Latest / Technology Now / How do we know if our AI… is really AI?

Transcript

How do we know if our AI… is really AI? · Technology Now

• BARADJI DIALLO I'm originally from Mali in West Africa. I never grew up imagining myself ever being in stage like this , so it's a true privilege to be where I am. So let alone being in tech and AI and all that stuff we're talking about today. So I really feel blessed being where I am. MICHAEL BIRD That was Baradji Diallo guest for this week. I’ve got to be honest with you Aubrey, he is a pretty cool dude. AUBREY LOVELL He seems really cool. I mean he probably is cooler than us, right? MICHAEL BIRD I mean, are we now elder millennials? Is that what you referred to us on a previous podcast? Two elder millennials. I, I think he's, he's significantly cooler than us. But his job is actually really quite amazing, too. And that is what we're to be focusing on this week. AUBREY LOVELL Alright… go on then, what does he do? MICHAEL BIRD He’s almost like an AI forensic scientist. He gets to look under the hood at products which claim to be using AI and find out if they are the real deal. AUBREY LOVELL Ooh that’s interesting, and pretty cool. I can’t wait to find out more! I'm Aubrey Lovell MICHAEL BIRD I’m Michael Bird And welcome to Technology Now from HPE. MICHAEL BIRD Since the AI revolution, we see it everywhere right? All sorts of products and solutions claim to be based around AI which shouldn’t come as a surprise when market data company Statista projected in march 2025 that the AI market is projected to reach well over 200 billion dollars this year and expected to increase to a market volume of over a trillion dollars by 2031. We have of course, linked to this in the shownotes.i AUBREY LOVELL That’s a mind blowing figure but as you said, it’s not a huge surprise when we see everyone claiming to use AI in what they’re making. I know that AI is pretty ubiquitous but surely not every product claiming to use it can be correct every time? MICHAEL BIRD Yeah exactly. And these mistakes aren’t necessarily malicious but they still need to be corrected, right? AUBREY LOVELL Oh for sure. It’s not good for anyone to have people and products incorrectly claiming to be AI based. MICHAEL BIRD Well as I said before, this week we are joined by Baradji Diallo who works in the office of the CTO as an AI Innovation Architect in Technology Strategy and Evaluation AUBREY LOVELL But before we get to Baradji, I want to take us back and look at how the world of AI almost never came to be which means it is, of course, time for… ... Technology Then. AUBREY LOVELL Michael, I have a question: have you heard of the AI Winter? MICHAEL BIRD Ooh. I haven’t, I can guess. I'm guessing it's maybe from the mid 70s to maybe the late noughties, early teens, something like that. AUBREY LOVELL Okay, getting warmer So as you've noticed, right, we're currently in an AI boom. This was, however, not always the case which led to the coining of the term AI Winter in 1984 during the annual meeting of the American Association of Artificial Intelligenceii. Now these winters are periods of low funding, low investment and importantly, low interest and these winters are, of course, part of the natural cycle of research and development but they can be devastating for companies which have invested in AI. So, after overly ambitious promises were broken, and fundamental limitations in the technology were identified, US government funding which had fed the AI industry during the ‘60s was pulled. And a 1973 report from mathematician James Lighthill criticising the field of research led to the UK to cease funding for AIiii in the early seventies. Now, by 1974, the first AI winter had begun and funding was almost impossible to obtain for more AI research.iv This first AI winter began to defrost with the advent of expert systems which began gaining traction in the early eightiesv however the boom was once again short lived as these systems ran into a serious problem: many couldn’t store the quantity of data needed for their models to work properly and they could completely break down when presented with unexpected information. Yikes.r By the late ‘80s, regular computers had become more powerful than specialised AI machines, and in 1987, the market collapsed.vi Some time between the mid-90s and the early 2000s, the second winter thawed as new forms of machine learning caught on and the physical hardware to run AI systems improved to where it needed to be. These new types of AI software needed specialist hardware to work and until this point, the hardware simply hadn’t existed. vii MICHAEL BIRD But now it does? AUBREY LOVELL Now it does, indeed. And I get the feeling that this sort of thing is what our guest this week spends his time looking into? MICHAEL BIRD Yes that’s absolutely right. Baradji Diallo who works in the office of the CTO as an AI Innovation Architect in Technology Strategy and Evaluation and with a title like that, I had to ask him: What does an AI Innovation Architect do? BARADJI DIALLO I like to think about my role as a bridge between imagination and execution. That's how I see my role as. With that being said, one of the biggest things that I'm hopeful is I'm here today helping HPE shape of trying to figure out what's real, what's useful, and what is worth