Welcome to this new episode of Taxiing Talks on Tour . I'm Sander and I'm at PegaWorld Inspire this week and I'm with Don Sherman , the CTO of PegaSystems . Well , welcome to the show . First of all , great to be here . Well , there's a lot of Gen EI going on at the moment . Right , I heard somebody on stage saying it's Gen EI , gen EI , gen EI , gen EI .
That's the future for development , at least , right ? What's your impression of where we are at the moment when it comes to AI or Gen EI more specifically ?
Yeah , I think we're at a really interesting kind of inflection point , right ? I think one of the things that we are constantly reminding our clients is , in some ways , ai is not new . No , it's in the 50s or something . There have been predictive models and statistical data models .
We are constantly reminding our clients is in some ways , ai is not new . No , they're ounces of fifties or something .
There there have been predictive models and statistical data models what we often call AI decisioning that you know in Pegas world we've been working with clients with for over a decade and , um , I think sometimes a lot of the hype around gen AI and we're going to use you know gen , add around Gen AI and we're going to use you know Gen how to do this , and
that ultimately comes down to well , actually , no , you probably want us a really good predictive model that you can build on your own data set and construct it relatively quickly . You don't need a whole bunch of NVIDIA chips in order to build this thing . It's transparent , it can explain what it's doing and it actually can add real value to your business .
So what I'm seeing is clients really starting to step back and think about how they're going to apply Gen AI , what are the use cases and challenges it can address , and then realizing that it's not a one-size-fits-all solution .
You need to apply different types of AI and different approaches depending on the problem you're trying to solve and the regulations and data world you're running in .
Would a cover term maybe pragmatism ? Maybe look at it from a more pragmatic point of view than last year when we all did all the big stuff .
Yeah , I mean , I think that AI is a transformational technology in much the same way right that the internet was a transformational technology , much of the same way right that the internet was a transformational technology . But we get to that transformation by taking pragmatic steps , yeah , right .
And I think what I'm starting to see organizations do is begin to look for and identify what are those pragmatic steps so that we can actually prove value , we can prove trust and we can begin to sort of build that groundswell needed to drive the transformation we want to see .
It's one of those things that people always say about this is that new technology , the impact of new technology , is overstated in the short run , but understated in the long run .
Yeah , and I think we're exactly in that , and I don't know if generative AI is going to directly follow the Gartner hype cycle . It feels like the hype cycle is moving so fast right now . It feels like a bunch of loop-de-loops as opposed to like a single kind of trough of disillusionment .
But I think we will move through that and we will see use cases and usages that kind of pop out into that plateau of productivity .
Yeah , and do you see , are there learning or learning points that you can take with you from the AI , the old-fashioned AI ? Maybe we should call it now into the Gen AI world that we don't make the same mistakes that we made in the past .
Yeah , Again , I think part of the use case , part of the lessons come back to identify the use case . Right At the end of the day , the goal of an organization shouldn't be to use Gen AI , no , no . In the same way , the goal of the organization shouldn't be to use Microsoft Word or Microsoft Excel .
The goal of the organization should be to drive their business . And if Gen AI can be a tool that adds productivity to their employees , that helps them tackle problems they couldn't tackle before , that helps them deliver better experiences to customers , great . But you have to put it in the context of what's the actual problem I'm trying to solve ?
What's the actual outcome ? I'm trying to drive .
But that's actually . That sounds very logical , right , and I think you made it a point on stage also this week that you said we don't we build it around the case , right , but especially with this kind of technology , the case is also a bit of a fluid kind of concept , right . So how do you do that ?
How can you build around the case if the case isn't yet very well defined ? If that makes any sense ?
Yeah , I think so . I think we've been trying to look at Gen AI through two lenses . One is there's a set of productivity use cases that I think we're all going to want to adopt , right ? A simple example of that is summarization . We know Gen AI is pretty good at summarization . Of text yes , Of text Not necessarily of voice , not necessarily of voice .
Well , if you can get the voice to text , then it can summarize it right . We have technology that can turn voice into text and it's pretty good at summarization . And especially , it's pretty good at summarization if the end consumer of that summarization is an employee who you're still going to hold accountable for .
to make sure the summarization is correct and valid . Make sure the summarization is correct and valid , and so I think we're going to see summarization in lots of places . I see it when I do a WebEx meeting . I get it in my email . We've already put summarization into the contact center app that we have .
That allows you to summarize customer interactions , or in case management at Pega , so that you can summarize what's happening in a case . That's a nice productivity hands-on right .
That's quite generic .
