[Music] Welcome to our broadcast brought to you by Dell expert Network I'm your host Kima Johnson from the channel company today we'll explore how the latest AI advancements are reshaping Industries and opening new opportunities for msps join me for this insightful discussion is Dr bronze Larson from Dell Technologies delighted to have you with me I really appreciate the opportunity to speak with you today as well let's get started with segment one AI 101 bronze what is AI and gen AI in simple terms there's two flavors of AI two modalities that it can be operated in and they're known as analysis and synthesis analysis is more traditional AI it's where you have a sensor system that's ingesting data you can imagine like a camera on an autonomous vehicle and it's ingesting information and then that AI system analyzes the images from that sensor to produce some sort of insight like maybe recognize a stop sign and classify it as different from a speed limit sign and it's based on whatever information has been used to train up that particular model but every model that's designed to do analysis can actually just be inverted and allow it to do synthesis and what I mean by that is imagine you had a model that if it's able to understand what a stop sign looks like to be able to classify it it has the elements to then create a stop sign or synthesize a stop sign or generate a stop sign which is known as generative AI so that's where you could actually say please generate a stop sign and it'll actually produce an instantiation and stop sign for you it can be applied to text so you can ask it to create answers which is what people think of with copilot and chat GPT where it's generating a response maybe one way to think about the these two is also something that you've been using for a long time video games if you think about the football video game Madden football there's a physics model in there for how football behaves when thrown in the air and you could use that model to analyze a football flying through the air and say based on where it is and how it's moving I know where it was thrown from the angle it was thrown the velocity it was thrown at and where it's going to land you could analyze the situation by taking images or information about the ball in flight but you can flip that same model around and input parameters this velocity this direction this asmith and it will synthesize and generate a football flying through the air according to that model and that's what goes on in video games when you play it so people have been using both Ai and generative AI probably for decades without necessarily even knowing it can you explain the different ways of AI and how it will impact Industries and msps if you think about any AI system which is really just designed to automate some process often it's something a human can do but it doesn't have to be there's kind of four elements it's really pretty Common Sense four elements that you need to make the system operate and do what humans do you have to be able to perceive learn abstract and reason and what I mean by that the first perceiving is you need some sort of sensor to take in data so a camera with the example I had the autonomous vehicle but it needs to be able to actually learn from that data to be able to make decision so learning is a key aspect of a I as well but it shouldn't just learn what it's seen before it should be able to actually operate and Abstract to things it's never been exposed to like imagine if you trained up on a base to a car an autonomous driving system to recognize a baseball a football a volleyball so that if it saw one of those crossing a road it might say whoa there might be a little kid chasing after it you know I need to stop but what if the new type of ball appears that it wasn't trained up on AI systems that they are today they tend to sort of fail and they might just think oh misclassified as a leaf and drive right through it potentially causing problem so the abstraction that third tenant is something humans can do very well and that's something AI systems need to do as well and the last is reasoning once you actually properly identify or create the intelligence about what we're looking at or what's what you're trying to to understand there has to be actions taken like suap the car or keep going if it's not a threat so that perceiving learning abstract and reasoning are the four tendons of AI and the waves of AI there's three of them have evolved to advance and add each of these different characteristics so wave one is the oldest it's been around well officially hundreds of years but you know really more in Earnest since about the 1960s and it's just based on rules it's only really got the perceiving of getting some input data and then reasoning and it's just human knowledge it takes the rules which we already know and it just encapsulates those in an algorithm you can think of like an ATM machine you know it's something a human used to do there were tellers in Bank where you go to deposit money take out money check your balance in 18 we knew the rules it was very deterministic so we encapsulated those with a series of automated you know tasks that allow you to go to an ATM machine and do something that a human used to be able to do I've already mentioned video games I mean the game of checkers we know the rules how Checkers work we know that they can only move diagonally and they can't move where there's a piece you could jump only diagonally you can't move backwards unless you're kinged so being able to automate those rules allows us to create video game AI systems that can play against you and follow those simple rules but what happens if you don't know the rules and that's where the learning concept really comes into play and if you don't know the rules you don't know the model one thing you can attempt to do is gather a bunch of data about the situation and pass it into a machine which hopefully will learn the model for you learn the rules and that's the second wave of AI and it's based on statistics it's a stochastic processing it's what we think of today with neuron Nets and deep learning chat GPT and generative AI is mostly in this domain and it's where you can actually pass in data like a bunch of pictures of a cat a bunch of pictures of a dog and it figures out the features of those and it can then be used to classify a dog versus a cat or in the case of generative AI it can generate a dog or a cat but it doesn't abstract very well so if you then only train it up on dogs and cats and you presented this system with a fish it might just fail or wrongly classify the fish as a dog or a cat it just can't think beyond what it's been trained up on and that's really where the third wave of AI comes in is this idea of abstraction