WEBVTT
00:00:00.160 --> 00:00:10.000
Uh first off, and you said this earlier, Walt, and we said it during the panel too, but all all innovations, all technology, these are tools for humans.
00:00:10.640 --> 00:00:18.079
This is here to enable, to uplift, to enhance uh humans and move us forward.
00:00:18.320 --> 00:00:21.839
And um uh none of us can predict the future.
00:00:22.000 --> 00:00:27.280
Uh uh, even our predictive models can't 100% predict the future.
00:00:27.519 --> 00:00:35.920
But I'll say that if we look into recent history, so I'm I'm 45, I just uh had a birthday a couple of weeks ago.
00:00:36.079 --> 00:00:52.560
I remember um I remember before the internet, I wasn't in the workforce, but I do remember a time before the internet, I remember before smartphones, I can I can remember what life was like before and after these other sorts of real disruptive technologies.
00:00:52.640 --> 00:00:58.399
And I feel like we've said the word disruptive so much in the last 10 years that like it's almost lost its meaning.
00:00:58.560 --> 00:01:08.079
Um but in these situations, we didn't have a net loss of opportunity uh for people to work, for people to engage with the technology.
00:01:08.159 --> 00:01:11.200
What we saw was new industries, right?
00:01:11.519 --> 00:01:15.599
Uh uh new types of technology, new ways for people to work.
00:01:15.760 --> 00:01:20.000
Uh a blockchain would not exist if it weren't for the internet, right?
00:01:20.159 --> 00:01:24.159
Like there's there's brand new things that come out of these technologies.
00:01:24.400 --> 00:01:30.799
And is it the the only thing that's safe to assume, I think, is that things will change.
00:01:30.959 --> 00:01:45.359
And we all need to be ready to adapt, to see what's possible and see what tools are out there, and find ways for us to use the tools to enhance our work rather than allow the tools to replace our work.
00:01:45.519 --> 00:01:47.439
That's that's the way I feel about that.
00:01:52.640 --> 00:02:00.640
You may know the American Railway Engineering and Maintenance of Way Association, or AREMA, as the quote, keepers of the manuals.
00:02:00.719 --> 00:02:05.840
You may know them as the quote, people behind the largest annual railroad conference in North America.
00:02:06.079 --> 00:02:08.639
Heck, you may not know about AREMA at all.
00:02:08.800 --> 00:02:19.840
This podcast is designed, no pun intended, to change your view of who AREMA is and how AREMA has changed the trajectory of many railway careers over its 100-plus year history.
00:02:20.080 --> 00:02:29.759
Welcome to Platform Chats with your host, Walt Lesser, where he takes a moment to discuss the impacts AREMA has had on the very people who are proud to be called members.
00:02:29.919 --> 00:02:53.439
Are you ready to roll with AREMA?com.
00:02:53.759 --> 00:02:57.840
Hey everybody, welcome back to another episode of Platform Chats.
00:02:57.919 --> 00:03:06.639
I'm your host, Walt Blesser, and wildly enough, we are on episode number 47, 46, 48, something like that.
00:03:06.719 --> 00:03:12.400
I know there's a ghost episode somewhere in my list that never quite got recorded, but we'll get to that one.
00:03:12.560 --> 00:03:15.120
Uh sometimes people have to go do other things.
00:03:15.360 --> 00:03:20.080
Um first time to do this at Platform Chats.
00:03:20.400 --> 00:03:30.479
We have decided to host a podcast as a follow-up to our emerging technologies panel that we hosted in Arima, at Arima in Indy.
00:03:31.039 --> 00:03:39.199
As many of you know, uh the Emerging Technologies panel is actually the birth child of Platform Chats.
00:03:39.360 --> 00:03:55.520
So Platform Chats started in five or so years ago as a podcast to reach out to a whole new host of members uh as well as non-members to potentially get interest in joining Arima.
00:03:56.000 --> 00:04:07.680
And after a couple years of that, there was a request to do one live, which that then turned into the genesis of the Emerging Technologies panel where we could do just a live panel discussion.
00:04:07.840 --> 00:04:09.360
So similar but different.
00:04:09.439 --> 00:04:14.080
Uh, and our first one was in 2022, I believe, uh, in Denver.
00:04:14.400 --> 00:04:18.800
And this that was on a Wednesday morning, and we had a we had a decent sized room.
00:04:18.879 --> 00:04:36.399
Uh, it wasn't the world's largest room, but uh we then have a lot of requests to Arima that hey, people liked it, they liked interacting with the panel, they liked asking live questions and getting live answers instead of just being talked to, like many of our presentations that we do.
