00:00:03 Tom Constable 

Hello and welcome to episode four of the Warfighter podcast I'm Tom Constable, and this is Colin Hillier, Mr Jetsetter. New in from Itsec 2022. And where are you now? 

00:00:15 Colin Hillier 

Just a few days off, but yeah, so recording from hotel room, I'd sort of forgotten that I was due to. 

00:00:21 Colin Hillier 

Some leave but and and I'm not sure what time zone I'm in sleeps difficult, but hey, crack down. 

00:00:28 Colin Hillier 

I've been hearing a bit more about your. 

00:00:30 Colin Hillier 

Fan mail Tom so what's the latest on that? 

00:00:32 Tom Constable 

Not fan mail, it's our emails that have people contacting us with feedback and suggestions, which I'm sound like I'm being sarcastic. 

00:00:40 Tom Constable 

I'm not like it, it's hugely valuable and I really, really enjoy receiving them, and I think we do because it like we see every single week we wanna test and adjust and we want to change what we do to make sure it's optimised for people that are actually listening to. 

00:00:52 Tom Constable 

It so don't be backwards and coming forwards. 

00:00:53 Tom Constable 

So this one colour note. You've actually heard this feedback yet. 

00:00:56 Tom Constable 

So it. 

00:00:57 Tom Constable 

Says Tom, thanks very much. You're amazing. I wish you could. No joke. Doesn't say that it says, Tom says Tom. 

00:01:03 Tom Constable 

Thanks very much for being involved in the podcast, which I'm thoroughly enjoying on the robotics episode. I thought you missed a slight trick in not discussing and distinguishing between live dry training and live fire training. Quite a lot of this discussion centred around the importance of training. 

00:01:17 Tom Constable 

In the field, can't believe anyone would disagree, but I wonder whether there's a follow up discussion to be had about the relative benefits of live fire versus live dry training, particularly when advanced technological engagement simulations can be given data on accurate and effectiveness. 

00:01:33 Tom Constable 

What do? 

00:01:33 Tom Constable 

You think? 

00:01:33 Colin Hillier 

Yeah, it it's a. It's a huge topic. This almost looks at some of the work. I know that's going on with the US Army on convergence. 

00:01:41 Colin Hillier 

It's really deep. It might be more than one episode if anyone wants to come and talk about that, we'd be really happy to have you. 

00:01:47 Colin Hillier 

So yeah, please you know just on that Tom. I think one of the episodes went out last week and was. 

00:01:53 Colin Hillier 

Already some engagement on LinkedIn, so I think we don't care where the messages come, but it's really good to have people still asking. 

00:02:00 Colin Hillier 

Missions and connecting maybe with some of our guests as well so. 

00:02:03 Colin Hillier 

That's really good to see. 

00:02:04 Tom Constable 

Yeah, the last episode we promised, we'd update listener on our visit to the improbable office and actually more recently, you've you've been to itsec as well, so you've seen the improbable offering in the flesh. 

00:02:15 Tom Constable 

How do you wanna do this? I cover our visit to the office and then you let me. 

00:02:18 Tom Constable 

Know how it actually, kind of. 

00:02:19 Colin Hillier 

Yeah, can you give us a bit of a briefing on what we saw at the Shoreditch at the improbable offices? 

00:02:25 Tom Constable 

We turned up prompt and on time, as you'd expect and kindly shown through to the Commons room. And it was. 

00:02:31 Tom Constable 

I mean the offices are cool. 

00:02:32 Tom Constable 

But the main effort was there was to launch that partnership, and so there were a whole bunch of different companies there, or what was nice to see was the kind of the mindset of the people there. 

00:02:41 Tom Constable 

Not only the kind of energy enthusiasm of the people in the room, but I felt like there was a a really good safe space to discuss ideas. And despite having these two crusty podcasters come in and talk about what? 

00:02:52 Tom Constable 

Do we also got a chance to go and see the development of the tech demo in person, speech development team and that was exciting and good fun and also see the the way that they designed the partnership day and all the different breakout rooms and brainstorming sessions were happening. So that was good day and also very good nibbles at lunchtime. I have to say. 

00:03:11 Colin Hillier 

Yeah, that always changes the point of view. I did get a chance to see the live demo. 

00:03:16 Colin Hillier 

At its sick on the show floor in between running around. You know it always looks a bit shiny, a bit more fantastic out there. 

00:03:23 Colin Hillier 

The booth looked great. There was definitely lots of interest. There's definitely decent sized crowds looking at that and probably add that this is one of the challenges that has been bugging simulation since ever. I've been involved in it, so it's almost like a little bit of a Holy Grail. 

00:03:37 Colin Hillier 

For simulation, what I think probably trying to do, which is great to see. And you know Tamara your comments. 

00:03:42 Colin Hillier 

And definitely a collaborative atmosphere. You know they had a lot of the partners on their booth and some of the people I know quite well. 

00:03:48 Colin Hillier 

They were definitely interested in saying like how do we collectively solve the problem as opposed to hey, we've got all the answers which I'd like to think. 

00:03:56 Colin Hillier 

That's generally the. 

00:03:57 Colin Hillier 

Way we're going that more collaborative approach in in our market, but that's me just being optimistic but. 

00:04:02 Colin Hillier 

If that's the way forward, then more power to. 

00:04:04 Tom Constable 

Briefly, can you describe what their kind of demo was? 

00:04:08 Colin Hillier 

Yeah, so put the best way to talk and and I will sort of steal words over some of the improbable guys that briefed us, but it was about connecting disparate simulations together and then having it built in sort of modular fashion. 

00:04:20 Colin Hillier 

So if you want to do more work in the sort of C2 space or intelligence analysis, or to do sort of sentiment analysis. 

00:04:28 Colin Hillier 

Or you do you can. 

00:04:29 Colin Hillier 

There about sort of whole world simulation. You'd be able to almost have a menu of different simulation applications that would plug in, and you know with very little effort. 

00:04:38 Colin Hillier 

So, and this has come out, some of the recent army publications as well having systems that can be spun up relatively quickly. So in sort of days or weeks as opposed to years, which is what's traditional so. 

00:04:50 Colin Hillier 

As I say, bit of a Holy Grail, if we can genuinely spin up the applications as as required and the number of. 

00:04:55 Colin Hillier 

Applications as well. 

00:04:56 Colin Hillier 

Because they don't only want the depth and breadth and they want the quantity, so it's like we've gotta do all these things. So not a small challenge. Not something you can. 

00:05:05 Colin Hillier 

Do at a small scale and needs serious investment so it's really interesting to see. 

00:05:10 Tom Constable 

I wish I was out there, I do. 

00:05:12 Tom Constable 

I had serious. 

00:05:13 Tom Constable 

FOMO last week looking all the posts on LinkedIn. 

00:05:17 Tom Constable 

So I'm moving on to our interview this week. It is another Corker and I generally think you'll enjoy it. 