turning into something tangible. So when we talk about by AI innovation architects tend to wear a dual hat. The first part is evaluating emerging technologies. So AI, of course, is one of the obvious one in that space. And the second one is slightly different but complementary, where I take raw innovation and try to turn it into something practical. MICHAEL BIRD your job is to look at products or ideas that come to you and say yeah yeah that works or maybe not BARADJI DIALLO to put it simply, my job is to help separate noise from value. There is a lot of noise in here and making a decision that can benefit our future at HPE here. Because a lot of this stuff have a lot of noise. question is, how do you separate all the noise for true value for something that's actually meaningful and impactful to all of us? MICHAEL BIRD So how do you actually do this? Talk me through the process. BARADJI DIALLO So yeah, part of the job is to test out if it's real or just hot air. MICHAEL BIRD Somebody comes to you with a product. I'm guessing we're talking here software. Somebody comes to you with bit of software And what do you do? Do you just sort of play around with it to begin with? Just sort of see how it works? BARADJI DIALLO Exactly. In a way, I like to think about it more like a treasure hunt. So a lot of time or come out with these great or bold ideas or make this claim. goal is to really find out if X really marked a spot or if it's just a bunch of empty promises. So going through treasure hunt, there's quite a bit of different things. There's not always the same path each time. There's different things I have to rely on meaning I can't fully rely just on my own expertise. This innovation space is so vast, so many companies coming, and so many different things. We need to play a role to stay ahead of the market. So sometimes I have to lead on different experts from the company, because that's what makes us great. Knowing I can't be the master of all, but I know those that might know this area, so leaning on them at time to truly see, because we talk about AI first. For example, to me, AI is not really about, okay, does it work? Because we passed that point now. You want to use AI. The question to me is... Ultimately, can you use it for the right task, for the right purpose, and in the right condition, versus just using it? And part of that is trying to cut through that noise and trying to see, are they just putting a wrapper around some kind of software? Yeah, that might be great, but the reality is, is that true innovation? There's nothing wrong with it, but is that where HPE is going? Not always, or is there something truly embedding the product where AI is ultimately making a decision? versus just putting a power by AI, but I'm like, you can put that on anything. But what is truly underneath, that's making that decision. And that's what I'm looking for ultimately. MICHAEL BIRD It sort of feels like the way AI sometimes goes, it sort of feels like back in the day when there was this search of like smart home products and it was like this microwave can connect to the internet and it's like okay, why? Why would you want to do that? BARADJI DIALLO That's the thing, right? because sometimes I think we have reached a time where everyone want to put AI on everything, which I get it. But some of them tend to be more of a marketing. And sometimes there's no need to put AI. Like you can have a real software. Because those are some of the stuff I look at too. Are you just putting AI for the sake of putting it? What is the true value of AI in here? MICHAEL BIRD Okay so you have this product, you've done a bit of an evaluation, are you then digging into the code? BARADJI DIALLO Yeah, that's the interesting things. as an innovation architect, one of the biggest thing is I love getting my hands on technology on the codes. Because sometimes it gave you a snippet of how the structure and things and how mature this company might be as well. But of course, not everybody wants to get your hands on their code. I have to be mindful of that. Sometimes in some of the scenario, I demo my work. But at the core of it, all I want to do, I want to work with them because to me, I like transparency.? What are you claiming to solve? And knowing those key things, now I can work with you to look at that code and say, okay, there's something reasonable here or if it's just a demo, looking at that demo to say, okay, is this truly solving what I need to solve and bringing in others as needed to truly evaluate the solution. MICHAEL BIRD I imagine with sort of AI products, that must be quite challenging because from what I understand, AI can be incredibly opaque to actually understand what on is going on. BARADJI DIALLO Yeah, yeah. People just want to put it everywhere. So definitely that can create some of those challenges. But the reality is, even if you have access to code, because sometimes people wonder, can I just look at that code and figure out if this is real AI or something like that? Not necessarily. think if more than anything, what you're looking for is figuring out the claim, even if it's powered by AI or embedded truly in their solution. The thing is, whenever AI is brought in, or you have a model bring into this area, it ultimately add up different challenges like hallucination and