00:04:36.560 --> 00:04:38.480
Uh very communication 1.0.
00:04:38.560 --> 00:04:40.399
We're trying to do more of a 2.0.
00:04:40.639 --> 00:04:45.600
And uh believe it or not, this year we got moved to the uh to the Monday session.
00:04:45.839 --> 00:04:52.240
We were given a full hour and we were given a very large room, and it was it was an honor to be able to moderate this panel.
00:04:52.319 --> 00:05:00.560
I'm I'm just really, really cool thing uh that we can do this, and uh that the reception uh has been so positive.
00:05:00.639 --> 00:05:08.720
So thank you all for your feedback and for pushing Arima headquarters to uh to do things that you guys want to hear about.
00:05:08.879 --> 00:05:11.040
And so this year uh we did AI.
00:05:11.199 --> 00:05:30.800
Uh we have done in the past, we've done battery electric uh locomotives, we've done hydrogen power, uh, we've done all kinds of interesting stuff that the goal being listeners and those that are in the audience can think about it uh and not have the answers.
00:05:31.120 --> 00:05:33.199
And frankly, the people on the stage don't have all the answers.
00:05:33.279 --> 00:05:34.000
And that's the point.
00:05:34.079 --> 00:05:34.800
That's the whole point.
00:05:34.879 --> 00:05:45.040
We like to play in a space where uh no one has all the answers because we very much want to uh get people excited about trying to figure things out.
00:05:45.120 --> 00:05:59.839
Um, there was a really cool advertisement during Sunday Night Football last night that I was watching with uh something about how this is a horrible time to be a problem, uh, because there's a lot of individuals out there that are trying to quote unquote find their problem to solve it.
00:06:00.000 --> 00:06:11.920
And uh that's inspiring to people like my kids, uh, and uh hopefully a lot of people that are even listening to this to figure out what that problem is and is there a way to apply technologies of today to solve it.
00:06:12.079 --> 00:06:21.120
Um and so with that, I did bring back our two panelists, uh, both Jason and Doran are here from um Railvision and Xtrack Object Computing.
00:06:21.279 --> 00:06:27.120
I'm gonna have them do a brief uh introduction for those of you that did not sit through the panel.
00:06:27.199 --> 00:06:43.439
Um we are not gonna do the full 20-minute PowerPoints, but they'll each do a brief introduction as to kind of who they are in the world and why they're in the roles that they're in and heck, why they're even interested in trying to apply AI to um to the rail industry.
00:06:43.680 --> 00:06:50.720
And then what we're gonna do is I've got a oh man, we've got so many questions uh that were submitted through the app.
00:06:50.879 --> 00:06:53.439
And we're gonna use that as the basis of today's podcast.
00:06:53.600 --> 00:06:57.839
Again, this is very different from uh what we've done in the past.
00:06:59.600 --> 00:07:01.439
And uh we'll start with Jason.
00:07:01.680 --> 00:07:06.399
Jason, if you want to go ahead and do an introduction, that'd be great.
00:07:06.959 --> 00:07:07.600
Absolutely.
00:07:07.759 --> 00:07:14.560
Thank you so much, Walt, for inviting me here today to talk to you, and then also for the invitation uh on the panel.
00:07:14.720 --> 00:07:20.160
That was an amazing experience for me, and so I'm just I'm so thankful for all of that.
00:07:20.319 --> 00:07:22.560
Um, so yeah, my name is Jason Schindler.
00:07:22.879 --> 00:07:31.600
I've been uh uh working at Object Computing for about 13 years, and I have about 20 years in the technology and innovation industry.
00:07:31.759 --> 00:07:39.600
Um, object computing is a technology consultancy company, and that's really where my work has been for the last 20 years.
00:07:39.680 --> 00:07:50.480
So I've spent 20 years helping organizations of all shapes and sizes uh identify and adopt the emerging technologies that's gonna help move their business forward.
00:07:50.639 --> 00:07:52.560
I've done that in a number of industries.
00:07:52.720 --> 00:08:01.360
Um, for the last five years, uh I've been working with the Xtrack platform, which I think we're gonna talk about a little bit more in a moment, perhaps.
00:08:01.439 --> 00:08:05.279
Um, but the Xtrack platform is a platform specific to Rail.
00:08:05.519 --> 00:08:12.240
It brings AI and ML insights into like operational and safety efficiencies within Rail.
00:08:12.480 --> 00:08:14.720
And that's that's why I'm here today.
00:08:17.680 --> 00:08:18.639
Outstanding.