00:05:23 Tom Constable 

So without further ado, let me introduce you to Chris Covert, the executive producer for gaming, exercising, modelling and simulation at Microsoft. 

00:05:33 Tom Constable 

Chris Covert welcome. 

00:05:35 Chris Covert 

Ohhh thank you, thank you for having me. 

00:05:36 Tom Constable 

I know you're a busy guy and. 

00:05:37 Tom Constable 

So on I. 

00:05:38 Tom Constable 

Really, do we really do appreciate you taking the time today? 

00:05:41 Tom Constable 

This interview is the first time where we are the warfighter. Podcast are united in location. So hopefully the sound quality will good enough. 

00:05:49 Tom Constable 

I hope you. 

00:05:49 Chris Covert 

Can hear us clear enough it's an honour to be a first here. Hopefully my New York based Wi-Fi is good enough to hold the call. 

00:05:55 Tom Constable 

Can't be any worse than my forest of Deans. Let's not get into that. Otherwise I'll start crying and throwing toys out the pram. 

00:06:01 Tom Constable 

So with tradition Chris, please introduce yourself and let us know why we should listen to what you have to say. 

00:06:06 Chris Covert 

Oh man, let's let's try. I'll do my best. Chris covert. Been in the modelling and SIM community for only about a decade younger in the industry, but currently at Microsoft doing a lot of our work behind modelling and simulation in an industry. We're calling gems, gaming, exercise modelling and SIM. After the US Defence Science Board pilot publication of the same name. 

00:06:26 Chris Covert 

Background in deep reinforcement learning and autonomy been doing SIM. 

00:06:29 Chris Covert 

For as long as I've been interested in computer science, so it's an honour to be doing it here. And oh man, do I have a lot of opinions? 

00:06:36 Chris Covert 

And and I'm glad we get to talk about it today. 

00:06:37 Colin Hillier 

We hear AI a lot. It's a very common phrase used, and probably, I suspect, by people that don't fully understand it. So let's just pretend we don't understand it. 

00:06:47 Tom Constable 

I don't, I I before we start I I'm gonna be the first. 

00:06:50 Tom Constable 

I don't understand. 

00:06:50 Colin Hillier 

We're gonna pretend. 

00:06:51 Tom Constable 

With it, you know we asked you for something you were passionate about and instantly you jumped to AI. You almost bit our hand off to say look AI is an area that I wanna. 

00:06:58 Tom Constable 

Wax lyrical about so we'll look forward to this. 

00:07:00 Colin Hillier 

I think the first area of interest is we're the sort of examples that you've come across where you used in anger AI and give us some of the examples where it's worked for you. 

00:07:10 Chris Covert 

Oh man, let's see to start before we even jump into example. 

00:07:14 Chris Covert 

I think we need to talk about how there's two different camps of AI when we say AI. Unfortunately, half the room thinks one thing and half thinks the other. 

00:07:23 Chris Covert 

Not uncommon with terms that are catch alls or generalisations for a tonne of different opportunities or a tonne of different applications. Bases for it I would say when we talk about game, AI or AI and. 

00:07:34 Chris Covert 

Modelling and simulation. 

00:07:36 Chris Covert 

It's when you think of traditional video game artificial intelligence versus artificial intelligence and machine learning, which is backed on either a data structure like you would have in a machine learning architecture or in a reinforcement learning architecture. They are two very different camps and have two different outcomes. Unfortunately, just by happenstance of our language. 

00:07:56 Chris Covert 

They have shared the same term and therefore get interchange. 

00:08:00 Chris Covert 

Changed a lot, so before we even jump into use cases talking about the value of going on the game AI route versus the value of using something more attuned to a reinforcement learning model has huge implications and ramifications. 

00:08:13 Colin Hillier 

Chris, I did a little. I mean I. I start from zero as well, but I did do a bit of background reading on this. Yeah, just a Wikipedia reading but but you know, under that title of A. 

00:08:24 Colin Hillier 

Guy, there's a whole swathe of technologies and techniques. I mean including things like computer vision, you know, looking at sort of decision trees, natural language processing, all of which are actually very different. 

00:08:36 Chris Covert 

Very different fortunately, and unfortunately the magic of AI as a catch all is that you can use one term and mean just about anything so. 

00:08:45 Chris Covert 

When you have a hard problem that needs to be done, you can probably put AI to it and to a lot of people. 

00:08:49 Chris Covert 

That's enough, it sets you on the right path forward. Unfortunately for the people that love AI, it means that there are a billion different architectures. 

00:08:56 Chris Covert 

You have to sift through before you find the one of value that hits that right requirement. You're spot on and we can go into machine learning. We can go into the difference between. 

00:09:06 Chris Covert 

Supervised versus unsupervised learning. The value of game AI in a lot of modelling and simulation I'll make. 

00:09:12 Chris Covert 

The I'll make the bold claims you're on the record. This is first of all to say the opinion of me myself. 

00:09:18 Chris Covert 

Chris Covert not the opinion of the company that I work for here. I would say a lot of modelling and simulation utilises game AI more than machine learning because you don't need true robust machine learning applications. 

00:09:29 Chris Covert 

For a lot. 

00:09:30 Chris Covert 

Of what we consider you know non playable character. 

00:09:33 Chris Covert 

Actions the benefits of game AI genuinely. There are a tonne of them because the value add of using an artificial intelligence system. 

00:09:42 Chris Covert 

In a modsim experience, is that you're prescribing actions that are meant to feel realistic to a non playable character. 

00:09:49 Chris Covert 

In order to do that, you don't need full machine learning or deep reinforcement learning applications. Now you can get better fidelity out of those sometimes, but there are trade-offs, and in a modsim environment it's all about trade-offs we know. 

00:10:01 Chris Covert 

That it's our industry. 

00:10:02 Chris Covert 

The value of game AI is it doesn't have to be pretty. 

00:10:04 Chris Covert 

Trained and in most cases when you're developing this AI, it's in environments where it really can't be because you are developing in an unfinished environment, you have very limited players. 

00:10:14 Chris Covert 

You have limited play throughs. You don't have a data set of all the possible actions or a headless version of this that you can run as a Monte Carlo and collect all of the optimised decisions that can be made. You're doing this in a very simple. 

00:10:26 Chris Covert 

Constrained environment so constrained game AI is more than OK. 

00:10:30 Tom Constable 

So why, why haven't? 

00:10:31 Tom Constable 

I yet come across a game or a simulation defence simulation where I am playing against an AI or training against an AI who are doing things realistically, doctrinally well, but also realistically, so if I suppress any. 

00:10:45 Tom Constable 

They're gonna behave like they're being suppressed. I haven't yet personally trained against an AI where I felt that if I do the realistic things well, I'm gonna get the outcome about this in this training scenario that I want. 

