Amanda Razani: Hello. I’m Amanda Razani, and I’m with Techstrong.ai. Excited to be here today with Harry Powell. He is the head of industry solutions at TigerGraph. How are you doing?
Harry Powell: I’m all right, thank you. How are you?
Amanda Razani: Doing well. Can you talk about TigerGraph, and what services do you provide?
Harry Powell: Okay. So TigerGraph is a platform for analytics and machine learning, but what makes it unique but unusual, is it’s on connected data. So, these are networks of information. It structures data in terms of the relationships, which is quite different from the ground up than a normal relational database that people might use, and that allows us to be very powerful when thinking about data, where relationships are important, for example, in fraud, where criminals might be connected, or in supply chains, or in money laundering or in any sort of application where understanding how things relate, one to another, is a really important thing.
Amanda Razani: Okay. Well, that’s a great lead into our topic today, which is data analytics and where generative AI fits into that. So, let’s go straight into our questions. How does data management need to evolve to meet the ever expanding capabilities of AI for businesses?
Harry Powell: Okay. So, in some sense, the whole promise of generative AI, which I guess where we’re talking to, particularly, at the moment, which is this ChatGPT and large language models and all that is… The promise of it is that data management per se could become irrelevant, right? Because you just need to ask any question, and somehow it’ll jumble through all your data and find an answer and return it in any language you choose, whatever, in Spanish or something. And in some sense then, who cares about data management? Because the AI will sort it out for you. Somehow the AI will – imagine that instead of an AI, you just had a data department of half a million people all just shuffling bits of information, bits of paper around, well of course you can’t do that, but the AI can, and that’s the promise of AI.
Whether of course it can deliver on that, I guess is up for debate and not just in terms of practically, does it have the power to do this thing where you just take a bunch of information and data management decides, unimportant, but there are all sorts of questions around it, like how do you know that it’s understood your question in the way that you intended it to be understood? And when it returns an answer to you, how do you know that it’s done what you expected it to, and that you’re then going to interpret it in the right kind of way? Because of course, if AI replaces data management, then some kind of AI is going to replace data managers. So what’s the point of doing that? It would only be in order to put any old person in charge of data management, just democratize it, get anyone to be a data manager, of course, then how can you be sure that they will understand what the AI is trying to tell them because they won’t have that kind of training or that kind of experience.
So I think you talked about how is AI going to affect data management. At the promise level, it says data management becomes irrelevant because somewhere in your data is a pattern. It will find that pattern, it will understand what you’ve asked it, and it’ll give you the answer back as if it were half a million clever people and a huge load of piles of paper in a big room. But the truth is that we actually need to have a degree of experience in understanding how to unambiguously ask a question and how in some sophisticated and subtle sense to respond to an answer.
Amanda Razani: Absolutely.
Harry Powell: That’s my thought anyway. I mean, who knows whether I’m right or not, but everyone thinks that it’s going to… the promise is that it’s going to really change the world. But I think there’s perhaps a little more subtlety to the question than that.
Amanda Razani: Yes. And as you said, we’re not there yet with this technology, and how do we know if the data is accurate, if the data is clean? And if we leave it all to AI, then there may be a lot of mistakes slipping through. So what is your suggestion for companies when they’re looking at AI and they’re looking at how to implement and harness AI? What is your suggestion to these companies? Where do they start?
Harry Powell: So as I say, when we’re talking about AI, I think we’re talking about this generative AI thing, right? To me, it feels like the place it could be best is in a task that you would otherwise automate. So where people have used kind of automation software in the past, then that’s obviously something where AI can perhaps do something pretty effective. It’s good at doing what relatively unskilled humans can do. So if you have a process that is currently run by people who are clicking boxes and moving information from one place to another, then AI can perhaps do that more effectively than, or at least in a slicker way than, perhaps the current crop of automation software might do.
And I think also it is useful where you have a corpus of information that you really understand. So chatbots might be pretty good, or if you have a knowledge graph where you have put a real reasonable amount of work into encoding what your organization is able to do, maybe with a huge amount of documents. Maybe you’re a big manufacturing company and you have a reasonable amount of structure already, and there are a reasonable number of questions you could ask. Maybe there are questions about the… What’s happened… how has the design of one car changed compared to the next car, right? Well, there’s a reasonable amount of structure around that. You’ve got a whole load of documentations. AI would be brilliant at in some sense, synthesizing an answer for you on that.
Perhaps where it’s going to be less useful is where someone says, how should I understand a change in customer behavior that can be observed in some sense through maybe a banking use case? How do I understand how my customers have changed the way that they interact with the bank? Bit too… quite an open question. The data is quite abstract, right? It’s not kind of words and stuff like that. And the answers need to… probably where answers are required to be quite precise, then maybe the current kind of crop of large language models aren’t really the right thing to do. If it really matters whether your interest rate… Perhaps you’re calculating an interest rate, it kind of really matters if it’s 1% or 5%, right? If are paying your mortgage, your property loan rate, someone said you’ve got 1% payment or 5% payment, you care about that. But a large language model, well, it’s kind of pretty close, it’ll be all right. You get the pricing wrong though, and suddenly your bank either sells no loans or it’s completely overwhelmed with people applying for a loan.
Amanda Razani: Yeah, absolutely. One small inaccuracy can become a big problem pretty quickly. So it’s still very important this human element. And so I do have to ask, how do business leaders approach training when it comes to implementing this technology? And also when they’re undergoing this change management to implement generative AI or any kind of technology, there’s sometimes employees that are going to buck against this. They don’t want to use this technology. So how do business leaders handle this?
