Media Week NYC theCUBE + NYSE Wired | Scott Hebner, theCUBE Research

Clips
News
More from Media Week NYC theCUBE + NYSE Wired

Scott Hebner

Principal Analyst, AI

theCUBE Research

Causal AI: The next evolution in machine learning

As the initial hype for artificial intelligence dies down, the question of which trends are going to stick around is on everyone’s minds.Causal AI and small language models are the answer, according to Scott Hebner (pictured), principal analyst of AI at theCUBE Research. Ultimately, he believes these trends will coalesce into a cohesive machine learning architecture.“We’re very much in the infancy of AI,” he said. “There’s some trends that are starting to develop. Now they are going to converge over time into more of an architectural approach, and I think that is the notion of causal AI, the actual ability to reason, problem-solve, understand cause and effect

play_circle_outline Introduction to coverage at NYSE with NYSE Wired community.

play_circle_outline Market growth and key players in causal AI industry.

play_circle_outline Call to engage with experts and companies in causal AI field.

play_circle_outline Empowering Companies Through Collaboration: Urgent Engagement Opportunities with Scott Hebner and theCUBE Media

Info
Transcript

Scott Hebner, theCUBE Research

Scott Hebner

Principal Analyst, AI theCUBE Research

TheCUBE is covering events in New York City at the New York Stock Exchange in collaboration with the NYSE Wired community. John Furrier, host of TheCUBE, and Scott Hebner, a new research analyst, discuss Scott's latest notes on AI. They touch on the growth of AI and the emergence of causal AI, which focuses on reasoning and problem-solving. The conversation also covers the importance of specialized small language models (SLMs) and the potential of agenetic AI. Scott emphasizes the need for AI models to understand cause and effect for effective problem-solving... Read more

explore Keep Exploring

What is the new note that Scott Hebner has done? add

What is the expected growth rate for the causal AI market by 2030? add

What kind of companies are being targeted for engagement by the speaker mentioned in the text? add

What company should be contacted if one wants to learn more about the research and insights being published for free on theCUBE Research site? add

bolt Powered by CUBE AI

Scott Hebner, theCUBE Research

search

>> Welcome back everyone to theCUBE's coverage here in New York City at our new location, the New York Stock Exchange. In collaboration with the NYSE Wired community, we're bringing you all the content here on the East Coast. This is our East Coast access point, the network of great experts and all the data coming out of theCUBE. I'm John Furrier, host of theCUBE, and I'm joined here with Scott Hebner, who's our new research analyst at theCUBE Research. As you know, siliconangle.com is where we have our publication and media. thecube.net is our video, and thecuberesearch.com is where our bench of seasoned industry analysts are breaking new stories and new research every day. We're going to talk about a new note that Scott has done. Scott, welcome to theCUBE here in our location in New York.

>> Thank you for having me. This is great.

>> This is your backyard.

>> It is.

>> We've known each other for years. You are executive multiple decades at IBM, product marketing, all across the gamut of many transformation journeys. We're super excited to have you on the SiliconANGLE Media CUBE team, on theCUBE Research, heading up and doing a lot of AI, digging into AI, the AI industry research. All of our big customers are going to love that advisory and all the reports. I saw your first one. We'll talk about it now. So, great work. Just a lot of great stuff.

>> Thank you. This is going to be a lot of fun.

>> Like the scene here.

>> I love it, yeah.

>> This is the center of capitalism in New York City, and of course Palo Alto is the center of technology, but New York has a huge renaissance in technology. IBM stored history. Tom Watson, they were the computer industry. Hewlett Packard on the West Coast. So, I feel like IBM, HP, kind of with that East Coast, West Coast kind of founded tech, the New York scene here is pretty amazing, and you see a lot of big tech companies coming here. A couple cycles ago it was mostly Massachusetts, but now you've got Silicon Valley, New York. So, it's booming and a lot of talking about AI, and this week is even more impressive because the UN's here for the General Assembly and you've got Climate Week. So the theme is really governance, big global issues from supply chain to governing AI to how do we make the earth better? And we've had great customers on. Google brought four customers. So, again, Google Cloud, thanks for them for coming on, but been a great week. This is an AI dream scenario where both the back end and the front end are being upended with disruptive enablers and there's a lot going on both sides of the equation. And you got to do both to be successful.