00:10:56 Tom Constable 

Invariably, you know I end up gamifying it, and you know if I shoot a window, I know that I'm gonna attract that soldier to the window and therefore I can shoot them in. 

00:11:04 Tom Constable 

Head, that's that's obviously unrealistic. If I was shooting a real you know window frame, that soldier's gonna be anywhere other than that window frame. And then that's a different thought process, so why haven't I yet come across that? 

00:11:15 Chris Covert 

It is a very good question. I would say it to over generalise it. It's because the state space of these free actions is just too large to capture it. It comes down to constraint realism so. 

00:11:29 Chris Covert 

You can think of I and I love the Alpha series of AI right Alpha. Go when you talk to some of these players or you look at their interviews on playing these super optimised machine learning or deep reinforcement learning agents that are playing well above the best of the best in their field. It feels like they're playing an alien because they're making human decisions. 

00:11:49 Chris Covert 

That don't feel human right? They're making optimised decisions, not realistic ones. 

00:11:53 Chris Covert 

And it's difficult when you the more advanced the machine learning model can get, the more you're throwing information into a black box for a lack of better terms, it's hard to prescribe what realistic looks like without overfitting that data set to things that you think might be realistic cause it's going to look for optimum. It's a mathematical model, it's looking for Max and min. The nice thing about. 

00:12:14 Chris Covert 

Game AI is that there are ways that you can throw those constraints and actually prescribe some of those transitions to make it feel more realistic without having to dive into a super deep machine learning architecture. It's again recognising. 

00:12:29 Chris Covert 

What that ideal state space looks like? What those ideal transitions look like and then creating an architecture that actually follows suit through those desired actions. 

00:12:38 Colin Hillier 

And I'm guessing one of the things that may force us down this sort of game AI or decision Tree route is for machine learning. 

00:12:45 Colin Hillier 

You have to have enough data and probably enough good data in a format that you can train the machine. 

00:12:51 Colin Hillier 

Is that a correct? 

00:12:51 Chris Covert 

Statement depends on the model. There are again at a very high level here. There's kind of two approaches to a machine learning model, let's say A. 

00:13:00 Chris Covert 

Supervised or unsupervised structure to it. And it's all about how that data is structured now this isn't considering reinforcement learning, which we'll get to for sure, but if you're going with a database and you wanna train a model off of that data, a supervised model says I understand what my data is through its label, right? 

00:13:16 Chris Covert 

I'm going to use a computer vision analogy here. If I have a data set with a whole bunch of dogs. 

00:13:20 Chris Covert 

And a whole bunch of cats in it. A supervised model says I have a data set that is labelled for each image of a cat that it is a cat. My data is self aware that it. 

00:13:29 Chris Covert 

As a cat then I trained that in the model and I can use that to differentiate cats versus not cats, right? 

00:13:35 Chris Covert 

These are used for things like classification or regression. An unsupervised model says I'm going to throw all of my data together and not differentiate between what is a cat or what isn't a cat in the data itself it's unlabeled, so I can then use clustering to determine this is group. 

00:13:50 Chris Covert 

A and this is Group B. Now a user would recognise that as group cat and group dog. It also is good for anomaly detection. 

00:13:57 Chris Covert 

If I throw a lizard into the mix, you'll have group cat, group dog and group lizard. You also might not. Right. These things aren't perfect. You might have group, cat, group dog and the lizard is. 

00:14:06 Chris Covert 

Group dog, but it is good for anomaly detection. Depending on how you format that data set, that becomes the more data. 

00:14:13 Chris Covert 

You have a very arduous and manual process in a lot of ways. The benefit of something like reinforcement learning. 

00:14:19 Chris Covert 

And there are a whole bunch of different reinforcement learning on or off policy models. You can run these without data sets and it will look to find. 

00:14:27 Chris Covert 

Optimal behaviours like. 

00:14:30 Chris Covert 

Sarsa learning is one of my favourites because the way you explain the acronym is actually just the name of the acronym. 

00:14:35 Chris Covert 

Sarsa stands for state action reward state action. So let's think of Mario right. Mario has a state he's standing on this 2D plane. 

00:14:43 Chris Covert 

We're going classic Mario. None of the crazy super fun 3D Marios. Only way to go we're jumping. 

00:14:47 Tom Constable 

Only way to go mate. 

00:14:49 Chris Covert 

Then nearly onto some Goombas. In this analogy, no rotation necessary state is. I know where I am in relation to my environment. My action is maybe I'm gonna move right? I'm gonna move left I'm gonna jump or I'm gonna. 

00:15:01 Chris Covert 

What my reward is, depending on what that action did, how valuable that reward was in the grand scheme of what I view as my overall weighting for. 

00:15:10 Chris Covert 

Personally, how I have my reward system set up so if I'm playing Mario I'm going to reward moving right and deprioritize moving left because it's a side scrolling game. If you go left you don't make it. 

00:15:22 Chris Covert 

If I get hit by a goomba, that's a negative reward. If I jump up and I smash it. 

00:15:27 Chris Covert 

Darrell, that's a positive reward. If I jump up and I smash a block. Maybe that's more of a neutral reward. 

00:15:32 Chris Covert 

If I get a mushroom great reward, so depending on that action you assigned a reward value to it and then you go through your state action pair again for your next state action and the Sarsa model is running through all of these different States and actions and gathering a total score based on a run, it knows what. 

00:15:49 Chris Covert 

Actions led to what rewards and it's therefore going to try and optimise through a series of runs that it's going to do the best, most optimal way to run through that environment, but. 

00:15:59 Chris Covert 

That's the key here. 

00:16:00 Chris Covert 

And I think that's the thing that the modsim community for reinforcement learning models and deep reinforcement as well is. 

00:16:06 Chris Covert 

It requires environmental knowledge, so it's difficult. It's not impossible, but it is. It's not turnkey to take a model trained in one environment and adapt it to a new environment. And as modsim enthusiast and maybe lifers on the call. 

00:16:20 Chris Covert 

Here today we understand just how difficult it is to model these operational environments. In particular, as the foundation of this, let alone use those to start developing agent behaviours. It's normally you want it the other way around. You're developing these behaviours and you're creating them. 

00:16:35 Chris Covert 

Environment to have a fully fledged environment that now your agents are trained to means they're not as modular, which could actually limit the value of that modsim environment, where again game AI doesn't have all of those constraints. It makes it a lot easier to be more plug and play with your logic. 

00:16:50 Colin Hillier 

So just going back to Tom's original question, where do you think the problem is in terms of developing that realistic AI? Something that feels you know it is a bad? 

00:17:00 Colin Hillier 

Dot trying be correct. What do you think's missing? Is it data or is it time is it? Is it sort of complexity of the model? 

00:17:05 Chris Covert 

What what are we missing here? I'll? I'll say all of the above and I'll throw it under another catch all term, just called scalability. 

