CONTRIBUTOR
General Manager and Editorial Director,
Techstrong Group

Synopsis

In this Digital CxO Leadership Insights series video, Mike Vizard speaks with Rachel Roumeliotis, VP of Content Strategy for O’Reilly Media, who explains how the AI skills gap will be narrowed.

 

Transcript

Mike Vizard: Welcome to another edition of the digital CXO video news interview. I’m Mike Vizard and we’re here with Rachel Roumeliotis. She’s vice president of content strategy for the O’Reilly Group, and they have a new survey out talking about the skill shortage as it relates to AI. Rachel, welcome to the show.

Rachel Roumeliotis: Thanks so much for having me.

Mike Vizard: Walk us through what the survey found exactly. We’re always seeing skill shortages these days, but what in particular about AI is unique as it relates to skills or our inability to find those skills?

Rachel Roumeliotis: Yeah. So, that is really the number one barrier to moving forward with AI projects and enterprise. And what it seems like is that there is a dearth, maybe not a dearth of data scientists and data engineers. But then on top of that, if you want to actually be building your own algorithms and your own sort of like custom products, you need an ML engineer. That is not always the case, I think as well, to be clear. That’s not always needed.

And so, I think one thing that you’re seeing and that we saw in this salary survey is that people are clear that they need to know more about the cloud. And the cloud is actually I think picking up the empty space that’s left by there being not enough AI engineers. And so, they’re providing services instead, or other companies like H2O, which was very high in terms of the salary that a person could get if they had knowledge of it, that is a whole platform that can help you not necessarily need a full team of AI engineers, but just some really great data scientists and data engineers so that they can plug that information in, use the tool, and then get what they need out.

Mike Vizard: So, it’s almost like IT abhors a vacuum, so we’re spending more time automating processes, so we don’t need as much dependency on people that have those skills. And is that kind of a continuum that we’re going to see or are we going to stay on this curve?

Rachel Roumeliotis: I think that we are going to continue to see cutting edge AI projects need real people. And those are going to be the fun stuff that probably comes out of Silicon Valley. Things like layering AI on top of analytics or a recommendation engine just to give a couple examples, those don’t necessarily need you to create a new algorithm. So, it’s more about making sure that you have your datasets, making sure you have the compute processing of the cloud to be able to do something.

And then the auto ML that is available in the clouds, being able to do that sort of production and governance to some extent and get that out there. I also think that we’re seeing sort of the next round of companies getting into AI and realizing that’s something that they should do. And I just don’t think it’s necessary, I guess for every one of them to have AI engineers, but they’re certainly not _____ _____ either.

Mike Vizard: So, what we’re starting to see is auto ML is a framework that is accessible to almost anybody with a reasonable amount of IT skills, and people are using that to democratize AI essentially, and use that to automate some very common everyday processes for my own company, and I don’t have to be a rocket scientist to figure that out.

Rachel Roumeliotis: Correct.

Mike Vizard: And then as part of that it also seems I don’t need as much data as I used to need for the training AI model either. And that’s part of the whole democratization process as well, right? We’re starting to see small datasets can be used with auto ML to train some specific function or task.

Rachel Roumeliotis: Yeah, absolutely. And what I find really cool too, and someone was saying it’s super important to use outside data as well. There’s these like data markets now that people are using. There’s the open data, which is why open data is important as well. So, you don’t always want to have to pay for it. But yeah, I think that goes along with the, unless you’re doing some super specialized AI project, you don’t necessarily need this huge library of data. Or you don’t need like so many data sets. Like good enough is good enough.

Mike Vizard: The downside of that for the AI specialist is it tends to act as a governor on their salary demands, because basically people are starting to say, “I don’t necessarily need to hire a specialist for every little thing. I’m going to reserve that for something that has massive impact and is a lot more complicated.”

