What are your top challenges implementing AI/ML?


1.3k views1 Upvote7 Comments

Head of Information and Data Analytics in Software, 5,001 - 10,000 employees
Using the basic statistical modeling and unsupervised learning doesn't help because cyber security doesn't have a label data concept.  You go to Netflix or Amazon.  They look at users behavior and they’re able to tune the algorithms, and increase accuracy. But in cyber security, there's no label data. The traditional approaches of figuring out signal versus noise did not work, so now we are looking at newer ways of doing things, where you're teaching the machines to think like a self driving car.  Meaning, not just through the unsupervised learning, but looking at how a really skilled human would go figure things out.  So that's where this is headed right now.
1 Reply
Sr. Director, Security Engineering & Threat Management, 10,001+ employees

There’s a challenge around...how do we eliminate misconfigurations from a security baseline and expedite the quality of high fidelity alerting to a limited set of analysts, so you don't have to throw bodies at problems that you historically had to solve for.

Sr. Director, Security Engineering & Threat Management, 10,001+ employees
How does one as a security practitioner or an engineering leader go about identifying capability and quality? There's a lot of new things getting released and dropped, but it's not just about evaluation of capabilities. How, as a security or engineering practitioner, do you go about doing your assessment to see where you add value? You could be in a really mature landscape where maybe some of those AI or ML type rules that certain products offer don't provide you as much value upon an assessment or a POC. Does that mean that the product really doesn't add value? Not necessarily.  So as we have this kind of dynamic, does that mean that AI and ML tools will always benefit a less mature landscape better? Possibly.
1
CISO in Software, 51 - 200 employees
I'm an infrastructure guy. I do some security, but mostly infrastructure. It was around 3-5 years ago where AI help desks and bots started coming along and maturing. At my last company, we kind of dabbled in it and pushed ahead with a company that helped us. They claimed that they had closed 40% of our tickets.  It didn't turn out that way.  Maybe it was the way that we actually designed it and the way we're inputting and outputting our tickets. We were in the middle of crazy growth too. So we didn't have time to concentrate on a lot of tools at the time, but what it did do for us, once we got our KBS and everything in there, people started using it and just asking basic questions, like where do I find my benefits? How do I get to this thing? How do I use zoom?
2 Replies
Chief Digital Officer, 201 - 500 employees

We invested in AI platforms for consumer behavior. So we've had it for a while and now we're taking it to the next level with a new CDP that we just got.  We're going to be doing a lot of prescriptive modeling and so forth.  That works perfectly because it’s what we use to do what's called lookalike modeling (i.e., go out and look for these kinds of people so I can market them).  That's one use case that has been very, very heavily used within our organization. There's two others that we're investing in right now. One is exactly what you're talking about, where people claim that they can do X things with the data that you have...until you put it in action. You don't have a lot of history, you don't have a lot of stuff, and the AI cannot learn.  So that's where we find it a little bit challenging: having that good data source, having a plethora of data, and then using that to solve for specific issues. But if you just put something in and turn it on and expect it to work, it really doesn't.

Head of Information and Data Analytics in Software, 5,001 - 10,000 employees

It's about looking at process time, right? When people say AI is snake oil, there are certain reasons behind that. People have built their products doing mostly unsupervised learning. What that means is, when they are coming and learning in your environment, initially, there'll be lots and lots of false positives because there's no level data. So unless somebody invests six months to a year in terms of reducing the false positives and tune the algorithm, they're not going to get to value. There's only two things here:   either you are ready to invest a year into that, or the older way of doing things where if a company has already invested their time in terms of creating these labels, you would get programming, which means these algorithms are already tuned so by the time they are applied, the time to value will be really, really quick. So, it depends.

Sr Director of Information Security and Compliance, 1,001 - 5,000 employees
We've got the same issues that some companies face, like visa does, where you're talking about millions or billions of transactions a day coming through the systems and you need to be able to quickly parse through there and figure out what's going on. So for us to make our products work and for the customers to get what they need out of it, the system has to be able to quickly discern what's normal and what's not. And come back and say, “Hey, this is something to look at.” And that's a constantly evolving process for us, making sure that that works in a way that not only is efficient, but it comes back with an answer that's not two days later. With attacks, it's fast. You need to know now. You need to be able to say, “something's happening,” two seconds ago, or “it’s happening right now,” instead of “hey, I crunched this overnight,and I think you may want to check on what happened yesterday.” And I think one of the big issues that we see there and why you see our big platforms are on the cloud is that it requires a lot of computing power. And, that's one of the big reasons that you see our product is now SAS-based. Because trying to do it on premise for customers means you can't get the horsepower that was required to get that data crunched fast enough to give people access to what they want without having huge gobs of machines. And the only way to do that today is to use the cloud.
1

Content you might like

Yes, AI has significantly reduced costs and improved customer experiences.4%

Somewhat, there have been some cost reductions and customer benefits, but there's room for improvement.82%

No, AI implementation has not yielded noticeable cost savings or substantial customer enhancements.11%

Not sure / I don't have enough information to assess AI's impact.4%


28 PARTICIPANTS

231 views

Yes, we're actively using it37%

No, but it's part of our roadmap52%

No, and we have no plans to10%

Unsure0%


205 PARTICIPANTS

958 views1 Comment