00:17:11 Chris Covert 

I think if you look at some of the examples of using reinforcement learning well in games full stop, I'd love to talk about Forza. 

00:17:21 Chris Covert 

The sports line of racing games. You have the more SIM side of the house and the more ARCADY side and horizon right motorsport and horizon are two different styles of racing games. 

00:17:30 Chris Covert 

One of them is more of a simulator, one of them more an arcade driving game, both of them popular Xbox games played by millions of people. 

00:17:38 Chris Covert 

The story I love to touch back to and I wasn't a part of this team. I would have loved to have been. I was probably in like middle school though. My favourite thing they did was. 

00:17:45 Chris Covert 

Between their shift and more on Prem cloud sources and their Azure cloud source, Xbox 360 to Xbox. 

00:17:52 Chris Covert 

That's one fact. Cheque me Wikipedia fact checkers if you're if you're listening to this, they had all of their drivers now being able to collect all of this data in a hyperscale cloud environment, right? 

00:18:02 Chris Covert 

You went from an on Prem data centre to a cloud based environment. You had millions of users running all of these laps and they're just sitting on all of this massive data. 

00:18:10 Chris Covert 

And you're trying to drive against AI competitors there as well, right? You're in your grid of 12 to 16. 

00:18:16 Chris Covert 

And as soon as the race starts, if you're playing single player in particular, you start driving and you have 11 other people driving right next to you that need to feel realistic, or you immediately break your immersion. So what they did was they actually ran. 

00:18:29 Chris Covert 

A whole bunch of reinforcement learning models to overhaul how they do predictive braking, line following and driver aggression. So this is going to be when you come out of a draught how quickly you break in and out of a turn when you actually take tight turns or wide turns, they ran the reinforcement learning model to make the computer feel more realistic in those three domains. 

00:18:50 Chris Covert 

Now the state space can be constrained to those 3 domains because those are really the most important that separate amateur from elite drivers, right? I am a really really, really bad go. 

00:19:00 Chris Covert 

Cart driver because I don't know when to break into a turn and I think I'm good at the other stuff. 

00:19:04 Chris Covert 

Maybe I'm a little aggressive as well, but genuinely breaking around a turn in line following is the difference it that shaves seconds off of each lap so to be able to run these models was interesting because you could create a general model and then do some. You know random number generation some slight. 

00:19:20 Chris Covert 

Perturbations of that model to make it feel more randomised to make it feel like each agent was operating on a different behaviour model, but they didn't, they actually took it down to the individual level they said. 

00:19:29 Chris Covert 

If I race a couple of laps on these tracks it will create a model of how I do preventative braking and line following and driver aggression so that if I go offline you guys can race against the version that feels like me even though I'm not there. 

00:19:42 Chris Covert 

That is to me a microcosm. It is a. It is an example of what is to come as we start to scale out the. 

00:19:49 Chris Covert 

Ability to run these models with much larger state spaces. Those are three parameters that they train these models on, and even that feels herculean sometimes. When you think about what has to go into that data, they also had. 

00:20:00 Chris Covert 

Millions of drivers rising all of these laps. So yes, I think we'll get to a point where you can have doctrinal agents running on a reinforcement learning model, but you have to think of just how complex each action has its dependencies. That state space grows so quickly, so fast. 

00:20:16 Tom Constable 

Probably a stupid question. Let's take Call of Duty or other games are out there. That and proper more Milsim armour squad, whatever it might be. They're all being played online and there are, you know, thousands of millions of engagements happening online constantly. 

00:20:33 Tom Constable 

And individuals must react in certain ways to the proximity of a certain explosion around landing near them being shot. 

00:20:40 Tom Constable 

Is there something to learn from the games industry and and a way that civilians fight could? Could there be something that we could draw out of that data to help us either create better AI or for even maybe? 

00:20:53 Tom Constable 

Great new military tactics. If we see a reoccurring pattern that actually if you I don't know, throw a flash bang and an HG grenade in a certain order and then running through the wall it through the window. 

00:21:04 Tom Constable 

Then we we've learned that 95% of the engagements and positively is something that we can pull. 

00:21:09 Tom Constable 

In the future, can pull out from that. 

00:21:11 Chris Covert 

I I hope so. I think that that is a natural transition from where we are now. I actually have seen it done in a couple of different ways. 

00:21:19 Chris Covert 

Unfortunately, none that I can openly talk about here, but hopefully that we'll all get to see in the next coming years. There is an art to using reinforcement learning models on like legacy footage. 

00:21:31 Chris Covert 

Or gameplays or play throughs or training footage to actually better understand what that state space looks like, those transition points. So if a then B. 

00:21:41 Chris Covert 

If you manually were to code that entire state space right, like in a finite state machine or in a hierarchical finite state machine, then you can run a decision tree through all of those actions. 

00:21:51 Chris Covert 

You could do it in real time. You could feel realistic, but you would be constrained by what you identified as a state and a transition out of that state. Your triggers running a reinforcement learning model to identify those triggers. 

00:22:01 Chris Covert 

And then putting that in a state space that you can run in real time is, to me, a really good blend of the two different type. 

00:22:08 Chris Covert 

Types of when we say AI had them coming together in harmony using reinforcement learning to say these are the possible reactions to that state, I'll go back to the Mario when I get close to a goomba. 

00:22:19 Chris Covert 

I need to jump. That is an overly simplified example, but in the modsim environment there are a lot of those I need to jump at a gumbo moment so it's grenade through the window. 

00:22:29 Chris Covert 

I duck to the left or you know, I back out of the room, not I look at the grenade for five seconds like most players might. 

00:22:36 Chris Covert 

Running that figuring out what is optimal and then using that to code into a a state space to me is a perfect blend and is probably a a good patch until we get to a full end to end reinforcement learning model in real time. 

00:22:48 Colin Hillier 

I mean, there's there's maybe an interesting parallel in development of self-driving or autonomous vehicles where there's a number of different organisations building synthetic environments purely to train these systems. 

00:23:00 Colin Hillier 

And yet the concept of the Dojo and the Dojo is essentially having to run on Super Compute architecture because that's the sort of scale. 

00:23:08 Colin Hillier 

It's a bit like any other model you know. We build economic models. The micro simulation models for. 

00:23:13 Colin Hillier 

Shifting policy outcomes in government. They run on super compute, but I don't think we're doing that on the sort of training and military side yet. 

00:23:22 Colin Hillier 

For that, maybe again, maybe there's some that is and and it's not not within our our other space. But is that where we need to get to developing that Dojo? That where we can train before we even start training to have the Dojo to train the. 

00:23:35 Chris Covert 

It is an awesome question because you think of some of those dojos exist already in the commercial space. You think of things like open AI. Gym is a hugely popular. 

00:23:44 Chris Covert 

There are these places where models can be trained surely, but again, the better the model, the more it's being trained in that operational environment. 