Harry Powell: Okay, so training, I mean, there’s going to be a degree of experimentation, I think actually, because it’s such fundamental change. I mean, I really think it is. If you think about the original kind of industrial revolution, I know we’re going back a little way there, but when production was moved away from cottage industries into factories and then the electrical revolution came in and you could start powering your factory in different kinds of ways or organizing it, and then you bring it… all these big changes… I think this AI thing may be doing the same for knowledge work as those other innovations were for production of things. And that probably means that the question of how we use AI is in some sense perhaps secondary to what we’re actually doing in businesses as a whole, right?
So big question. Imagine AI, and we know it can, is brilliant at pulling together slide decks, right? It’s brilliant at pulling together slide decks. It just routes through all the information in the world and comes out with a slide deck that tells you how you can brilliantly do change management in your business. I’m going to play it back to you. Let’s do a ChatGPT. Pull me together a slide deck on how to manage AI change in businesses, and it produces you this deck. And that’s amazing. And maybe something you might think then is, okay, that’s great. The first thing we should do is get ChatGPT to do slide decks. That’ll save a whole load of work. But in fact, maybe the real answer should be, if ChatGPT can produce these slide decks, should we be producing slide decks, right? Suddenly.
Because in some sense, when you look at a whole load of work that’s done in offices, we’re replicating with a word processor, what actually someone with a typing pool and a whole load of carbon paper did 50 years ago, we’re still passing these information documents around. And in some sense, the large language models are sort of in our face going, look, if a computer can do that, should we really be doing that? Is that really adding any value? So in terms of change, what I’d say is yes, they’re amazing these things. And yes, your earlier point about accuracy, be careful of these things because they’re not necessarily thinking linearly and in structured ways in the way that human beings do. They could produce all sorts of funny stuff.
But even so, if you imagine that they actually did work in a way that human beings could understand, perhaps this is the prompt to actually change the way we work. Because if I can on demand, pull up all the information I need to know in order to do a change program, why do we need to be producing it in a slide deck at all, right? All of the stuff that’s produced probably not important. But I guess my slightly odd point there is this may be a prompt to actually start thinking rather differently about how we use computers and how we use information, because most of the reason that people have a computer on their desk in an organization can probably be replaced by ChatGPT. At which point do we really want to replace them with ChatGPT? Or do we just want to move on from that style of work, overall?
Amanda Razani: Yes. So kind of thinking what do the humans bring that’s unique compared to this technology, and harnessing it as a tool in the areas where it will be helpful, but again, we still need that unique quality that humans have.
Harry Powell: Or it’s a bit frightening, because I think a lot of people are sitting around going, what is that quality? Is that really a quality? Are we deluding ourselves that it’s a quality? Or will we all be replaced with ChatGPT? The experience of the human race has been that whenever we’ve come up with an innovation, we’ve ended up with more work to do than before the innovation. I mean, there are many more jobs available now than there were back in the Victorian days, and chiropodists and barristers, the coffee people, and all of that, right, they never existed 150… maybe they did 150 years ago. They certainly didn’t exist on every street corner. We’ve invented a whole load of new things to do, and that’s probably what we’ll end up doing.
Amanda Razani: So the data analysts probably don’t need to be afraid that AI is going to take their job anytime soon. And I mean, certain jobs probably might eventually be taken over by AI, but that’s just going to create new jobs – a load of new jobs is what you’re saying.
Harry Powell: Yeah. I think actually data analysts… Look, I’m a data scientist, so maybe I would say this, right? But I think data analysts, because of the importance of clarity and precision in their work, are going to be… And clarity and precision in an environment of uncertainty, that may not be the kind of thing that ChatGPT is very good at. And I know that lots of big companies have put a lot of effort into this idea of, I’m going to have a front end on my analytical platform that says, “Just enter something in plain language and it’ll come up with the answer.” But I think I would be loath to make big business decisions based on that, because there’s so many fundamental things that a well-trained analyst does, which is think deeply about a problem, right? ChatGPT ain’t thinking deeply about a problem, right? It’s just getting what it thinks the information is.
Dealing with the subtleties of say, experimental design. Am I exactly asking this or am I asking that? And how is my answer going to be impacted by relatively subtle changes in experimental design and those kinds of things? How I then interpret that. When I get my answer back, what do I then do with it? All those things are actually much more important than the relatively mechanical thing of pulling out information and running a statistical procedure on it.
Amanda Razani: So assuming that as fast as this technology is advancing though, assuming that it’s going to continue advancing at this rate, where do you see the future of the enterprise two years from now as far as AI, generative AI? Where do you see us going?
Harry Powell: So I think we’ll all be exhausted if AI keeps moving at the rate that we’re talking about. I mean, the rate has just been, it’s hard to keep up with that. I certainly see AI, as I said earlier, really taking on a lot of automation processes because it’s so much easier to set up. And then I think we’ll rethink how we organize businesses and enterprises around the kind of earthly old-fashioned behavior we have. And what will that mean? I think the people will become more precise and technical, because those are the kind of questions we are going to need to ask, that we can’t do otherwise. All that kind of clerical middle probably is under threat because we won’t need huge amounts of people producing documents, PowerPoint presentations, reports and all that stuff.
Amanda Razani: It will be interesting to see what happens. I appreciate you coming on the show today and sharing your insights with us.
Harry Powell: Well it’s very kind of you to ask me. Thank you.