rules of Engagement so I'll

quickly go into each of these and just

give you a view on what they are and

what we are doing as X and Pro in this

space so the first one is how do I

create um a digital twin and the the

reason that we see or the the benefit of

this is really quite often people are

stuck at how do I get started so we this

it helps you with with getting started

it helps you to collaborate around key

objectives that this that your solution

need to do and don't wander off around

one person's bias or a technology that's

really interesting it Focus around key

objectives it helps you build these

things a lot faster we anticipate around

about a 50% so we're still in testing

all of this but we anticipate around a

50% faster development overall um right

from requirements Gathering to get it

into into running and it helps you

sometimes to think out the box because

we are so trained and biased around you

know the world that we

know um that similar to what they had in

London New York and places like that

previously do is work in progress for us

so uh it you'll see this being released

throughout the course of the year um all

the things that I'll be sharing with you

so looking at this generator process we

broke it down into six steps so to get a

co-pilot to help us um defining the

requirements doing the data management

things like generative integration how

can I discover autodiscover apis and

actually build the mapping for me um

come up with what are the

recommendations that I that will help me

to identify the problem and and and and

um help me um provide prescriptive

analytics some ideas around what should

be on the ux how do I align that with

the user stories um and then actually

create the configuration actually create

the ex in our instance XM Pro Data

stream configuration files um and some

of the other IDE defacts that we need

the rec the recommendation files so that

those can be imported into the process

um and help you speed up

so uh and also gr documentation and then

lastly make sure that we can put

security and governance auditing use

this as an auditor to put that over this

as well so here's an

example um this is inside X and pro and

again I think this process is pretty

generic how we've implemented or how we

are implementing this is there are

multiple different use cases that you

see here you can see the five steps at

the Top If I choose one of them I can

then go into it there are the six

objectives and each of these will then

take me through a series of question

questions and um and prompts I can

prompt and it can prompt me back so if I

look at for example identifying key

objectives and

challenges so it'll start off by how do

you describe the business of problems

and so and there's a set of of questions

built on our methodology that we have

found very effective in working with

customers so how can we bring that and

make that available to anyone who's

trying to do this in this instance for

my cheese manufacturing organization you

can see a number of number of questions

and then at the end of it for step one

it'll start summarizing the results for

me so but that is learning so that I can

put this all together at the end and

actually create that configuration file

so and here are just some of the outputs

um of that so once that is done I can go

into the next um Step of this process so

being able to generate twins we

anticipate about a 50% reduction in time

and again this is new um you can help

you identify unanticipated needs it's

one of the benefits that we saw out of

it sometimes you don't think of it and

and uh the intelligence that's in this

agent allows you to to or will

interrogate will ask you questions which

you might not have thought of and the

real thing is that it can help you build

a digital twins around the real problem

statement here's an example of what came

out of a session like that where this is

the problem statement around the you

know what what where who why you know

what is the problem why you're doing it

how you going to do do this who is it

for and what are the key benefits it is

really good at taking all of those

questions from multiple inputs in

different stakeholders and helping you

to put this together so that is how we

build the twin the next part is how do

we put some intelligence into the twin

um how do we how do we um create smarter

twins so we don't just build them but

now we make them smarter and um I know

it's a Cheesy story but um I'll get back

to kind of the life of um

a day in the life of a cheese plant

operator who's assisted by co-pilot and

this co-pilot is really good as a

process advisor as you will see so

here's our stock standard dashboard for

a or digital twin um representation of a

um a cheddar manufacturing

um uh uh operation and you can see

across the top of the process I can see

different batches where they are what

the statuses oee all the normal things

that I that I like to see inside um my

normal application at the top right I

have an advisor if I click on that it

now opens up um a area where it will

actually look at what is currently there

what is what are the current um um uh

instances and what you'll see on the

recommendation side on the right hand

side out of the box X Andro allows you

to create recommendations with

recommendation rules and some of our

customers have actually got millions of

recommendations over three or four five

year period that they've gathered we can

interrogate this also with the digital

twin in this instance looking at the

process data um it can recommend what

queries you may want to run but I want