00:23:51 Chris Covert 

So you would need for the best models today to be training it in the place you want to deploy it, which is sometimes infeasible, sometimes cost and tractable, sometimes just based on timing, not possible. I think that. 

00:24:04 Chris Covert 

As these especially 3. 

00:24:06 Chris Covert 

D and I'll say, you know, 3D geospatial georeferenced environments become more mod SIM standard. We'll have a better global, I'll say like AI. 

00:24:16 Chris Covert 

Jim Dojo 2.0 to move forward too. But right now if the industry's still a little sparse on adopting that as an open standard for how we build out these Sims. 

00:24:26 Chris Covert 

I think there is still a little bit of of ambiguity around what these environments look like and feel like to have a common model system, a model Dojo, to help train a lot of these models. It's still a lot of application specific training. 

00:24:40 Chris Covert 

I definitely see a future in in combining that. 

00:24:43 Colin Hillier 

Though, well, what's really interesting is you know historically, geodata has been under some classification. 

00:24:48 Colin Hillier 

What we're seeing is more and more the high resolution data is actually just commercially available. So what's missing is things like the kinetic data, the ammunition data and effects, and that's what we probably might never have. 

00:25:00 Colin Hillier 

But I guess it's sort of the military to combine that the the publicly available data in high definition with that effects data which only they have and pull that together to create that environment where we can train our I and, you know, discover you know. Going back to Tom's point, how do you discover new approaches that you never thought of new? 

00:25:19 Chris Covert 

New techniques the best part about AI is just how quickly it grows. I left my masters programme, you know. Not too many moons ago and it's almost unrecognisable from where I left it. 

00:25:31 Chris Covert 

Each individual field of anything you could classify as machine learning, deep reinforcement learning has evolved year over year at like record. 

00:25:39 Chris Covert 

Breaking pace, I am generating images using diffusion models. 

00:25:44 Chris Covert 

Putting on, you know, all of the fun image generation tools through AI right now that I could possibly get my hands on because genuinely hour by hour, the industry is reinventing itself. 

00:25:53 Chris Covert 

The open source support the commercial off the shelf availability and the government off the shelf. In our case, applications for a lot of the tools we're using are exponentially growing year over year. 

00:26:04 Chris Covert 

When I started doing Sims about a decade ago. 

00:26:07 Chris Covert 

Geospatial was not anything new, but from a decade ago to now I almost have. I'm spoiled for choice on how much geospatial data sources I can pull in open, how easy it is to work with partners that are geospatial providers, both on Prem and in the cloud. 

00:26:22 Chris Covert 

It's great, I think the Modsim community is going there. We're vectoring in a very good direction for building these more open. 

00:26:29 Chris Covert 

That we can train generalised models on and man I wait until a year from now where this comment seems Super Archaic in that we only had spoiled for choice in geospatial. 

00:26:39 Chris Covert 

It's a matter of time before we hit that and modsim across the community the way. 

00:26:42 Tom Constable 

We're growing now. Stupid question. #3 and I promise this may or may not be the last one. Where's AI going to go where? 

00:26:49 Tom Constable 

Could it end up? Is there Skynet or whatever it's called, you know is it is AI gonna take over the world or actually is that just something that's made-up for films and actually AI is? 

00:27:00 Tom Constable 

Not as clever as we like to lead ourselves to believe, and we're nowhere near this sentient being that starts, you know, work out ways to end humankind. 

00:27:09 Chris Covert 

Oh man, I think pessimistic. If anyone listening to this knows me, they'll know that I'm I'm pragmatic. I think human use of AI is going to lead to our demise well before any general artificial intelligence or or general intelligence model. I think personally that the state of AI will see in the not too distant future is just. 

00:27:29 Chris Covert 

More ubiquity, right? We use AI for so many things. 

00:27:32 Chris Covert 

That don't feel like they're AI right now. They feel incredibly natural and a natural part of our lives. I love to see the adoption of things like digital assistants. 

00:27:42 Chris Covert 

I don't know a person who doesn't use a digital assistant like it's something that's always been around. Watching my own family, especially ones who are older. They don't know how to send a text message, but they can talk to their digital assistants. 

00:27:53 Chris Covert 

Like they're they're neighbours, they're they're roommates. They go way back, I think through things like transcription and translation. We've seen AI just become adopted by society. I'm super excited for a time where in experiential AI, meaning, like game, AI or or action behaviour. 

00:28:10 Chris Covert 

That's things just feel incredibly natural everywhere. 

00:28:13 Colin Hillier 

Probably worth just sort of winding one second back because you you know you. 

00:28:16 Colin Hillier 

You mentioned there's things that we're using everyday and they're good because we don't realise that AI and and use one example, and that's things like Google Translate and actually, Google Translate isn't just a rules based process, but it's actually developed from. 

00:28:30 Colin Hillier 

Analysis, You know, artificial intelligence and machine learning, analysis of lots and lots of use of common language, and I use every day and it's it's quite interesting because you can tell what language. 

00:28:40 Colin Hillier 

That the AI has been looking at by its responses is fascinating. 

00:28:45 Chris Covert 

Again, how models are trained. Google has been doing AI for a very, very long time. I have a couple of really good friends at Google. 

00:28:51 Chris Covert 

They will eat, sleep, breathe AI. I imagine until the you know the end. 

00:28:55 Chris Covert 

Of their careers. 

00:28:56 Chris Covert 

The best part of AI is when you can smuggle in the innovation right. Overly forward AI. Again, the reason we're even having this conversation is because AI. 

00:29:04 Chris Covert 

Is a very industry hot term. Programmes are sold on their ability to leverage AI and that's it. Leveraging AI period is in a requirement in some docs and you hate to see that as the AI guy. 

00:29:14 Chris Covert 

So you you look at that and you go. Oh man, we have a lot of work to do. We have a lot of education when you can make a I feel like it's not a I like it's just a very simple computer programme. 

00:29:24 Chris Covert 

You've hit a really good sweet spot. Now that to me can be done at smaller scales because models can be shrunk and put into mobile experiences for large scale mod SIM, we're still at that point. 

00:29:35 Chris Covert 

Especially when you're running these in game engines where you have to adjudicate these decisions super quickly so it's just not there at scale yet you you feel when you're running a deep reinforcement learning model in a mod SIM. 

00:29:46 Chris Covert 

Because your frames start to crawl, but that won't be the case forever. And when we start to hit that threshold, which I think we're actually at now, you will start to see advancement in these models. 

00:29:56 Chris Covert 

Probably blow us all away, right? We'll look back at a time where this wasn't industry standard and be like how did we even do this? How did Modsim survive for decades and? 

00:30:04 Chris Covert 

Decades and decades. 

00:30:04 Colin Hillier 

I think you're right. You know I have that feeling as well where we look at this and I I feel where we're sort of slightly looking bleary eyed at this and trying to get our heads around AI and and I think. 