I mean , I think everybody would like to have that right , right and I think , over time , it's going to be something that we just expect , like spell-checking we just expect to have as part of the various apps and tools that we use , and I think it's important for us to do that because , frankly , I think it's going to become table stakes .
But I also think enterprises need to understand that they're not going to differentiate their bank from another bank because they have summarization tools available to their employees .
But do you think it will remain at that point of just the relatively low-level , very tightly defined kind of use cases ? Productivity tools yeah .
So I think those represent a great place to start , because they keep human in the loop , they're safe , there's some measurable business value attached to them . So by all means , deploy them , test them , validate them , but I think the enterprises that are able to push up into more transformational uses of it right .
So , like one of the one of the ways that we see that happening and and again , I think part of what we need to have in this time period is a little bit of an openness to experimentation we see what works and see what doesn't .
But one of the places where we've we see real opportunity is helping organizations think differently about some of the core aspects of their business . You know pega , we are a workflow and decision company , so we tend to think about workflows a lot .
We tend to think about decisions a lot and we know from our experience and probably it's the experience of anybody who's ever been through a process reengineering effort that getting people to align on what a workflow should actually be can be pretty painstaking .
If you're not careful you can end up sort of repaving the cow path right Like there are suboptimal outcomes that can happen .
Well , if I can use Gen AI in partnership with some best practice documentation to recommend a starting point if I want to implement a new workflow , that can both accelerate that kind of requirements design process , so that's good , but it also can push the business to think in new and different ways .
It's funny you mentioned push , because that was one of my follow-up questions . It's like do you experience pushback in that ? Because this is also a human kind of issue that you need to solve to actually start working with Gen AI .
Right , Because there needs to be trust , there needs to be all the things that you need as an employee or as an organization as a whole to say , well , okay , this sounds good , but this is not . Is it good for us ?
And that's why , for example , the tool that I'm talking about here is called Pegagen AI Blueprint , and we just made it available free on our website , and part of the reason for that is we want people to be able to touch it , see it , play with it , respond to it and start to understand . This isn't a magic trick . We're here in Las Vegas .
This isn't like a magic show where I'm trying to astound you and amaze you . I'm just trying to give you a tool and I'm trying to help you see that . Hey , I can think of , in this instance , Gen AI as like a really smart sounding board that I can throw an idea off and it might come back with a different suggestion .
So in that respect , it's good to realize that this is just another tool that you can use . I mean , we shouldn't make too much of it , right .
I mean , I think in general right as a software industry , we like to hype everything right and everything that happens is going to transform everything . That's what I always try to de-hype .
Right , that's what I always try to de-hype .
Right , and I think to your point earlier . I really do think in the long term we're going to look back on this sort of cusp and flip that we've had with Gen AI and say , wow , that changed a lot of things . Yeah , but I think the way you get to that is almost , as you say , by de-hyping it .
Yeah , by saying it's technology , it's a tool , there's always going to be new technology . How do I apply this to the really meaningful problems that I have ? How do I test that , validate it , learn from it , get experience with it ?
And if I do that enough in a continuous cycle , I'm going to look back and realize , oh my gosh , we completely shifted the way we've been working .
But that's the way of the world anyway , right , it's also with your own knowledge . I mean it also builds up over time and after all I know quite a bit about this now . I didn't know anything 10 years ago .
That's more or less the same . It's like , look , if you want to train and run a half marathon , you go out and you run a mile and then you go to run a mile and a half and then you run two and next thing you know it's like , wow , I can run 10 miles . I didn't know I could do that . It's the same kind of idea . So there is a .
I mean just yes . No question Is Gen AI something for every type of company , or is it yes ?
unless or yes , I think . Again , I keep coming back to the analogy of Excel . I really do think in many ways this is going to become a prevalent tool . I don't think any company exists without using some sort of spreadsheet or office program .
If you use that analogy , then also there's the lack of adoption of underusing . Excel is also the problem . Because Excel can do so many things that most people don't even know . You don't want to have that , to repeat that mistake , but you also don't want to overuse Excel , right ?
I'm sure we've all opened up Excel workbooks that have so many macros going on inside them that the thing can't even open on your computer , right ? So , as with any tech it's , how do we apply it ? How do we use it for what it's really good for ?
How do we attach it to a business use case , and how do we not shove so much into it that we're overwhelming our capacity to appreciate ?
it , so that's sort of a process-oriented kind of thing . Right , you say , well , how do we want to use it ? So you need to think about how you want to use it . Right , you say , well , how do we want to use it , so you need to think about how you want to use it .