00:08:18.879 --> 00:08:23.120
And um we'll get to your products here in a second, Jason.
00:08:23.199 --> 00:08:26.079
So thank you for uh for the quick introduction.
00:08:26.319 --> 00:08:30.079
Uh Doran is our other guest who is with Rail Vision.
00:08:30.319 --> 00:08:32.159
Doran's based in uh Israel.
00:08:32.320 --> 00:08:33.519
Doran, are you on there?
00:08:33.840 --> 00:08:34.639
I hope.
00:08:34.960 --> 00:08:36.320
Can you hear me, guys?
00:08:36.639 --> 00:08:37.679
Absolutely.
00:08:38.000 --> 00:08:38.639
Good.
00:08:39.759 --> 00:08:44.879
So um again, thank you as well for joining, uh, for joining this uh podcast.
00:08:44.960 --> 00:08:48.080
And again, I had a great time with you guys uh at the panel.
00:08:48.240 --> 00:08:52.080
It was very interesting, and I'm really happy on the opportunity to continue this on the panel.
00:08:52.240 --> 00:08:56.960
I'm Doron Coadier, I'm the VP business development here at RailVision.
00:08:57.360 --> 00:09:02.159
Um actually, I came from the world of AI uh in my previous roles.
00:09:02.320 --> 00:09:06.320
I used to be working in the defense industry, the automotive, and now the rails.
00:09:06.639 --> 00:09:16.159
So I saw it in different areas uh in different perspectives, and you know, I truly believe that AI has a lot to bring to the railway industry.
00:09:16.399 --> 00:09:19.120
Uh, we're gonna talk in a minute as well about RailVision.
00:09:19.279 --> 00:09:30.320
RailVision basically provides uh very unique sensors to the rail industry in the field of uh uh operational efficiency, safety, uh, but we'll talk about it in a second.
00:09:30.399 --> 00:09:34.879
So I'm very very proud here to join uh this podcast.
00:09:36.000 --> 00:09:41.120
So if we stay on the theme of uh find your problem, which I really enjoyed that theme.
00:09:41.200 --> 00:09:43.679
I I gotta figure out who's behind that advertisement.
00:09:43.840 --> 00:09:52.480
Um so again, entrepreneurs, startups, etc., they all start with a problem statement, right?
00:09:52.639 --> 00:09:59.600
We all look at something in the industry and say, I think we can do this better or more efficient or safer.
00:10:00.080 --> 00:10:01.759
Uh I'm gonna go to you, Jason.
00:10:01.840 --> 00:10:06.720
What was your problem statement for what you're doing uh at Xtrack?
00:10:07.519 --> 00:10:18.080
It was really about finding ways to bring more modern uh data insight tools to the rail industry.
00:10:18.320 --> 00:10:54.879
And we were um we we started off with a single problem, which was identifying the root cause of unscheduled breaking events by consuming PTC device logs and building an ML uh model to read the logs, identify the root cause, and then we have a human in the loop check at the end where if an individual feels that that uh diagnosis is wrong, uh they can provide reasons for what it should be, and then we retrain the model based on that information, and then the next version of it would include those things as well.
00:10:55.039 --> 00:10:58.799
So that's that's where we got started with that very specific use case.
00:10:58.960 --> 00:11:12.720
And that was very helpful because it gave us the opportunity to work directly with folks in the rail industry instead of kind of a lot of a lot of people when they're building tools, they're also imagining problems instead of to your point finding them.
00:11:12.799 --> 00:11:21.840
Uh so they just they imagine a problem and then they go solve that, and then they find out there's no customers, uh, right, because it's not actually a problem, or they didn't imagine it correctly, or whatever.
00:11:22.080 --> 00:11:36.480
Um, a strong belief of ours is we need to partner with the people that have the problems, be able to confirm them, know how to best solve them, and how to best put our uh solution out in order to where it can be consumed and impactful.
00:11:37.360 --> 00:11:38.720
Outstanding, yes.
00:11:38.960 --> 00:11:42.639
And I I I wholeheartedly agree with that uh approach.
00:11:42.720 --> 00:11:49.919
Um, it's coming up with a problem that doesn't exist could lead you to uh not a very long-lasting startup.
00:11:50.720 --> 00:11:51.840
A lot of work, at least, yeah.
00:11:52.080 --> 00:11:52.399
Yeah.
00:11:52.639 --> 00:11:54.320
How about you at RailVision?
00:11:54.480 --> 00:11:59.840
What did you uh how did you envision the issue in the rail industry and how you were gonna apply AI?