00:30:13 Colin Hillier 

This years for those our listeners that are embarking on these projects and looking at how they should start with AI. 

00:30:20 Colin Hillier 

What's your sort of top 345 or whatever you know tips to someone that's never used it before? 

00:30:26 Colin Hillier 

Well, the obvious mistakes people make that you would recommend they be aware. 

00:30:29 Chris Covert 

Of I would say #1 is just understand the taxonomy. Understand what's out there and the the different applications of AI. 

00:30:36 Chris Covert 

Don't start with theory unless you have a super mathematical background and that's all you care about. Start with an application that interests you. 

00:30:42 Chris Covert 

For me, it's Gans, generative adversarial networks. Gans are super fascinating to me. I love working with them. It is what gets me inspired to start identifying models that are open that I might want to use for my own purposes, like upsampling or, you know, super resolution models taking old photos from the the 80s, the 90s, the 60s, the 70s for my parents. 

00:31:04 Chris Covert 

That ages me for the colours on the the line here. Sorry guys and running them through a super resolution to make them feel like modern photos. 

00:31:10 Chris Covert 

Super fun and then from there you can start to tinker with the model. You can start to maybe train your own versions of these models if you have the right capabilities and you can really get in from the application side and not really have to care about the theory until you really want to make some severe modifications of it. 

00:31:25 Chris Covert 

I love that because not every model and not every application is for everybody. There are parts of AI that. 

00:31:30 Chris Covert 

To me are. 

00:31:31 Chris Covert 

We're boring and I probably will not touch them because there are alternatives to implementing those models that I prefer in other forms on the game AI side. 

00:31:39 Chris Covert 

I think there's also a way to not have to worry about artificial intelligence slash machine learning, which will just bin that as the ML side of the house and do the more decision tree side. 

00:31:49 Chris Covert 

I think there are really accessible ways to get involved in those communities as well. There are tonnes of game. 

00:31:53 Chris Covert 

And opportunities or providers out there that you can just jump right in and start developing today. Massive communities to help you just get off. 

00:32:00 Chris Covert 

To ground with I need an environment with an agent and an adversary and I just want to start having that adversary make decisions. 

00:32:06 Chris Covert 

One of the things right now it's not a new topic that I love to talk about with people interested in game AI is this thing called Goal, Goal oriented Action Planning. Because it's a really fun way of visualising how humans make decisions. 

00:32:20 Chris Covert 

And it feels super fun and modular. It has its own drawbacks like everything but Gope is a great way to start if you're looking into game AI. 

00:32:26 Colin Hillier 

So to summarise, I think what you're saying is dig a little, understand the text. 

00:32:29 Colin Hillier 

Economy and use the right tool for the job before you dive in and go. Hey this is AI cause not. 

00:32:35 Colin Hillier 

Not all AI is is the. 

00:32:37 Chris Covert 

Same, not all is created equal. They're very different outcomes as well. If you really want to get into AI and you drive really really deep and you dive deeply into one field at the end of that tunnel might not be the thing that you thought it was, or the thing that is relevant to what you want to do with it. 

00:32:53 Chris Covert 

So just understand the landscape. It's such a big landscape. The value of deep reinforcement learning versus reinforcement learning. The value of neural. 

00:33:00 Chris Covert 

That's and all of the different flavours of them. What they can do in conjunction with other models. A lot of AI is done in pipelines, so just start to understand at a high level what the different applications are and you can start to piece together your own pipeline of what you want it to do. And there's a community to support. 

00:33:15 Chris Covert 

You, throughout the process, very polite way of. 

00:33:17 Colin Hillier 

Saying get educated. 

00:33:18 Tom Constable 

That's like hard work that does. 

00:33:19 Colin Hillier 

Sounds like a lot. 

00:33:20 Tom Constable 

My last dumb question and then we can wrap up when does a line of code which is an if then else statement or whatever it might be. When does something like that? 

00:33:29 Tom Constable 

Switch to becoming AI, because if I'm creating a product it's a smart wearable and it can tell my heart rate and I want to know if there's a been a rip in the mesh and my heart wearable and the heart rate increases or decreases. 

00:33:43 Tom Constable 

And then I wanna send a message back to the OPS room to say that there's a potential risk of injury is. 

00:33:48 Tom Constable 

That AI or is that just? 

00:33:50 Tom Constable 

And if they're not. 

00:33:51 Tom Constable 

Process that a normal algorithm is written and where does it the line. 

00:33:55 Chris Covert 

Merge, it's the ultimate question and the reason we're talking today, right? 

00:33:59 Chris Covert 

To some people that say I linear regression is AI to people with a background in mathematical AI ML, they'll spit on the floor and go. Hey, that's that is not AI, that's linear regression. I don't think that matters, I think. 

00:34:11 Chris Covert 

Analytics there is no fine line between what is linear regression versus a decision tree versus a reinforcement learning model. It's the nuance in driving down is going to be application specific in everything. 

00:34:24 Tom Constable 

So I can say that I'm an API developer when I'm when I'm creating these you are now I am now and I I'm choosing to call myself out. I'm gonna put on my LinkedIn profile. I'm a business card, I'm pretty. 

00:34:34 Tom Constable 

Business card now that's. 

00:34:35 

Perfect. 

00:34:35 Tom Constable 

That's thanks so much that I think this is the kind of thing we can have. Well, we should, and we'll have numerous episodes on. 

00:34:41 Tom Constable 

I'm picking probably going starting to jump into that and go game AI right? This? Let's do an episode on. 

00:34:47 Tom Constable 

And talk about that in more detail and hopefully of our understanding, and hopefully therefore the listeners understanding. So thank you so much for your time and your patience with me. 

00:34:57 Tom Constable 

More than Colin I. 

00:34:58 Chris Covert 

Think of course I. I look forward to listening to all of the people coming to debunk everything. I just. 

00:35:03 Chris Covert 

Said line by line. 

00:35:04 Colin Hillier 

The thing is. 

00:35:05 

Gonna happen so. 

00:35:05 Tom Constable 

Normally this stage, is there anything else you'd like to add to this conversation before we wrap up? 

00:35:09 Chris Covert 

Oh, that's a good question. If I had to add anything, I would say that this is definitely an unfinished topic. 

00:35:14 Chris Covert 

There's so much more we could talk about AI. I'd look forward to listening to everyone you can bring on to talk about the subject. 

00:35:20 Chris Covert 

Both game AI and AI ML if you want to talk more with me in particular, you can always find me on LinkedIn. I'm pretty active outside of actual posts. I have no other social media, so LinkedIn. 

00:35:29 Chris Covert 

Christopher Covert, you can find me there. Look forward to the chat and I wish you the best luck on the podcast. You have another avid listener. 

00:35:36 Colin Hillier 

Well, thank you. I'll be adding you and joining in your conversation. 

00:35:41 Tom Constable 

Yes, I love it when you know an interview just feels like there's flow and his excitement and his engagement throughout. 