Rachel Roumeliotis: Yeah, I think so. What I generally saw was, and I think you see this across sort of technology. I don’t know, you tell me if you disagree, but when something is new and specialized, the salary goes higher. As it sort of comes into the mainstream the salary goes down a little bit, and as it exits to mainstream, it goes up a little bit again, because people forget that knowledge. And that’s really what I saw sort of like through the salary survey specifically. Because you looked at things like, what do we have here?

So just the programming languages, for instance, right? So Rust, which is a relatively new language, the people who use it love it, but it’s a small group of people right now, but that was the highest average salary. If you know Rust, there are not enough of you. Go the next one, a little bit more, Scala. And the next one below that, Scala, I think is high because it is like the native language of Spark.

But I think it’s more, it has to do with supply and demand. So, it’s about getting, in terms of like data tools, what are people using? Did someone walk into H2O for instance, which I saw was really high, but not so very many people use it. So, there’s not a lot of people around to take care of all of your H2O needs.

Mike Vizard: So, what’s your sense of what is the impact of all this going to be? It almost sounds like if I want to automate something end to end to the nth degree, I probably need a bunch of specialists and I’m gonna invest a lot to do that, but it also feels like there’s a middle ground now that’s emerging where I can automate a wide range of things to augment people, and even the average mere mortal can do that now. And that we are going to see some advances in automation and productivity, but not to the point where we’re replacing people as much as we’re just letting them do more in a smaller amount of time.

Rachel Roumeliotis: Yeah. I think there’s going to be two paths. I think for the, let’s say it’s the 80/20 rule, like 80 percent, I think eventually will find that the tools the cloud providers give you in terms of the auto ML, that’s going to be enough for the AI needs of your particular company. Then there’s going to be 20 percent, they may take advantage of some of those things, but need to delve deeper into the guts of the AI project. So, I think that’s what we’re seeing. And I think it’s the 20 percent that have been doing a lot of the work upfront, obviously right now.

And I think that 80 percent, we did another survey about AI adoption in the enterprise, and there weren’t, I think there were maybe like 20, 30 percent of people looking at the auto ML solution. So, I think there’s a huge upside coming in the next five to ten years in that area as business folks really understand what like automation means. I think one of the early things that we’re seeing right now in like regular sort of business world, not necessarily like manufacturing, is this switch from monitoring your sort of infrastructure to observing it, right?

And observing it as proactive, it’s automated in many cases, there’s AI layered on top of that. That we’re seeing people actually, there’s been an uptick in people reading about that, wanting information on that. So, I think as AI and ML break down into realities, I think that we will see more use of AI overall, but I do think the faster growth area is in the auto ML.

Mike Vizard: Alright. As you look down the path, then do you think that C-level executives are overestimating or underestimating the impact of AI as they’re on their processes in general, or they may be banking on having too few employees or are they buying into the science fiction or do you think that they got it right?

Rachel Roumeliotis: So, I think maybe underestimating a little bit. I know for us, we know about it and we still probably could use more people that are fluent in AI and figuring out how we can incorporate that into business. Like I said, I think there’s still like, oh, we’ll do AI for analytics and we’ll communicate with our customer. It’s not a standard yet. And so, I think until people see that observability stops Facebook from going down for six hours or something, that people won’t understand its worth.

So, I also think that the term sort of AI and ML sort of isn’t really used. I don’t think, like we talked about like observability. I don’t necessarily think people think, oh, that’s AI. I think that there’s going to be lots of AI seeping into business. I don’t think it’s halfway there yet, but I don’t think it’s going to be called AI either. I think it’s just going to be about like automating processes. I think it’s going to be about, this is how we do analytics now. You know what I mean?

And it’s just like, I think what people do understand, and this is a little bit from COVID honestly with people I’m seeing like predictive analytics and stuff like that and it becoming mainstream is that the data sets are just, they are becoming huge. And if you want to process them then you’re going to need some AI support to do that. So, I think in decision-making it makes sense, like people are starting to get that, that, like I would like to use AI for that, but what does that mean? I still think there’s a ways to go there.