to and in this instance I'm asking it

what pasturization temperature and

duration yielded the best result in the

past so I'm looking for this golden

batch um in all my previous ones and see

if I so that I can that that I can build

a solution around or um a golden batch

approach so it will then interrogate the

data that is sitting inside our

application and this is um and it

provide me different ways of

demonstrating and showing that um

information whether I wanted graft or

and again this is all built into how I

can ask it certain questions around this

I may see that there's an issue with the

pasturization so you can see at the top

at the top there's a red box that um

shows me there's a there's an issue um

so that pops up and that's all part of

our standard recommendations you don't

have to do anything to get that

functionality there no AI involved in

that you can if you want uh put some AI

into the recommendations but those are

stock standard recommendations what I

tells me there's an issue with this

machine I can now go in and I can use

traditional AI where I do remaining

useful life and run regression models

and classification models and all those

things which is again standard in X

Andro you can get get that right now

there are some previous webinars that

discuss that in this instance um what

I'd like to do and having a trained on

our own company data our own operations

manuals our own recommendations uh I can

now start interrogating this so show me

that the temperature Trend around this

for the last 12 months and um with some

interpretations of what does it mean so

two instances were out of had anomalies

or whatever the case might be and then

um lastly show me this in a in a plot or

something these are all natural language

questions you can ask on your data at

any point in time so that's adding

intelligence to the digital twin now

just imagine so this is a water

treatment plant you can see there's a

camera with video on the side that

actually just imagine what it's like if

you could get the if you could get that

advisor to interpret what is actually

happening in that video and what does it

mean um so building that expertise at

becoming really really valuable assets

and workers in the organization it's not

just on the front end part of it but

expro has notebooks and we've had this

capability for a while where I plug in

ch GPT and I can do my analysis inside a

notebook which is where data scientists

engineers and people like that like to

take the expra data and and um and and

build models or evaluate or or or or

look for things and um in this instance

um I can ask it a question so

in Notebook using um exm Pro there's a

uh I can ask it how to how do I create a

correlation Matrix and again this is how

it creates the the python code for me I

can I run that then in the notebook and

here there's my visualization I didn't

have to know I don't know how to write

the the the code to actually create a

correlation Matrix um CH GPT did that

for me so that's the other area of a

co-pilot it's not just about your data

but also how to help you um um create

more intelligence or better questions so

again looking at this we see we

anticipate around about a 25%

productivity Improvement for people

using this we it's actually should

actually be more but um it also um help

you better decisions lower

cost better employees satisfaction and

it helps to create the self-service

analysis instead of going to an IT

organization and get them to write you a

SQL with a report that sits behind it so

that self that self-service analysis

really coming to the front the last part

is around how gen AI will impact digital

twins in 2024 is now it should do the

work not only do we create it we make it

smart but how can we get multi-agents to

do this now this is really um at the

Forefront of of where research and

everything is um it's a fascinating area

I personally believe this is the area

where there's most value because we're

taking the robot out of people and

making the world more human and that is

according to Leslie W Cox at London

School of Economics I love it um in

terms of what that approach means for XM

Pro we've seen that decision

intelligence is not just about

dashboards which is the support or the

decision augmentation which is kind of

bringing in these co-pilots but how can

we actually get in certain operating

conditions the the application to or the

the digital twin to run in a safe

operating environment for certain tasks

actually perform those tasks

autonomously in a lights out way and we

see this going from a informant approach

to a performant um solution going

forward so just imagine that cheese

factory again and in this cheese factory

I've got multiple agents now in a

traditional rule-based system I would

have to write all the rules with what

we're doing is the production agent the

process agent quality agent safety agent

have large language models behind them

so they are really smart around how to

interact how to plan and they have a lot

of knowledge around cheese they have a

lot of knowledge around safety and they

collaborate amongst each other in order

to achieve certain tasks now if you

expand that it's not just about the

production process but I could have the

same in the goods receiving packaging

and functional safety and I do need to

put some rules around it so these these

Bots or agents need to interact with

each other in that they virtual bots in

or or agents that interact with each

other in a collaborative way and they

use large language models as their way

of interacting and deciding what to do