>> That's absolutely right. And you're right about the tech renaissance here in New York is definitely happening. You mentioned IBM. They just opened a beautiful new building in New York City. So, renovating their presence here. And then you have all these startups here, and like you mentioned, what's going on at the UN. And AI touches all of this, right? We talk about climate and sustainability. Well, the energy costs alone for at least today's AI models is pretty dramatic.

>> Well, I got to say, when we announced that you were coming on board, I know that a lot of people in our community and the tech industry were very energized about you joining the team. You've had a passion for AI for a long time. You've been looking at the new frontier of AI. What's shaping your views?

>> Well, I think we're very much in the infancy of AI. Really in the infancy here, and it's unbelievably impressive what can be done today, but it's really only the start, I think. And I think there's some trends that are starting to develop. Now they are going to converge over time into more of an architectural approach, and I think that is the notion of causal AI, the actual ability to reason, problem-solve, understand cause and effect. I think the long tail of the small language model is going to become more and more dominant. And when you say small language model, you can actually substitute the S for sovereign, secure, specialized, small. If you think about all that, it becomes increasingly important if you want to optimize your model for your unique business, your unique clients, your region. So I think that one. And then agenetic AI, this notion of a bunch of network agents that are collaborating together to actually solve problems, take all those things together, and while the pieces are there today, and of course you mix in things like generative AI and so on, large language models, what's missing today is the architectural confluence of all that into, and that's what I think is going to start to shape the next two to four years.

>> Yeah, I totally love that. They've got the building blocks. It's almost like an operating system is needed at a runtime to kind of make things happen and generate action. I love your comment about SLMs, for small language model, but substituting S for those multiple faceted categories. It's interesting. LLMs, large language models, is another term that people use, large language models, when actually they've been called frontier models because it's not just language, it's computer vision. So it's multimodal. So I like how you went multifaceted with the S, because sovereign cloud, for instance, is an issue. You mentioned sovereign, secure.

>> Yep.

>> Data relevance and data quality is different based upon the situation it's needed in. This seems to be the pocket of area where you start to see people not freak out about, but pay attention to. An SLM could be, "I have a small data set for security, or if I don't have a small data, someone else might have it." We have a small cube set. People might want to use that. So this is where models start to get involved with each other.

>> Yeah, you kind of go back to the fundamentals, the principles or the tenets, if you will, of AI and business. Needs to understand the unique language of your business. It's going to be used for automation of your processes, your workflows, how your customers interact, how you engage with the world, which is always unique per business. And it's got to be trusted. And trust is a lot of different words from sovereign to secure, but explainable, how and why did you do this kind of thing. So you take those three things, and yeah, you get into this idea of the SLM. You can actually throw in small too as being what most people say, but I think it's more about specialized and tend to be behind the firewall. And by the way, they feed off the LLMs, so it's not like either/or. It's architecturally going to come together.

>> Yeah, I think of the work that theCUBE Research did, me and Dave and the team, with the power law. That was two years ago. It's actually playing out to be relevant. I want to get into your research that you just put in and what you're working on. I think folks are hungry for advisory and navigating through the innovation cycle we're in. But before we do that, I want to get your take on something because, knowing your history, you've seen many innovation waves.

>> Yes.

>> Over the years. Transformational. What's the big transformational set of things that people need to pay attention to now, from your perspective? Looking back at your experience of transformational waves, we've seen the movie before, but it's different every time, but now it's even more different because of the relevance of the backend-frontend. What are you seeing? What does your experience tell you about the current state of this wave of generative AI and all the impact operationally, to user experience, to societal change?