Mike Vizard: Do you think organizations have a good handle on their processes? The thing about AI is you want some level of structure in these processes that I’m going to automate. And I would posit that most of these businesses, their processes are decades old. There might be more exceptions than there are rules in the processes. So, do you think at some point organizations are going to have to take a step back and go, we need to clean up our processes before we can even think about applying AI to something that doesn’t really have a logical flow in the way that a machine is expecting?

Rachel Roumeliotis: Yeah, definitely. I think there’s a lot of cleanup to do with data and the processes around it. I think that one of the early things to do, well, first have like a data governance policy, right? But then on top of that, there’s an AI governance policy. So, I do think that there is a lot of work to be done before you even get to, we’re going to incorporate AI. But I do think you need to look at it through a lens of we want to use AI and this is why it’s so important. The data quality, the lineage, all of that type of stuff.

Mike Vizard: So, what’s your best advice to folks if you’re a business leader or an IT leader right now? What would you be thinking about as it relates to AI? Am I just doing random experimentation or am I thinking of a strategy or how do I get there?

Rachel Roumeliotis: I think the best-case scenario is obviously you want to be able to understand the impact that it can have. And I think you want to understand how much impact data has on AI. So, understanding that first, then I would look out to case studies and what other people are doing in industry. So, that gives you a better understanding and sort of an anchor into starting to think through what can this actually do?

Like I said, there’s just this little bit of a divide that is closing, but it’s not closed between how does this, not only how does this affect my business, but how can it improve or how can it innovate? We’re seeing banks do that all the time. Upstart is a company that is now doing loans through like an AI process. So, I think it’s understanding. I think it’s bringing that together. And then once you know that, I guess I would start with looking at, how is your data health, right? I would start with that.

And then, yeah, you have to get some really savvy data engineers. The other side of the things that we haven’t talked about is like that really understand the cloud, because that also, I feel like is what makes AI possible through the compute power there. And that’s something we saw on the survey too, which was that I was surprised by that, that more people than we expected had like certifications that were data engineers that had certifications in the cloud.

Mike Vizard: It’s like that African proverb, if you want to go fast, go alone; if you want to go far go together, right?

Rachel Roumeliotis: Yeah. Totally.

Mike Vizard: All right. Last question to you. Do you think that we get the legal ramifications of all this just yet? Or a lot of business people I talk to are excited about AI, but they’re hesitant because they’re not quite sure what the compliance regulations may allow them or not allow them to do. And that’s got them pulling back a little bit, ‘cause they don’t want to build something or invest in something only to have to take it apart a few years from now. What’s your sense of where are the governments, the legislators around AI? Is this all to be worked out? And how far are we?

Rachel Roumeliotis: I don’t think we’re far at all. I do think that, I just don’t think we’re far at all. Again, if you look at this like government, I think is behind on adopting AI and paying people to work on AI. I think, I’m sorry to say, I think it’s going to be a while before government comes in and gets its hands on AI.

But I think in the meantime, there is a lot of discussion, although it’s not like the first thing people talk about, about transparency and ethics and bias. And I think you’re going to see, honestly, the people I hope in like regulated industries, like health and finance and insurance say, realize that there is a need for that and move forward.

Otherwise that’s sort of I think when the government’s going to come in like Upstart or something, who seems great, I don’t know. Maybe they’re like, wait, how are you doing that? And if they can’t explain that, then that’s going to be a problem. So, I think, unfortunately we’ll probably continue to have, like I don’t think you should be afraid to move forward, but I think there will be periodically sort of terrible news stories that come up about this, but I don’t think it will stop you in your tracks with what you’re doing with AI.

Mike Vizard: Well, to quote Ronald Reagan, nine scariest words. “Hi, I’m here from the government and I’m here to help.” Let’s see how it all turns out. Hey Rachel, thanks for being on the show.

Rachel Roumeliotis: All right. Thanks so much.