How internet search algorithms could help critical infrastructure

Bill Kay: When you imagine ranking websites by importance, the same sort of mechanism goes into determining what is important about infrastructure assets when you consider the system as a whole, the grid where things are linked up.

Tom Temin: In other words, you’re looking at dependencies from what I’ve read. That is to say, this water pump is not the biggest part of the infrastructure, but it could have a lot of effect on things downstream of it, for example, should it get jammed.

Bill Kay: Yeah, absolutely. A power plant gives power to a water treatment plant gives water to a hospital, right? And if the power plant fails, then the water plant can fail. And if the water plant fails, then the hospital can’t provide water to patients.

Tom Temin: All right. So we have critical infrastructure, let’s say the electrical grid cascading is a problem that we know from history has caused widespread failures. One thing leads to another and next thing you know, too much current is cascaded down on something that can’t handle it and then there’s a blackout. I lived through a couple of them, actually remember the blackouts. And what is the database of the dependencies that you have to evaluate in the first place because in internet searches, there’s activity that has created a history that gives importance rankings to the algorithm. How do you establish that in, say, a chain of critical infrastructure components?

Bill Kay: The work that we did used one particular data set, the all-hazards analysis dataset was curated by domain experts. And so this is the kind of thing where we wanted to demonstrate that our method was a good way to stop cascading failures in a network. But this is the kind of thing where maybe domain experts would come to us and say, ‘We have a network, we’ve written down the dependencies and we are concerned about what cascading failures spreading throughout this network could do. Can you tell us which are the most important things to stop from failing?’

Tom Temin: So they know, therefore, what their dependencies are. What is the element they don’t know then? If they know the dependencies, shouldn’t they say, ‘Well, this is what absolutely can’t fail?’ In other words, it sounds like there’s a ranking here missing.

Bill Kay: The place that this kind of came from was a Google search algorithm called PageRank, right? And once upon a time, websites were ranked highly if a lot of other webpages pointed to them just directly with hyperlinks and people figured this out. And what they did was they spun up a bunch of fake webpages that didn’t say anything to link to their webpage to artificially inflate the ranking. And so what Google said was a webpage shouldn’t just have a bunch of links to it. It should have a bunch of webpages that themselves are important point to it and those things are important because lots of things point to them. And so you can’t just look at one asset and say, ‘What are the things that point at it?’ It needs to depend on a lot of things that are themselves important and those things need to depend on a lot of things themselves important. So Google’s algorithm takes a holistic view of the entire network structure and ranks the webpages based on how important they are in the whole system rather than based on a local thing.

Tom Temin: We’re speaking with Bill Kay. He’s a mathematician with the Pacific Northwest National Laboratory. And did you have Google’s cooperation in this investigation because that algorithm is kind of an important asset for them?

Bill Kay: Oh yeah, there are implementations of PageRank in a lot of scientific computing things. So it’s, it’s it is not a guarded secret. I mean, probably the nuts and bolts of how Google’s internal algorithm actually works is theirs, but how PageRank works is just a linear algebra thing. It’s got a Wikipedia page. It has implementation.

Tom Temin: So it’s almost an open source type of thing.

Bill Kay: Yeah, absolutely, yeah.

Tom Temin: And just to be clear, you used the more recent holistic version of the algorithm and not the original sheer page rank based on link’s version.

Bill Kay: That’s correct, yeah.

Tom Temin: All right and give us an example of how this might work. I mean, I’m a critical infrastructure operator. I want to know what’s important, what thing I absolutely don’t let fail. What’s my process? What’s presented to me to be able to effectuate that?

Bill Kay: It’s less that you are operating one infrastructure unit and more. Let’s say you’re in charge of distributing resources to defend infrastructure and you’ve got, let say your budget limited or time limited or something and you can protect 5% of infrastructure assets to stop them from failing. What we can do is look at the dependency network. And say which things are the most likely to get caught up in a cascading failure. So if something random fails and the failure is spreading, this thing is likely to hit and also if this thing fails, the failure will cascade the farthest. And so what we said was we can rank all of the infrastructures in, say, a power grid or something like this. Say if you’re budget limited, let’s grab these top 5% and make sure that they have backups or something like that.

Tom Temin: You can take almost what they would call a risk management approach to it.

Bill Kay: Exactly.

Tom Temin: And do you think this works in other than physical infrastructure domains, for example, the banking system where cascading failures or real estate, that type of thing?

Bill Kay: PageRank has been used to analyze a number of different network constructions, but one thing that we did that was, I think, somewhat novel is that we actually cared about both things getting caught up in the Cascade and how far they magnified the cascade going forward. I looked and I haven’t found anything that has considered both of those simultaneously. And so I think the part of the really the novelty of this work is that we care about getting caught and spreading a failure.

Tom Temin: Sure, and just briefly tell us a little bit about yourself. You are a mathematician. Where did you come from and how did you end up at PNNL?

Bill Kay: Yeah, that’s right. So I’m actually a pure mathematician by training. I got my Ph.D. At Emory University in Atlanta and the field that I studied was, it’s called graph theory, but if you ask people in the applied space, it would be called network science. And I was sort of headed towards academia and took a different track and wound up at a lab and I get to work on these really cool applied graph theory problems that hopefully can be impactful and helpful to keep things working.

Tom Temin: By the way, is the chalkboard still a useful tool in the pursuit of mathematics?

Bill Kay: Yeah, I got one on my wall. I got a chalkboard and Hagoromo chalk, the fancy Japanese chalk. I don’t know if you’ve heard about this.

Tom Temin: Not personally, no.

Bill Kay: Yeah, OK, yeah. Mathematicians love it. Doesn’t squeak, doesn’t get dust all over you, you don’t get the white lung.

Tom Temin: All right. Well if I ever need a blackboard, I’ll try it.