how does this meet your constraint. So you have to look at all those constraints like accuracy, explainability, fairness, ethic. So many things that goes into it. Those are some of the elements I'm looking for to see does the model behave accordingly versus okay, powered by AI. That doesn't really excite me no more. We have passed that time already. MICHAEL BIRD Yeah, AI can be just a bit of a buzzword that means nothing. Unless you can actually prove it's doing something useful. I mean, how can you tell if something really is, for example, like generative AI? Does the code have a certain look? Are there signs? Are there little clues that you're looking for? BARADJI DIALLO looking at the code really, I wouldn't say there's certain look when it comes to AI, because when you think about AI, there's different level to this space. I like to think about it as a three layer, the AI stack. So when you think about the AI stack at the very top of the layers, you typically have a few organization that can truly train and build these models and these spaces require tons and tons of compute and serious amount of capital. So there's not many people that can play there. So if somebody's telling me they're playing there, I'm like, wait a minute, what company are you? So that one raised flag. And in the bottom, typically that's where a lot of time we have a lot of teams just use models out of the box and then via APIs or via a web tool, plug in, play easy to adapt but less flexible to control. of course, if you're playing in that space, nothing wrong with that. I have to be mindful of there to where is a true innovation because you're just using something. But I'm looking at the middle layer typically. That's where the real complexity and that's where the real opportunity lies because that's where Teams takes in these models, embed them into their use case, fine-tune them and embed them into to real workflow while navigating shifting business constraint requirement. And that's the area I'm looking for more than the code. What are you actually doing? Because you can add all kind of fancy code there, but if it's not solving a real problem, I'm like, what are you truly doing? And that's the question I want to answer. And how is it serving the people we want to serve, which is ultimately our customers. MICHAEL BIRD Presumably if they're using an off-the-shelf model, there's data challenges there? BARADJI DIALLO Exactly, because that's part of the problem, right? If you're using some of this off-the-shelf model, because that's the one thing I tell people. You can't just take a model, you can, and put a wrapper on it, but the reality is, I don't... personally consider that the true innovation is serve your purpose. That's good. But a lot of time, that real opportunity and innovation comes when you can take these models, fine tune them, and then chain them together and using the right model for the right task. Because you have hundreds of thousands of models available now, both open source and closed source models. So it's very challenging deciding which models to choose for the right task so that’s those key things MICHAEL BIRD And generally speaking, how long does this process take you to review and get to a stage where say, yep, this is good innovation or... BARADJI DIALLO No, that's a great question. Sometimes this can take up to a few days or sometimes it can lead up to weeks or even months because my goal is to give everyone a first shot. Especially when it comes to evaluating. This is a complex space and when it comes to innovation, you can't just brush it off because I don't understand because sometimes they come to me, I truly get what they're saying and sometimes I have to know, I'm not an expert at everything. How can I connect with other experts at HPE because you have a massive organization with a lot of brain mind in there. So being able to collaborate and coordinate with them. Sometimes it might take time, but I like to be transparent with anyone I work with. Each step of that way and get them involved wherever it makes sense so they know why am I making some of that claim and why I don't like certain things. That transparency is key. And sometimes they come back and say, I don't agree with you, Baraji. There's nothing wrong with that, to agree, but that was my experience and I'm sharing with them and some are grateful for it and regardless what it is for me, I just want to stay truthful to myself and be as transparent as possible. MICHAEL BIRD So our listeners are probably going through a similar process with reviewing products, particularly reviewing AI related products. what sort of tips could you give our listeners the best way to evaluate if the product that is being put in front of them is what it says it is? What are those maybe top five things that you do? BARADJI DIALLO you need to start thinking about when it comes to AI as a multi-stage process, when it comes to choosing the right model for the right task and the right constraint, you have to approach it like a way that mirror how we approach hire people for critical role. Like the first step, for example. That step you have to think about what are you truly solving for? And then the second tips I will give them to benchmark the basics. So this is where teams usually tend to start, where we try