>> Well, since I've been around the scene, which is almost 30 years or so, going all the way back to God knows how long ago, kind of reminds me like the Game of Thrones, multiple seasons. To me, it's been one big transformation where we start off by interconnecting everything, then we instrumented everything, and now we're in that phase of infusing intelligence. So the three I's, right? And there's been some, like you said, it's the same movie or plot, just different characters. And there's some common characteristics to each one of these phases, I think. I think what we're seeing today is, again, common phases, is that there's not going to be a one size fits all. It's not going to be dominated by one player. And open standards, open technology that allows you to wire these things together is, what's in the end, going to dominate. So, there's a whole focus these days on open source models versus closed source models. Are they direct, indirect on how you consume them, the pricing models and all that? And if you look at the long tail of these models, and I think Hugging face is approaching a million models on that site. And then, by the way, there's two sides of the coin. There's the marketplace around models, and then there's the marketplace around AI data sets, which are two sides of the same coin and they're both developing at rapid speed in terms of marketplaces. And so, how you wire these things together, and that's why I'm on this notion of architectural approach. It happened with the cloud era, it happened with e-business, and it happened with client server where you start off with the piece parts and then they start to come together, and then before you know it, parts of it commoditize out, like the browsers. Are the large language models going to become the browsers of the future? And I'm talking not this year, next year, but-

>> Their role in the infrastructure of this user experience. That's what you're getting.

>> I think the value over time is going to be more not in these large language models from the closed source ones, but it's going to be in the SLMs.

>> Yeah. And I like your stance on agenetic. As you know, we all, on our team, the Cube Research, love the agenetic wave. We think it's going to be very relevant to SaaS. Category is going to change that category significantly, so we'll see some innovation there. And then guests we've been having on theCUBE here in New York City because of Climate Week, they're solving problems with cloud scale, with the hyper scalers and with supercomputing-like capabilities being democratized. They're solving new, hard problems that are categorically new, that they couldn't do before without specialized high-performance computing. So you're seeing a whole nother category at the top of the stack, large-scale problems and then SaaS evolved into agenetic. I just see the application security, the application platform paradigm, that movie's happening with agents. But this new category. We heard ex-Stanford scientists formed Atlas AI. You met them prior to coming on here. They're building an earth instrumentation engine to solve logistics and supply chain. They're doing two things. Those are hard things, but they can do it with Google Cloud. They're cloud provider. So, again, this is a beautiful time to do this.

>> Yep.

>> What's your research doing now? Take us through your research that's coming out and then what's down the pipe?

>> Well, you just said the key word. They're solving. They're problem-solving. And I think one of the challenges with today's large language models and generative AI in general is that it's based on a correlative design. It correlates across. It basically swims in a large data lake and correlates to identify patterns, associations, anomalies, and they can then predict or forecast or generate something, right? It can do the what. It can tell you what, but it can't tell you how it went about deciding that's the right answer and can't tell you how to do it, and it doesn't do a good job at it telling you why that is the best option among others. So, think of the design point today. It's based on statistical probabilities. Probabilities tell you our beliefs in a static world based on past historical experience or data. Causality tells you how those probabilities change when the world around you changes. For AI to help human beings actually problem-solve and plan, it needs to understand causality and the science of why things happen. And that's what causal AI is all about. It starts out with the notion of reasoning. Everyone knows today's can't reason. OpenAI just came out with their O-One model, and they proclaim that it starts to allow AI that reasons. And I think it's a big advancement on the journey to AI reasoning systems, but it's really, the journey has just begun because it does not understand cause and effect.

>> So reasoning right now is just reasoning off statistical calculations and causal AI, you're suggesting, which is your research area, is going to be where the action is?

>> I think that's going to be complimentary in the roadmap ahead. So, the way I would frame it from a technical perspective is they came out with this new technology called chain of thought. So, in terms for mere mortal thinking, basically what it does is it looks at a complex problem like a math problem. It breaks that problem down into what's called COT tokens, and then it starts to assemble those and process them to solve the problem. But then it will go back in time, re-evaluate its previous steps. It spends a lot of time reasoning, going back. Reinforcement learning's behind the scenes, both in what it does as well as human feedback into it. And over time it gets stronger and stronger to solving these problems, but what's neat about it is it will go back and re-evaluate the problems and do things differently. And it takes a lot of time and inferencing to do that. So, it's actually reasoning, but you're right, it's still based on a statistical model. Really sophisticated, very compute-intensive to do it. And if even if you go ask ChatGPT O-one, "Do you understand cause and effect?" It will tell you no. So, therefore, it's a step in the right direction. And by the way, there's others doing it like DeepMind. They have a thing called STAR self-taught reasoning or something like that. It's the same concept, but to me it's a building block on a longer journey to systems that will actually be able to reason, reason being defined as at some point you have to understand cause and effect.

>> So take us through some of your research you're putting out now. What could people expect?