the ability to actually um not have to have seen something before and still be able to operate much in the way that we do and if you think about human learning I mean I don't need to show a child a thousand images of the letter K before the child recognizes the letter K which is what current AI needs lots and lots of data I can draw one pencil drawing of a letter K and show it to a kid and that kid's probably going to know the letter k for the rest of their life whether the K is inverted whether it's red green blue cursive Times New Roman aerial font even if a cloud kind of looks like letter K and then if it spotted a new letter like a letter P it might not know exactly what it is but it would probably say hey you know what I think it's part of that alphabet system you described and it probably is assigned to another phonetic sound so humans are able to abstract without needing lots of data and so this third wve of AI actually has some attributes that it's not databased and that's pretty huge if you think about the impacts that AI has on the environment with all the training and all the data all the storage all the GPU processing and everything that goes into making a current AI system work that kind of goes away it even kind of impacts um AI ethics and the things we talk about because without data you don't have bias without data you have explainability so a lot of the issues that we talk about today privacy and storing of people's data and usage rights they kind of go away because it's not Based on data so this wave three is sort of new and examples of it are you know there's biosystems out there that can actually detect bacteria viruses and toxins even if they've never existed before and that's pretty important because viruses mutate so fast that having a constant up-to-date database that you can compare is really difficult so having a system that's capable of really saying I see a new bug let me explain the attributes of it this is its toxicity this is its viral low this is uh you know transmissibility and here's what it means in the class and the family that is part of so that humans can start to kind of say this is something we need to worry about so anyway it really wave three really has applications anywhere because almost every system that's out there today is going to suffer from the lack of being able to abstract and so you know it's something that really applies to anyi system today as that ship moves from kind of second wave to the third wave what are the key drivers behind the rapid evolution of AI and how can msps capitalize on these Trends it's kind of interesting that the technology has been around and in use for many decades but what really changed was the key driver was that it was made easy generative AI has been around for a long time I think I did my first generative AI image and video back in 1996 so it's a powerful tool that's been out there but it was never commoditized and really brought to the user in an easy way that people can use and kind of resonated with them um most of that has been in the language domain with famous open AI chat gbt or Microsoft's co-pilot instantiation but they really made it easy just like a search engine and so one thing to keep in mind as you start to use this is that you know keep in mind it really is still a stylizer you know it can create the style of conversational language but it's not necessarily a search engine so if you wanted to do creative tasks like create images for marketing campaigns um it's really great at that because it'll produce all sorts of different types of ideas for you that maybe you didn't think about and you can ask it to even be in the style of mon or the style of whatever you could ask for an essay to be written and it'll give you the format you can ask for software code to be generated and it will actually generate functions and it'll lay out the architecture for the goal you're trying to achieve it'll actually do it for you and so you know that's kind of um a pretty powerful capability now you have to be cautious that it still is based on statistics and it's creating amalgams from the internet from all the different input data that's been used for training so it's sort of a summary but it might also merge things in ways that could be a little dangerous and an example there is I've used it by accident to ask for certain mathematical theorems I'm a mathematician my training one time I was on being asking for a theorem I just needed to reference it and by accident I typed it into chat GPT I I looked away and when I came back there's the mathematical theorem I was looking for but it wasn't the mathematical theorem it was an amalgam of math theorems that looked perfect it had almost all the same words but there were some key elements like less than and greater than signs that were flipped and certain elements in it and so it wasn't correct I can't say that generative AI or co-pilot did it wrong they did what it's supposed to do which is create new stuff create Styles and wrap something in a way like an essay on a topic um and format the way that an essay should be written a softw uh Frameworks that outline and it does a great job at that but you have to be cautious that the actual nitty-gritty details within it are still going to be an amalgam of stuff from all over the world so you know you might ask it about heart health and it pulls from the heartest Health letter and the Berkeley Wellness letter and the Mayo Clinic but it might also pull from a uh popular vlogger in the Caribbean who loves salt and recommends that you eat lots of salt for your heart and maybe that's not the best so you have to be a little cautious about the details but it's a fantastic tool um for a lot of other creative areas and I think that fun and creative part of it is one of the reasons why it exploded onto the scene because people were just so amazed I mean even musical groups like Duran Duran they created an entire video uh that was generative AI it was a a s called invisible you can go look it up and that was pretty neat it kind of shocks people that you know this system can actually make videos now granted that was creative so there was kind of no wrong answer uh matter of fact the band said that they liked it when it sort of came up with wacky stuff because that was more artistic and creative but that's really the reason why right now it became so prevalent is everyone got their hands on it and they could kind of see what AI could do for the first time how is Dell leveraging gen AI within its product offerings and solutions for msps we are also on a learning path as well I mean this technology is not completely understood but we've really dove in head first and we've created an office dedicated to uh learning what AI can do for us we also have an