Podcast: Siloed IT and OT, outdated wireless among top hurdles to scaling industrial AI

Below is an excerpt from the podcast:

Scott Achelpohl: Samuel, my read of the new Cisco report, where industry industrial AI is concerned, is that the findings place outside emphasis on modernization of technical and network infrastructure in most manufacturing operations, for AI to be utilized at scale and on the need for IT and OT staff and systems to converge and collaborate.

I guess I'll open it to you to talk about the report. Is this your interpretation of the study? What else would you like to add?

See also: New Cisco AI study sees widening execution gap, strain on manufacturing infrastructure

Samuel Pasquier: For the audience, everyone knows about Cisco as an IT company building networks. We helped to build the internet for the last 40 years, but we also have been building industrial networks for 20 years.

So, we are helping our customers in manufacturing environments to build their infrastructure, to connect their machines, to connect their plant, connect the factory floor, and two years ago we wanted to get a little bit of a state of industrial network, and we did a similar report, and what was very interesting for us.

See also: Workshop confronts manufacturing execs with the big stakes that ride on proper cybersecurity protocols

At that time, the majority of the respondents told us that AI will have the biggest impact on industrial network over the next five years. So, we are three years along, and we thought, you know what, let's double-check. Let's understand what's happening with AI in industrial environments and what the impact are that we see on industrial networks.

So, we talked to 350 manufacturing customers, really operational leaders in manufacturing environments, to try to understand what's happening and what we are delivering in this report. We can go through it a little bit and talk about it today.

SA: OK, Samuel, let's get into it some more. We have some questions, as you might imagine. Samuel, what do you make of the disconnect identified in the report between AI deployments, as we mentioned, 68%, and the much lower percentage, 19%, that regard their deployments as “mature”?

SP: The key things about that, and you know, we have to think about maybe the obstacle, right, to be able to scale AI. So, we see a lot of people—we talked about it in the previous part that we've done together—but really what is very clear out of the report is there's a few things that are hindering the deployment at scale, infrastructure limitation.

We talked last time about the use of machine vision, quality inspection, those kinds of things. Once you want to have a more global view of that around your entire infrastructure, then you need to have more performance, you need to have more bandwidth, you need to be able to store more data, to be able to have more correlation between the different information.

See also: Implementing a true approach to PLM, the cornerstone of manufacturing

That’s one of the limiting factors. The second thing that has been an obstacle is really security. As you connect more and more smart assets, which means assets that are talking, which means they are connected to the network, you’re increasing to some degree your attack surface.

So, how do customers reduce the “blast radius” so they can get smart things that can talk, while at the same time not increasing their exposure to threats? And the last, which is really linked to this one, is you need to connect more things. You need to care about security, but there is a fundamental skill gap.

You need to have people who understand the industrial environment, to understand what you need to do with your machine. You know, we think about manufacturing, industrial automation.

See also: Rockwell: Top-performing OEMs likelier to use tech such as digital twins, cobots to cut downtime, speed recovery

But at the same time, you need to have people who have this security mindset and understand what needs to be done in security. And this skill gap, having the large number of people at scale to be able to do that, I think that's one of the things that is limiting the massive adoption or the scale of some of those AI use cases, right?