>> Well, I think what's happening now is you have the advancements in, as we said before, the SLMs. So they're becoming more specialized. There's data sets that are coming along with that. You got the whole agenetic AI thing that's maturing at a rapid pace too. And you have these technologies that are slowly but surely going to start to allow these AI models to do a better job of reasoning. OpenAI, with their chain of thought, is one of the building blocks. At the same time, there's a new market developing called the causal AI market. And depending on who you look at, I think the consensus for you, it's about a 40 to 50% compounded growth rate between now and 2030. I think it's going to get up close to at least a billion dollars, but some of the estimates are a little bit less. I think it comes down how do you define it? But if you define it in terms of vendors or companies that are providing new tools, new methods, and new platforms, it's about a 40% growth rate I don't know, 10 or 12 companies out there today that are doing quite well. And you're hearing it, you heard it in some of the interviews today that people are starting to do this. So I think that starts to get fused together. I'm going to write another research note. My next one will be about all the different methods under the umbrella of causal AI, and then again, architecturally, they come together.

>> All right. Well, I really appreciate you coming in on this day and getting the update on the research. Obviously we love what we do. We do a lot more CUBE interviews together with Dave Vellante and the team. And again, this is going to be an area that's going to open up its aperture big time. Take a minute to talk to the audience about what you're looking for, who you want to engage with, kind of briefings you're doing. How do folks engage with you? Who do you want to talk to? Take a minute to put a plug in for what you're working on and how people can connect and add value and collaborate with you.

>> Yeah, one of the things that I pride myself on over the years is I have a big network of experts. They're business leaders, they're technologists. I love engaging across that network. I want to continue to build it because my view is strategy is only as good as the insights and the consensus is built upon, and never think alone. Never go it alone. So, I love to build partnerships with people and learn from them. I'm going to focus these initially on, as you kind of pointed out, the next frontier of AI, what you need to start preparing for now to be ready in the next year or two. There's a lot of people talking about gen AI. I want to try to fill that spot, that don't be a laggard in the next phase. The movement around AI reasoning and causal AI, I think is not only necessary, but it's inevitable. People are going to want their AI to help them problem solve. It's going to happen, guaranteed, a hundred percent.

>> So you're looking for experts to answer your network. What about companies out there? Looking for certain kinds of companies, and if they're watching, what's going on in their world that would motivate them to contact you to engage?

>> Yeah. Well, one is, do you want to get some insights into where this is all heading? Two is, do you want to get introduced and understand the 12 or so platform companies that are now building causal capabilities? There's another whole wave of dozens of companies that are pharmaceutical and so on that are infusing causal. We just met one, for sustainability. I would like to learn more about what they're doing and how we can help them. And then, finally, any company, business that's actually starting to use this to learn from and where we can help. And again, there's a ton of companies. CausaLens just had a conference in the UK about two days ago. They have 26 businesses there showing off what they're doing with causal AI. So, this is all happening and I want to engage across that community that's doing it to learn and then take that learning and transfer it to other companies that want to start to get involved.

>> And just one final question. The speed of acceleration here is scope the speed of the change. You mentioned a causal event that you attended. What's the speed? What's the pace of play?

>> I think it's real now. I think there's a lot of experimentation. There's real use cases that have real ROI that have already been achieved by many pioneers. I would say the next couple of years is when it's really going to accelerate. I'll tell you the Gartner, if you look at the Gartner hype, they have it as a high-impact technology in I think the two to a four year timeframe. I think it's going to be a little quicker than that. And things like what OpenAI just did is just going to be a spark for people to think more about. Like, "Wow, these things don't reason."

>> So, if you're a company, you got to act now. Contact Scott, of course. Thecuberesearch.com is the site. Go check out under the research piece, the content's being published. It's all free. We have free insights. Of course, we do engagements and of course they host on theCUBE, and again, we're bringing the real-time, high-frequency insights as best we can to you. It's our mission here at theCUBE and SiliconANGLE Media and theCUBE Research. We do our best. Engage with us. We'd love your support. And again, love the collaboration. Okay, thanks for watching here. It's our NYSE location, Cube East. I'm John Furrier, host of theCUBE with Scott Hebner, AI analyst with theCUBE Research. Thanks for watching.