to evaluate some kind of models against public metrics and benchmark available but this is still limited. It gave you part of the picture. Kind of like if I have to utilize my HR or hiring process. So this is like when you're reviewing a resume. It gave you a signal but never the full picture. So you get an inside of it. And the third tips in that aspect will be you need to simulate your world. Because that's what I do when they're coming to me too. How does this model actually behave on your environment? Your data and on your constraint. Because you might behave great in finance domains, but I mean the healthcare and IT, it might not serve me well. So this is the last stage where you bring somebody in for interview. So they're not just solving a task, now they're solving your task. So that one is about simulating your world. And the last tips I would love to give in this area is you have to evaluate for trust reaching an area of cannot be something you just put it to last in because the reality is, accuracy does not equals trust. Because sometimes that's the conversation I get with. You can't tell me the model is 100 % accurate so that doesn't give me the confidence to trust this model, because that's the space we're in right now. So we need to stress test for robustness and make sure it's safe to use these models. Because this is when you hire the employee can they do that job reliably and with confidence and that's the key there because each of these stage kind of give you a sharper lens but together we kind of give you confidence that not only that the model works but that it works for you and for your specific use case. So those are the tips I'll give to people looking at the space take a look at hopefully it will serve them well. AUBREY LOVELL What an interesting guy. MICHAEL BIRD I think he's got such a fascinating job. AUBREY LOVELL Yeah, what a fun job. I love what he said in the beginning about when you are doing an analysis and kind of figuring out emerging technologies with AI, separating the noise from the value and kind of how it is a treasure hunt. And it really is because it's just like this explosion of options. And yeah, you get the marketing aspect of that. But what really actually helps you? Where's the value for your business? MICHAEL BIRD yeah and he said, he said something in the lines of powered by AI isn't really a big selling point anymore. A bit like, I think I gave the analogy of, back in the day it was like connected to the internet and you know, just because, just because you can, be powered by AI doesn't mean that that provides anything useful. Like, you know, the same analogy of the microwave. If my microwave was connected to an AI model, I'm not sure what additional value it would give. I sort of feels like we're in that phase of yeah there's potentially quite a lot of stuff out there but actually trying to figure out what is the thing that's of most value. AUBREY LOVELL Yeah, for sure. It's like a Renaissance period, if you will. That's kind of the way I see it. It's like, right now we're figuring out what works, what doesn't work, know, throwing it all out there and then seeing what catches. MICHAEL BIRD Yeah, I think he also said one of things he looks at is, is it solving what needs to be solved? Again, going to that point of are you just providing a solution without a problem? AUBREY LOVELL Right, exactly. MICHAEL BIRD Now, Baradji is clearly someone who enjoys what he does immensely so I thought it would be nice to end with one of his favourite things about his role. BARADJI DIALLO for me, a lot of stuff I do and a lot of the innovations all come down to making innovation reachable and making it feel real. Because as I mentioned, that's how I began my introduction. I came from a place and a background that this opportunity didn't exist. So I've been put in a space right now where I feel like it's my duty, no matter what I do, on the project I'm working on. How do you make it feel real and reachable and part of the project that I'm leading and most of them are in those spaces to help our customers to help our team members actually see themselves in places they didn't see themselves before and actually feel like they can reach this innovation and that's a core thing that I enjoy doing and very grateful to be where I am today. MUSIC STING AUBREY LOVELL Okay that brings us to the end of Technology Now for this week. Thank you to our guest, Baradji Diallo, And of course, to our listeners. Thank you so much for joining us. MICHAEL BIRD If you’ve enjoyed this episode, please do let us know – rate and review us wherever you listen to episodes and if you want to get in contact with us, send us an email to technology now AT hpe.com and don’t forget to subscribe so you can listen first every week. Technology Now is hosted by Aubrey Lovell and myself, Michael Bird This episode was produced by Harry Lampert and Izzie Clarke with production support from Alysha Kempson-Taylor, Beckie Bird, Alison Gaito, Alissa Mitry and Renee Edwards. AUBREY LOVELL Our social editorial team is Rebecca Wissinger, Judy-Anne Goldman and Jacqueline Green and our social media designers are Alejandra Garcia, and Ambar Maldonado. MICHAEL BIRD Technology Now is a Fresh Air Production for Hewlett Packard Enterprise. (and) we’ll see you next week. Cheers!