next

um and and we then put the rules around

to say Within These in within these

rules you are allowed to interact with

each other and I'll get back to some of

what these what these

mean the this is how we see the

factories of the future where there's

certain tasks of this that can be done

and you know they can optimize a large

number of variables they're always on

goal seeking and they can reason like

humans in a

way this capability of machines being

able to manage more variables has been

proven by MIT at the machine

intelligence for manufacturing

operations side uh suggest that you have

a look at that but machines are much

better at optimizing a large number of

variables to what a human is there's

research done at places like standard

and Google Deep Mind and others around

this agent based approach how can agents

use large language models as their way

and we don't have to build all these

rules around them they can

discover um this is an example of and

these are real bots so this is a

physical what we are doing is the

virtual how do we build a digital twin

versions of this but these are eight

Bots that um interact with each other

they clean a room and

their focus is you know without having

procedural descriptions at the back but

using large language models on what I

should do what the task is how to break

it down how to actually perform it

fascinating stuff again if you're

interested have a look at Auto RT at

Google deep mine this is not what we're

doing we're doing this

around digital twins how can I create a

virtual approach that is similar to that

how can I get the digital twin to

actually run it instead of having robots

running around and what we've done at XM

Pro come up with a architecture which is

ideally suited to our stream hosts and

how we build applications right now so

we have data streams and stream hosts

and and the infrastructure that we can

plug these agents who can interact with

each other they can use exm Pro

listeners they can perceive in the

middle we have um the the actual we

train them on your company specific

domain knowledge um they have large

language models for their broader

understanding and there's reinforcement

learning to make sure that you know when

they reflect and plan that they that

they learn from what they do like we do

as humans like we do as when a new

person starts in our business it's

exactly the same process and then the

actions could be recommendation so

advising or we can actually um actuate

and switch things on and off um so

that's the that's the future this is

where we're heading and as I said X and

pro ideally suited to do this because we

already have the infrastructure event

based architecture so we are event

driven and we have data streams and data

and and a platform where we can plug

these things in

um this is all uh work from a the um

around the core of digital twins so we

have the digital twins we're building

the the generative agents to be able to

use the digital twins to interact and

and and um do this algorithmic business

and then what we're also doing is make

sure that we have the Rules of

Engagement in a computable way so the

agents can actually understand it so

what are the rules of engagement and

some of them may follow things like

deontic rules or more um so without

these rules you can't play it's the most

important part actually of this whole

thing is how do we make sure that it's

that we that we set that up and um the

way that we do that same same um outline

but it's the obligations prohibitions

permissions delegations all of those

kind of policies and how and how that

applies and how do we make it machine

readable and interpretable and enforced

by the multi-agent so again in the

multi-agent environment we're not trying

to automate not the job but the task so

let's say I've got two agents they need

to fill the milk tanks um they need to

consider power um milk levels

electricity prices maintenance schedules

production so between the operations

agent and the reliability agent they

need to negotiate they may have

competing objectives and they need to

need to negotiate around when to fall we

may then also later on introduce a

quality agent so again how do we

introduce this without procedural rules

that sit at the back um so that we can

address that challenge and XM Pro can

cover this whole Spectrum this whole

Continuum from one end to the other end

um starting with um where a lot of

organizations starts you don't have to

start at the right hand side start on

the left hand side and incrementally

increase and build the capability

towards that area where you can cover

the whole spectrum of what you need to

to support people to augment the

decisions and to take um so again in

terms of the impact of this is massive I

personally believe this is the biggest

thing we can do in this industry is to

help with that initial skills Gap and

the productivity challenge is to take

some of the T tasks that are repetitive

and don't add value um

to um and and and help address them by

helping to build helping to make it

smarter and helping to automate um the

whole process thank you very much um I

hope this has made sense um for us the

cheese hasn't moved we know where it is

um we are we we are going after it

unfortunately um we've run out of time

so there won't be time for questions um

but please contact me if you're

interested in exploring how we can help

you with this kind of problem and how

this could fit into your

organization um thank you very much

appreciate your time and look forward to

seeing you on a future

webinar