Interview: Christian Morin on AI – ‘We cannot just use it without understanding what it truly is’

So first of all, I think there’s a growing movement towards the cloud that’s going to continue; not necessarily a full-on move to cloud, but more that customers start incorporating some aspects of it in their environment. Our position on that is – is we’re on a hybrid first trajectory. Therefore, we believe that everything’s going to be hybrid at one point in time. It’s not necessarily a question of full cloud versus no cloud. In many contexts, you actually need to have something that works onsite, because it offers physical security, right? The physical aspect is rather important.

I think there’s still a lot of AI hype, and I think we haven’t seen the last of people using AI in a not-so-successful fashion, but we’re going to continue seeing AI being incorporated throughout different stacks. And I think this is more of a macro trend that we’re seeing in technology, throughout our industry.

I see it more used as a replacement or enhancement of computer vision. I don’t see it used extensively outside of that space just yet. We’re working on potentially changing that, so we’re going to actually see more AI stuff. So hybrid AI and regulations … Regulation is actually something that’s interesting. We’re looking at a combination of regulations from the security and privacy perspective in some places, and a lot of regulations as they pertain to protectionism.

What’s happening in the US with the Cloud Act, and specifically in the European Union, I would say we’re going to start seeing an acceleration of potentially moving away from American cloud service providers and more so towards sovereign clouds. So that’s going to possibly even put a little bit of a dampener on the full-on go to cloud aspect, because there are not that many options.

So the hybrid option, where you can potentially store your data where you want, is going to have a play. So this is what I see: three big things in the not-too-distant future that are going to start accelerating or continue on their trajectory.

Cyber Daily: Speaking of AI, for a moment, there are two general viewpoints we hear a lot. That is from the AI evangelist, who believes it’ll change everything, and then there’s the other side, which is basically that AI is big enough to fail and the bubble will burst. Where do you stand on that?

VIEW ALL

Christian Morin: I’m somewhat kind of right on the fence on that one.

I believe that AI is, in general … It’s a great tool, right? We just have to use it wisely, and we cannot just use it without understanding what it truly is, right? It’s all based on math, and if you understand how AI actually works, you know, there’s a limit to how people actually understand how it works.

But we kind of understand, generally speaking, the mathematics behind it – how it infers and guesses at the best possible solution based on what you’re inputting. So there are limits to its capability. So if you have good training data, if the problem is well-bound, you can actually use it quite efficiently.

And if you ensure that for more mission-critical, like safety-critical type situations, that you put a human in the loop, I think there’s quite a lot of good that can come out of it. As I said in my previous comment, I think we still have, unfortunately, some people who are using it without really knowing what they’re doing. I think we haven’t heard the last of stories of how AI did stupid things.

Cyber Daily: We can believe that! Tell me more about how Genetec is using AI, especially in that vision sector.

Christian Morin: For us, AI is not, first of all, the be-all and end-all. It’s not a magical silver bullet. So you’ll never hear Genetec market AI this, AI that, AI data … What we’ve coined is intelligent automation, which implies that there’s a human in the loop, and we’re helping the human be more efficient, helping automate a lot of the tedious tasks, and ultimately augmenting their capabilities.

With respect to computer vision, what’s great with AI is actually there’s not just one mega AI model – it’s actually a lot of different models – so it’s how you can layer the models together and pipe information in a cost-effective way, which is actually quite important, and also leveraging as much as possible.

Our strategy is to try to distribute computing as much as we can, because it is very power-hungry, and not just boil the ocean either. So how can we use the information that’s provided by the edge? In cameras nowadays, object classification and a whole bunch of analytics are done directly. So this provides a stream of metadata, and that stream gets, I would say, augmented over time.

What we’re doing is that we’re seeing how we can use various models to augment that data, and how we can also augment that information in conjunction with potentially other sensors that are in the same proximity or vicinity of the camera, and that’s when you start layering things on top of each other. So as you’re presenting the information to the operators, as they’re looking for something, you’re able to suggest items.

For example, if you’re conducting an investigation, you’re able to recommend or suggest other nearby individuals, not just solely on the video data, but potentially on other data that was captured by the system. It could be access control, could be license plate data, could be whatever it might be. It’s a combination of layering, not reinventing the wheel, not going and taking the job of the camera vendors – the camera vendors continue making even better analytics.

And the more that you feed us data, the more specific, and the less heat or energy our own models will consume. And then we can have an overall better solution.

Cyber Daily: It’s refreshing to hear someone talking about how there is a need to not boil the ocean. How do you tackle that, particularly in a world where AI is absolutely doing that?

Christian Morin: It’s how you train the models, right? So when I’m looking at how I’m going to query video, what do security professionals typically look for?

There are typical types of queries, and there are typical types of things that we look for – you know, a man, a person, a van, the colour of the van, it might be a make and model, it might be a delivery truck, it might be FedEx, it might be something else. But what are you going to be looking specifically for?

Suppose you have a good sense of what people actually search for. In that case, you’re able to optimise your model so that it doesn’t just classify anything and everything, and ultimately, literally, just boil the ocean. So that’s typically the brute force model. You don’t need these ultra-generalist, foundational models that really have everything. You can actually narrow it down to something more fit for purpose, and this is what’s going to – I believe – distinguish the men from the boys down the road; how you can actually use these tools in a more efficient way.

Not only does it allow you to consume less energy – and generally speaking, that’s a good thing – but it also allows you to run some of these models, maybe closer to the edge and distribute them more efficiently, and also makes them more broadly available to users, because they’re less expensive to run.