00:35:48 Tom Constable 

So thank you again, Chris. Your ability to take complex concepts. Create a simplistic analogy or scenario, or relate it back to something that everyone can understand and then kind of grow from that was. 

00:35:59 Tom Constable 

Pleasure to listen to so thank you. 

00:36:01 Colin Hillier 

Especially as I think AI is one of those things that people refer. 

00:36:05 Colin Hillier 

To often and. 

00:36:06 Colin Hillier 

Really don't aren't very specific about what flavour they're talking about. 

00:36:11 Colin Hillier 

You know it's a tool set, isn't it? It's about having the right tool for. 

00:36:14 Colin Hillier 

The right job and. 

00:36:15 Colin Hillier 

Just go, oh we'll just fix that with AI doesn't really help you. 

00:36:18 Tom Constable 

Yeah, and I hope that people that maybe aren't building anything to do with AI. But you know when the next person pitches back to them that they're gonna use AI to solve the problem. 

00:36:26 Tom Constable 

Then at least they'll know where to start in terms of asking questions, which hopefully is one of the values. 

00:36:31 Tom Constable 

Of this podcast. 

00:36:32 Colin Hillier 

Yeah so onwards to our team. Journalist Andy Fawkes, who's gonna give us a bit of a an update on the latest happenings in simulation and training world. 

00:36:42 Speaker 3 

Hi Colin hi Tom. 

00:36:43 Speaker 3 

Really good, really good to be back. 

00:36:45 Colin Hillier 

The first story we've got is very topical one because I am in Florida, tending a little show called I hit second. 

00:36:51 Colin Hillier 

Don't know if you've heard of it, and if you tell us a bit more. 

00:36:53 Colin Hillier 

About it second, what's going on? 

00:36:55 Speaker 3 

I have heard of it. I went 15 years in a row and we perhaps we could talk about that later, and I think we've explained to our listeners before. It's the interservice stroke industry training simulation. 

00:37:06 Speaker 3 

Education conference and in fact it's been running in one shape or form since 1966, which is a long time. It started as a naval event go Navy, but over the years became into service. 

00:37:16 Speaker 3 

Although it's to do with training systems, it's now as you've said, it includes simulation education as well. This year I believe it's 17,000 people will be attending. According to the article in MSNBC. 

00:37:28 Colin Hillier 

Goes down there picking my badge up estate. And yeah, you definitely get your foot miles in your pedometer, miles in around that place and looks pretty well attended. You know, after the goings on of the past few years it's nice to be. 

00:37:41 Colin Hillier 

There in person, but I think Andy you reflecting on one thing that you felt might be missing from these sort of events. 

00:37:47 Speaker 3 

If you want to jump straight into that, I'm professionally disappointed that there is seemingly no hybrid aspect to this, although I guess some of the videos will be made available afterwards. 

00:37:56 Speaker 3 

I do feel that as I said, I went for 15 years in a row and in those days it's not that long ago, but it's obviously pre pandemic. 

00:38:04 Speaker 3 

And there was no alternative, and I think now everyone does know there's an alternative and that's doing things by hybrid. 

00:38:11 Speaker 3 

So sure, if people want to meet in person, they can. But if you can't for whatever reason can't afford it, you're unable. 

00:38:16 Speaker 3 

And I think personally there should be a hybrid options to these events, particularly when we're talking about Metaverse talking about distributed training. 

00:38:24 Speaker 3 

That given the. 

00:38:25 Colin Hillier 

And and, and there's an interesting article in in Ms and T about what the themes are. Maybe you can. 

00:38:31 Speaker 3 

Yeah, there's there's an article in MSNT that provides a sort of good context to the whole event. Marty Carjack interviewed the NHTSA president as Admiral Jim Rock retired. 

00:38:41 Speaker 3 

That's the national Training Simulation Association in the US, so that was really good to hear from Admiral Rob. I think he put a very interesting context to this about the challenges. 

00:38:51 Speaker 3 

Obviously, he reflected on you know what's going on in Ukraine, and we've covered that before. That's obviously part of the context, but he he was talking about the effect of inflation on. 

00:39:00 Speaker 3 

The technology things becoming more expensive. He talked about difficulties in the supply chain I, I guess cutting off supplies to maybe certain Chinese or maybe even Russian technology, maybe causing supply chain issues. 

00:39:12 Speaker 3 

And last but not least, it's something for people to talk about for years is the challenge of building capabilities in what he called. 

00:39:20 Speaker 3 

A timely manner, but that was really some good good context to the event that's going on this week in Orlando. I don't think you said it's. 

00:39:27 Speaker 3 

In it's actually. 

00:39:27 Speaker 3 

In Orlando, yeah. 

00:39:28 Colin Hillier 

So it's in all it'll. 

00:39:29 Colin Hillier 

So yes, I didn't. I didn't go that far. 

00:39:30 Tom Constable 

Yeah, yeah, I just want to go back to your point about the hybrid event. Yeah, I get your feeling. 

00:39:35 Tom Constable 

One thing I wanted to mention because I am I define myself as a reluctant eco warrior, is that I think there's a very big argument for that now as well. 

00:39:42 Tom Constable 

And I think we've gotta make sure that we travel because you should, and there's a need to. You gotta get out. 

00:39:46 Tom Constable 

They gotta see the. 

00:39:46 Tom Constable 

Kit that you gotta hold it, you gotta touch it. But if there is not that need to touch it. 

00:39:50 Tom Constable 

And you just wanna attend it and understand. And and then we've got to provide those options. 

00:39:54 Tom Constable 

For all the other reasons you mentioned, but also why send out 20 people from UK Modi to go and do A to fly out to Orlando to have a few meetings they could have kind of learned and and absorb through a remote link. 

00:40:04 Speaker 3 

Yeah, I saw for more than. 

00:40:05 Speaker 3 

20 though. 

00:40:07 Colin Hillier 

I'm going around the show having a bit of a preview. You know, there's definitely a theme about the meta versus meta theme about distributed training as ever. 

00:40:15 Colin Hillier 

Yeah, it's slightly incongruous to talk about that in terms of the training and then not do it. You know, not have the actions to do it during the conference, you know. And that's not to pick on. 

00:40:24 Colin Hillier 

It's actually that's a problem. Yeah, many conferences that you know they're the same issue, so yeah. 

00:40:29 Colin Hillier 

It's great to talk to people. It's great to sort of have that discussion face to face, but there are many people that can contribute virtually as we have been doing well for the last three years. Twice and that available. 

00:40:39 Tom Constable 

Why do we not think there is a hybrid option here? Is there a fear that if we start going more hybrid the footfall drops then people stop going to the physical location? And is that why the events are trying to move back from the precipice of that hybrid event? 

00:40:51 Speaker 3 

Well, personally I think the. 