office that is dedicated to what AI can do for its customer customers and we've been on this journey as a company for probably about a year now and we are still learning what you can and can't do with it we're also have a committee dedicated to governance because governance is changing really quickly and that's something that's scaring a lot of the customers partners msps that how do we stay on top of it it's changing so fast so Dell has actually got a very sophisticated process from monitoring that understanding it basically distilling information from these rules that are being put out there by region down so that we can actually say here's what you can do here's what you can't do so I think that's a powerful initial tool that we'll be able to offer to msps to make sure they don't misstep to make sure that they don't deploy technology and violate a rule because the penalties are actually quite stiff um you know the new uh EU AI act comes up with a you know per instantiation penalty which is pretty Monumental so being able to navigate that is something that we need to understand for ourselves and that information I think would flow down immediately to a lot of our um you know MSP and our partners and our customers to be able to help them navigate that and we have misstepped and we've learned along the way as well we at at times deployed some software throughout the company and found that even though we had done our a good due diligence the rules had changed from the time that we first evaluated it and we ended up finding that there was a violation which we had to go in and fix and software before we could deploy it again so that is something that is you know very costly to companies and they need to get it right the first time and monitor it um as well that brings us to segment to harnessing AI for transformation can you give some examples of AI applications that are driving significant value across Industries in the generative AI space or even traditional AI it's really on internal business optimizations it's these helper tools and one of the reasons why that is the case is because if you're doing traditional AI where you're the one inputting the data you're the one who's asking the question trying to develop the model you know companies themselves internally own that data they don't have to go out and acquire and so it allows them to actually build the models and they also have the insight and the technical expertise and the business expertise to know when the model is good enough so often the internal business optimizations are where things um happen the most I mean here at Dell we know we put it into our products so we monitor our products and have ai running to for instance heat management to make sure that these systems are running cool that they're running quiet and that that means they're less likely to break or have problems we use it for parts failure prediction so that we can actually estimate that our products which are out in the wild based on AI we can say there's a likelihood that there's going to be a problem maybe with this particular product or in this particular region and it helps us stock our repair rooms accordingly and that helps minimize downtime for our customers when they have a problem and helps us actually minimize cost which we can then pass on to the consumer as well in general generative AI has been very helpful with repetitive tasks by freeing up their time I use the software development by being able to create a software framework rapidly and being able to then allow the developers to kind of Dive In fine-tune it means they have more time for doing Downstream debugging product enhancements we've heard a lot about people having using gener AI to generate new business leads um and something that's much faster something that would take a day now they can do it five minutes because it sort of Aggregates information about a client summarizes it and allows even real-time conversations with customers to sort of uh you know be more in tune with what they're really looking for any type of risk assessment uh is again something that is very powerful to be able to um analyze when there might be a investment risk and identifying early marketing Trends and marketing recommendations for targeted marketing is something that is very very powerful for but in general uh the chat GPT is still mostly language based there there are some organizations that are using the image creation for creating marketing material realtime email respon responses as well as something that uh that people are seeing significant help with but it usually is kind of right now the manual tasks that are taking up a large portion of people's time and it's doing those allowing these people to be more productive and and expand what they could normally accomplish in a day the thing that is interesting though is the applications also have to be somewhat low risk because there there are error rates and because of those errors because of those hallucinations and the problem with hallucinations is they sure do sound convincing companies are finding that they have to maintain they're not reducing headcount they have to keep a human in the loop to basically uh review everything that these co-pilots are creating so it's Expediting the creation of content with humans still in a loop it's not a replacement for humans how can msps integrate these Solutions into their offerings you know it's still very much in the exploratory phase and it's going to really come down to each business with their area of expertise analyzing these tools to find if and where it helps my advice is to at this stage play with it play with it and find out how it works and how you may be able to utilize it talk to people in your industry that might be using it to find out how they're using it and it's going to take each organization and Company's area of expertise to assess it no one outside can do that and I used the example of software development a year ago everything on the internet was saying it's going to gender AI is replace software developers in these sort of parlor tricks online we're showing video games being made in 10 minutes just by describing the features the software developers dove in and use this in earn what they've sort of found out is it does a good job at kind of gross organization you know macro scale software Frameworks it does a fantastic job actually at that and sort of outlining how functions should operate but the fine details are just wrong and it's actually sometimes harder to go in and then fix someone else's code so what they've been able to understand by playing with it and by utilizing it is they found the functional boundaries for their area of expertise to say here's where I can use it and it's a helper tool here's where