From the course: DJ Patil: Ask Me Anything

How can AI and machine learning (ML) help cybersecurity?

From the course: DJ Patil: Ask Me Anything

Start my 1-month free trial

How can AI and machine learning (ML) help cybersecurity?

“

(soft music) - [Interviewer] What insights can AI machine learning give us about cyber security? How can it help us? - Yeah it's actually might be one of the most pressing problems right now. So first how do we see machine learning and data science and all this come together for cyber security? Well one of the interesting insights was actually the early days of LinkedIn. We had a great data science team, much of which had been built out of people who came out of National Security after 9-11, so finding signal in the noise. People who had worked at Ebay and PayPal finding fraud and fighting fraud systems. And so one of the things that we realized is, well, we're building all this technology and techniques in data science to build these products, we can also use that to fight the bad guys who are trying to attack. And use that to create leverage to stay ahead of them. It's a constant evolving game. In the same way right now we have a real challenge because one of the biggest impacts where you can have, and have benefit from, AI and machine learning, is that you have an attacker who's constantly changing. So they've evolving over time. So if you create a static model, you have to update that model as you realize how the attacker is changing. But you're always lagging behind the attacker. This is the attacker, you're kind of always slightly trailing. So you're always losing. Machine learning allows you to stay much closer and adapt much more effectively to how they're operating. And so one of the things that we built and many people have iterated on now, is a lot of these fraud systems are machine learning based, they are AI based. And that's how we rapidly retrain them and get them to fight fraud. The bad guys have figured this out too. And as a result what they've done is they've realized, hey wait, I could use AI and machine learning to attack. And that attack happens in a few ways. One is, if you're just trying to poke and see where holes are in a system, that's a great problem for a computer. 'Cause it's like test, test, test, test, iterate, what did I learn, what did I learn? Figure out, modify, try again. And so finding a hole in your system is a machine learning, AI problem. So how do you fight that? The only way to fight that is going to be with machine learning. The other problem that we're starting to see is new attack vectors that are created using AI and machine learning. So for example, these ransomware attacks that have been happening. Well one of the things that we've started to see is, the way you get into a system by providing malware and then being able to have the system locked up and telling people, hey you've got to pay us a certain amount to get it unlocked, how do they do that? One of the ways they start using is social media, like links on Twitter and other places, that are adaptively tuned based on machine learning. So they try a certain messaging, did it work, did it not work, let's try another one. And because the volume is out there they can test so many different variations, they find the optimal one that is going to get you to basically click on a link. Similar versions are starting to happen when you get all these robocalls. Is they're trying to figure out, hey, is that a real human there, is it not? And then start trying to guess about this. And you've seen these chat bots that have been, people have a hard telling is this a human or not? So all of that type of social dynamics is being deployed to insert traditional malware. And that's a new form, so how do we fight that? Once again, it's going to have to be machine learning and AI. (gentle music)

Contents