00:40:52 Speaker 3 

There is that element to it. I think. There's also obviously the event organisers went through a difficult time during the pandemic and they just want to get things back. 

00:41:01 Speaker 3 

Yeah, to what they would call normal, but I I think over time it's just inevitable that consumers will want more hybrid events because I want to be able to pick and choose which events I go to. I was delighted to go to Bath in the UK. 

00:41:13 Speaker 3 

Few weeks ago. 

00:41:13 Speaker 3 

For a NATO conference and it it was great to meet people in person, but I think I, you know, shout out to NATO on that occasion anyway. 

00:41:20 Speaker 3 

Thing where they did actually make it hybrid so people could watch it. And of course that wasn't a commercial event, it was NATO, so I certainly think government sponsored events should go hybrid as a matter of principle and the reasons you've said. 

00:41:31 Speaker 3 

Also, Tom, about the carbon footprint of these events. So I totally agree with that. I think over time, as Dan I say at the Metaverse in all these technologies would just get better and better. The reasons for not doing any hybrid is cannot be defensible. 

00:41:43 Speaker 3 

Then it's off for. 

00:41:44 Colin Hillier 

A while maybe one one will come back to in. 

00:41:47 Colin Hillier 

The future. 

00:41:47 Speaker 3 

So our senior correspondent, Valter Ulrich, who was a previous editor of MST, has written. 

00:41:53 Speaker 3 

A book called all. 

00:41:54 Speaker 3 

But flying assimilation, which sounds really interesting to me. 

00:41:56 Speaker 3 

Anyway, I, I'm sure if you sat in a conference of. 

00:42:00 Speaker 3 

Be kind and not just a history. When anyone touches on old simulation, they will show a picture of a French gentleman sitting in a barrel and then the barrel is on top of another barrel. 

00:42:10 Speaker 3 

The reason this is shown off deemed to be the first ever flight simulator from 1910, so a long time ago. 

00:42:17 Speaker 3 

Even I'm not that old so but it's a long time ago. I think what's interesting about this story. 

00:42:21 Speaker 3 

It's definitely worth a read is it wasn't because they were just trying to recreate the whole experience. What they were trying to do was. 

00:42:29 Speaker 3 

To give the pilots or the trainee pilots a feeling of how to use the controls for this aircraft of the day. 

00:42:36 Speaker 3 

Because it was designed specifically for an end to Annette aeroplane and it had rather unintuitive control. So if you look at the pictures on the right hand side, you turn a wheel to control the elevator on the left you use the wheel to control the lateral movement. 

00:42:52 Speaker 3 

So it's not like a joystick which is actually quite an intuitive thing. This was, you know, you had to turn one and then turn the other, your your left and right were. 

00:43:00 Speaker 3 

Essentially, having to control the aircraft because you use your feet to control like a rudder. There were no fancy hydraulics or electrical it the there's a whole bunch of guys standing around it controlling the aircraft, and when you see the picture, yeah, I guess they were shouted at the pilot saying no wrong, we turned it the wrong way. But sadly the gentleman is actually sitting in this barrel died three weeks later. 

00:43:20 Speaker 3 

Really, in an aircraft accident was rather tragic. The lesson for us now is in the kind of still the role of humans instruction and also how simulation can really try and give much better feel for this kind of human factors of these devices perform not necessarily a flight simulator, any device, so you know the driving need was to try and understand these controllers. 

00:43:42 Speaker 3 

Before they went into the air. 

00:43:43 Colin Hillier 

It is fascinating because we think when we say simulation, we think our stuff that happened in the last 20 years and actually it's it's at least 100 years old, probably probably older. 

00:43:53 Speaker 3 

Yeah yeah, according to this this 1910 this, this simulator and I would say the rate of progress and simulation technology after that wasn't exactly fast, even though the First World War there was obviously need to train on the ground before the pilots took to the air due to the horrendous number of deaths for pilots, the rate of progress in technology then. 

00:44:13 Speaker 3 

Wasn't very fast at all, perhaps compared to now, but whether our ability to embrace technologies and get them into service quickly, as any much better now I I don't know, that's the. 

00:44:22 Speaker 3 

Project in itself. 

00:44:23 Tom Constable 

I love that we've picked a topic to discuss that is based on an image, so I'll do my best to make sure the show notes have a picture in it, or at least a link to the picture so that we can. 

00:44:33 Tom Constable 

You know, if you would like to join us in the journey of the man in a barrel on a barrel doing the patting his head and rub his chest scenario that actually turned out to unfortunately seemed to be the chappy died three weeks later. 

00:44:43 Colin Hillier 

He's he still died. 

00:44:43 Tom Constable 

And we reach that. 

00:44:45 Tom Constable 

Yeah, so maybe those are more innovations to be done on that on that barrel, but if you'd like that then you'll be in the shows. 

00:44:48 

Maybe that's. 

00:44:51 Speaker 3 

Also, the lesson is that he was the actual inventor of the simulator. 

00:44:56 

That's so. 

00:44:57 Speaker 3 

Yeah, but what we don't know is that. 

00:44:59 Speaker 3 

These aircraft because. 

00:45:00 Speaker 3 

Of the sort of the way they were designed, they tended to fill up for in the air anyway. 

00:45:04 Speaker 3 

So you can't unnecessarily blame the simulator in this case. 

00:45:08 Tom Constable 

That's horrendous, Annie. Thank you so much as always. Font of all Knowledge, Colin, I hope you have fun at itsec. 

00:45:14 Tom Constable 

I'm absolutely having a lot of FOMO and I wish I was out there with you right now. Everything we've discussed, including the picture, will be in the show notes and I I look forward to chatting to you next episode, Andy. 

00:45:24 Speaker 3 

Great stuff Tom Colin. Thank you. 

00:45:28 Tom Constable 

Wonderful this week I think just to tail off the podcast. I'm gonna ask the listeners to provide us kind of specific feedback. 

00:45:36 Tom Constable 

So on our news, how do we make our news section better? Are there any topics you want us to cover or any way in which we, as a three of us, can make it more dynamic or interesting from your perspective? 

00:45:47 Colin Hillier 

Yeah, absolutely Tom. Discussing this with Andy before the show. MSNT guys cover a lot of ground. Fantastic job, but they're not everywhere. 

00:45:55 Colin Hillier 

So there's some interesting stories, especially from sort of outside our little corner of the world. Then we would love to hear more about that, and potentially Andy can feature on that segment. So do get in touch. 

00:46:05 Tom Constable 

And the community is continuing to grow on the LinkedIn page. So just search for Warfighter podcast on LinkedIn and Press follow on the page. 

00:46:13 Tom Constable 

Or alternatively, again, look in the show notes. Have a look at our website or even send us an e-mail but old school contact at warfighterpodcast.com. That's all from me. Anything from you, Colin? 

00:46:23 Colin Hillier 

No, nothing from me. See you in two weeks.