But are there also technical challenges for companies thinking about starting to use Gen AI in their daily lives , for example , data access , or you know ?
Well , I think one of the important things to understand , or at least in my mind , is the distinction between this decisioning or statistical AI , which companies can and will train on their own data . We heard from the International Australia Bank this morning using Customer Decision Hub to improve customer decisioning based on their own data about customers .
Generative AI is very different . Enterprises , even really , really big enterprises , do not have enough data to train their own gen ai models . They don't have the data , they don't have the chips , they don't have the talent . So gen ai is more going to be like a power source that we plug into . The models are already built , we just use them them .
So the question , I think , for the enterprise becomes how do you use those in a way that interact with your data ? but do it in a way that is safe .
Yeah , and also especially when it comes to RAG , which is , I think , the most popular use case of Gen AI ever since it was introduced last year somewhere last year .
Yeah , so that RAG pattern retrieval , augmented generation , where you basically apply your own content into GenI ?
Yeah , but then you need to decide what is the data that this RAG is going to use , right ? I mean you need to think about that as well as a company you do .
And again , that to me comes back to use case right . So that RAG may be a set of knowledge content that I take and I use that to drive prompts into the Gen AI model .
To summarize for me and that might be really good if I'm trying to solve the access of my customer service agents to content and information , that RAG may be large amounts of customer history because I want to be able to ask questions and inquiry the gen AI about what customer history patterns are happening , right .
So so pulling that data into the rag and I think you know companies are they're gonna need to decide do they build some skills around doing that using technology like embeddings and vector databases ?
and all the pieces that are needed to make that work , or do they sometimes look at cloud-based services , like that's what we have with what we call Pega , knowledge Buddy , which kind of packages that all up ?
So you focus more instead and say , well , what's the content I want to feed into that and how do I make sure that I can govern that content in an ongoing way ?
But then you still need to be able to trust what it gives back , right , obviously , you have this sense of this is my own data , so the data points are correct .
You have this sense of this is my own data , so the data points are correct , because you probably curated it before you say , look there for the answer , but it can still make wrong connections between it can right .
We've seen that . I think there have been a couple of publicized incidents where even RAG models have recommended wrong things , and that's why I do think , even with stuff , stuff like that , it's going to be an employee in the loop for a period of time because we want to know .
I I think most enterprises are fine with giving an employee as a rag model that's giving some answers . Will you train the employee to say hey , by the way , you're going to get this answer . You should validate it . You're going to get , as part of the rag link back to the source content .
So you have the tools you need to validate it , but you need to feel accountable for validating that , which is very , very different than just saying , oh , we're going to put a RAG-based chatbot on the website and let customers ask it questions .
So I think those are some things that we have to build up a maturity curve to be able to get to , and also maybe you need to define best practices and all these things around it as well , how to do it properly .
Absolutely . And what are the guardrails ? What are the audits that need to be put in place ? That's why we've spent a lot of time thinking about okay , if you want to do RAG at an enterprise scale , what governance do you need to wrap around it ? What auditing do you need to wrap around it ? What security protocols do you need to wrap around it ?
Because to me , that's where the challenge of making this stuff work is .
Yeah , I totally agree , and that's obviously the other part of AI in general and I think that it also plays a role in in gen . Ai is that I remember five years ago was in . I was here as well and it was all about the . It was back then .
It was a lot statistical , yeah yeah and it was all about explainability and predictability and and that was very important and and it was all the craze and it's still important , but gen AI is very hard to explain and to predict right , and I think one of the things that's really important again .
This is why it kind of is almost fingernails on the blackboard sometimes when I hear organizations talk about their AI strategy , because what you really need is an AI's strategy , because there are multiple AIs that you apply in different ways . The statistical AI that we were just talking about can be very explainable .
The other thing that's also really really important about it is it's deterministic , so in other words , it's making probabilistic models .
And if you do it three times you .
But if I feed the same data into the model , the model will give me the same answer three times . Generative AI is not deterministic , so you have to be comfortable leveraging it in use cases where the fact that it's not going to give you the same answer maybe even is a good thing .
That's why I think using generative AI , for example , to suggest workflow designs , is great , because you can ask it , throw an idea off If you don't like it , ask it again and it'll come back and give you a different variation Again . It's like a brainstorming partner that you can work with .
What I really like about that basically , actually , is that it also can help prevent bias Right , because it can give you stuff that maybe from yourself , looking at it from your perspective , you wouldn't think of .
Exactly .
No-transcript , exactly they can actually give you something that you didn't expect .