I need to dive in and have my human expertise take over so I really really encourage you and every organization to just start playing just use Microsoft co-pilot look up YouTube examples of how you could use it talk to people in your industry and start down that Journey that path the other thing I will say is start with a problem to solve don't assume you need AI um what you actually find that you know there's a kind of a danger when you have a solution looking for a problem to solve but unfortunately gen is sort of in that space we kind of say hey here it is world now the world's running around going well like I guess I need it I guess I need it don't fall into that trap start with the problems identify the list of things that your organization needs to solve that are bottlenecks things you're struggling with and then just try to solve those problems if it leads you to AI great if it leads you to gen great that's something should come in response to a specific need and one thing we're seeing out there when we talk to a lot of folks is people are saying well I've got gen and I'm going to go try it over here will that fail I try it over here will that fail over here kind of work well they're they're just sort of flailing around and it's just you know they're going to try to almost fit a round peg into a square hole so if I were to advise people right now maybe play with the tools that are currently out there understand the limitations maybe hold off a little bit on diving in too deep until you really understand the problems you need to solve until maybe there's a little more solidification and understanding of the false alarm rates how to evaluate them and kind of who are the winners in this race and at that point if you're developing and playing on the side and you understand where it may or may not apply at that point you'll be able to identify the proper solution maybe it's an on-prem solution because of privacy issues that are important to you security issues which are important to you more fine-tuned accurate results which you can do by developing things yourself versus using um just the tools that are out there or maybe what you have out there already is good enough and so that's where I think it's changing so fast I would really dive in to understand it and but maybe be a little cautious and make sure you're approaching it from a problem maybe a solution as well and then governance is changing so fast as well and these ethics and Regulatory issues that you know you also want to sort of be cautious about diving in too strongly in one area that all of a sudden you find who you kind of decided to work with you know is facing some pretty serious problems and by the way just know you are not behind the curve it is the wild west out there right now it costs a loot right now starting now to explore is not you're not behind the curve you are you're right where most people are these days what industries stand to benefit the most from AI adoption and what specific challenges do they face in implementing Aid driven solution since gend AI is really just a stylizer I kind of keep coming back to that when it's applied to language it tends to be very articulate because it's stylized from all the conversational well pretty much English it's out there it's actually an issue that English is the dominant language on the internet and so it does a better job at a conversation English but since its output is based on statistically popular language it's not necessarily focused on producing the most accurate results and so what happens is that well spoken characteristic can often fool people into thinking that's more intelligent than it is and that can be dangerous and due to the fact that these neural Nets are black boxes based on some stochastic approaches you never really know when or where the errors will manifest themselves until the question is asked there's kind of no generative AI model resume out there that says if you use this for this application space it's going to be 90% correct 100% correct um and so that's a problem facing um the adoption right now also um there are a lot of AI scientists that sort of worry that people don't understand really how long it may take and how much it may cost for an application to become technologically and financially mature a lot of times people want quick Roi they maybe say give me you know you got six months to year to you know implement this when in reality there's a lot of applications maybe 10 years or more and that can cause some adoption to sort of fall off that's why most people right now are sort of in that exploratory phase um and they're sort of trying to figure out what is the ROI and that's hard to quantify there's also um bottlenecks Downstream there's some real applications as I've already mentioned of people uh having improvements in terms of maybe surfacing much more business leads than they were able to before faster than they were able to before but all of a sudden there's been Downstream bottlenecks there's nobody to process those new leads at the speed that they need to actually turn into a reality so there's sort of the entire business stack is kind of needs to be updated in order to really take advantage of a generative Ai and the power that it is starting to show promise for um and and you see that actually in some applications where uh you know there are some actual accuracy issues like General Motors and apple both got out of the Tous driving business because some of the error rates were not they were not able to get them down to a level that meets uh kind of governance requirements and performance requirements um and even cost issues can come come into play CVS and Walmart they're getting rid of their self checkouts because they found that the AI systems and the maintenance of um was actually more costly than just having human Checkers so there's a balance and there's a lot of learning going on right now and that's why I kind of come back to um play with it to find the areas where it works for you and then think about it from the entire business stack to say if I were to improve this a segment do I have the ability to actually turn that into uh overall P business let's move to segment three building strategic Partnerships in the AI era what resources and support does Dell offer to help MS PS develop AI competency and expertise my advice to people is reach out to your Dell resources you've explained gen Ai and AI so clearly where can we learn more I would just start with the links that are provided there and they'll be able to connect you with the right resource to help you on your journey Ron thank you so much for your time today oh thank you I really appreciate it and thank you at home for watching for the channel company I'm Kenna Johnson
2024-08-20