I was saying this morning that I think inclusivity and innovation go hand in hand , and the reason that is is the more voices , more perspectives you bring in , the more ideas that you might not have available in your perspective set are available to you , and GMI can open that a little bit .
But then it also muddies the water a little bit . Open that a little bit , but then it also gets . It also muddies the water a little bit , in a sense . Because what do you leave out ? Because you probably don't like .
For example , if you do the Gen AI blueprint , you ask it to come up with parts of your application that you may want to build , but then you need to decide which parts you want to keep , which parts don't you want to keep , which parts don't you want to keep . But that's still a trial and error kind of thing .
But that's still a trial and error , but that's stuff that you're going to have to do anyway if you wanted to build an enterprise application .
Yeah , that's true .
And I'd much rather you start with an idea , a starting point that feels optimal , that Gen AI and some of our best practices can help you provide , and then have that conversation about what to put in , what to leave out , rather than spending like nine weeks to get to the point that you're ready to have that conversation .
But doesn't the lack of predictability and explainability prevent Gen AI to be part of really the core processes and the core ?
of companies . Well , I think .
Because back in the day explainability was sort of a must , especially for a highly regulated kind of area .
Again , I think it really , really depends on the use case . Like we're talking about using Gen AI design time to help you design your business process . Once you run it , I actually want to use statistical AI to predict what might happen next . Where you have explainability , you have determinism .
You get back to the AIs Right exactly .
There are places like I might use generative AI during a process to draft an email that I want to send . But I want to keep a human in the loop to say , hey , we've drafted this email , you need to validate it and approve it and the process is going to audit that . You did it so that nobody can say Gen AI is responsible .
No , you got to audit it and you did it . So I think that's why it really comes down to the use case , the user and the controls you wrap around it .
So in that sense , statistical AI may have its own identity to a certain extent , but Gen AI won't , because there's always somebody who has to sign off .
I think in the near term , right Gen AI is always going to have that sort of sign-off wrapped around it All right , it's funny you mention it because we're almost out of time , so maybe on stage this week you were talking about the future of AI , which is completely unpredictable .
Completely unpredictable . I think my whole point was that nobody really knows .
You already mentioned some stuff . About 10 years from now , we're going to look back at this point and say look how far we've come . Do you see that ? The Because that's why I wanted to ask this question is based on , I think , research you commissioned ?
One of the key findings was that Gen AI also made AI so the old-fashioned AI more popular or being adopted more Exactly right . So do you see this as sort of a step up for proper AI ? Maybe that's the wrong word .
Statistical . AI or decisioning AI . So those of us who have been around AI for a while Proparator maybe that's the wrong word Statistical AI or decisioning AI those of us who have been around AI for a while will know that we've been through a lot of AI cycles . Ai was a hype in the 60s and 70s when it was expert systems . Then it went away .
Then it came back and it was predictive models . Then we entered the AI winner and it went away . And then Google . Initially AI was back again with Google , and then Salesforce , einstein , and then it disappeared . And then Google . Initially AI was back again with sort of Google and then Salesforce .
Einstein and then it disappeared again and then it came back with J8 , right ? So I think we go through these hype cycles and these waves , as we do with , frankly , every technology .
Like 3D movies . Right Exactly 3D movies .
3d movies I was thinking about looking back , right . So one of the great we talk about this AI being almost like the internet . One of the great examples of the internet sort of bubble bursting was petscom , you remember petscom . That's a symbol of the bubble bursting . I order all of my dog food on chewycom , so it's like we end up back there .
We end up at the transformational point . Eventually it just we have to make a supple . Just takes a couple of false steps , false runs to get to the end point .
So no big predictions about where we're going in terms of AI , I mean look , I think we're going to see it in the enterprise .
I think we're going to see AI injected into transformational thinking and help us kind of push transformational thinking , and I do think there's a huge opportunity as we've been talking a little bit about in this conference especially for enterprises that struggle with things like legacy modernization and legacy transformation , to plug AI in as a tool to help accelerate
that . So I wouldn't be surprised if , in a year from now , we look back and say , boy , we accelerated a lot of our legacy transformation in ways that we didn't think were possible , because you can pull in a lot of your legacy stuff . You can pull in a lot of your legacy stuff and help transform it into something different .
That's a good note to end this conversation on , I think , because it's a positive one .
I always like that .
So , thank you for for being on the on the show . Thank you very much . Pleasure to see you looking forward to next year and see what , how much , how much , how big the impact has actually been of what we've been talking about this week yeah , it'll be , it'll be exciting . I hope so .