00:12:00.480 --> 00:12:04.799
Yeah, well again, RailVision was founded to solve a very big problem.
00:12:04.960 --> 00:12:10.240
There's a huge amount of accidents worldwide uh in the rail industry.
00:12:10.480 --> 00:12:22.559
Just to give you some figure facts, in 2024 there were almost a thousand rail fatalities, almost 7,000 non-fatal rail injuries, more than a thousand derailments.
00:12:22.799 --> 00:12:26.720
So it's a big problem, not only obviously in the US, worldwide.
00:12:26.960 --> 00:12:41.200
And uh I even I even mentioned, I think, in the on the on the panel itself, that the the AAR president at the time, a few years ago, said that every three hours in the US, a person or a vehicle is struck by train.
00:12:41.360 --> 00:12:43.679
So this is what we're trying to solve.
00:12:43.919 --> 00:12:52.240
And but not only we're trying to solve it, we understand that one of the big challenges is the human cause behind this problem.
00:12:52.720 --> 00:12:58.639
Uh, either the limited eyesight of a human being, uh, the limited human resources.
00:12:58.720 --> 00:13:15.039
In some areas, they're not enough, for example, train engineers or drivers, so they work longer hours, so you know, putting more load on these uh on these individuals, harsh environment, you know, snow, fog, rain, loss of attentions.
00:13:15.200 --> 00:13:28.399
So the many, many problems coming from the human cause, and what we're trying, we're trying to assist uh and to come with the you know, we're bringing AI in order to solve this uh big problem.
00:13:31.679 --> 00:13:38.240
So it's and again, it feels very much like a tool in the toolbox, though, right?
00:13:38.399 --> 00:13:48.240
We're not trying to replace folks, we're trying hard to uh help give them necessary tools to make their jobs easier, better, faster.
00:13:48.399 --> 00:13:49.759
Is that a true statement?
00:13:50.960 --> 00:13:53.919
Yes, this is okay for more effective.
00:13:54.000 --> 00:13:55.360
Yes, it's it's a true statement.
00:13:55.519 --> 00:14:00.799
We want to actually, you know, human beings are limited in what uh the their abilities.
00:14:00.879 --> 00:14:03.360
We're trying to, you know, increase their abilities, okay?
00:14:03.519 --> 00:14:12.000
See beyond what a human person can do in order to assist them to do it in more safer, operationally uh more efficient.
00:14:12.639 --> 00:14:16.000
Uh yes, so it's it's quite a true true statement here.
00:14:16.720 --> 00:14:25.759
And Jason, you did a really good job at the panel setting the stage for before because before we go any further, there's a lot of questions here about AI.
00:14:25.919 --> 00:14:26.240
Okay.
00:14:26.480 --> 00:14:29.679
Um, and when we use that term, it's very, very broad.
00:14:30.000 --> 00:14:35.919
And when I think of AI, I think of anything from machine learning, I think of uh neural networks.
00:14:36.159 --> 00:14:42.080
Uh when people hear that term heck, I think some people just start thinking about data centers.
00:14:42.399 --> 00:14:42.960
Right.
00:14:43.360 --> 00:14:47.759
So uh and video chips and et cetera, et cetera, and and semiconductors.
00:14:47.840 --> 00:15:07.039
And so when you when we use the term AI, could you just maybe give us a 20,000-foot overview of your view of what AI is if you say you're at the cocktail party having a conversation with someone who knows nothing about it, and then let's drill down into how we're using it or applying it uh here.
00:15:07.200 --> 00:15:08.000
That'd be great.
00:15:08.399 --> 00:15:08.639
Sure.
00:15:08.799 --> 00:15:09.679
Yeah, happy to.
00:15:09.919 --> 00:15:16.399
So um AI itself is a broad, like it's it's a umbrella uh field of study, right?
00:15:16.480 --> 00:15:24.480
So it's been around since the 50s, and it's really focused on designing systems which can mimic certain aspects of human intelligence.
00:15:24.639 --> 00:15:29.279
And that like AI, artificial intelligence, that that's really as an umbrella term what it is.
00:15:29.440 --> 00:15:33.440
Um it's it's in our vernacular a lot right now.
00:15:33.600 --> 00:15:37.279
It's a very buzzword thing uh for good reason.
00:15:37.519 --> 00:15:55.600
Um uh but ML or machine learning is a specific kind of AI, and it uses uh like statistical models and algorithms to like uh predict elements of future unseen data based on training data that it's already received.
00:15:55.759 --> 00:16:11.919
Uh so an example, um, I don't recall if this was the one I used at the panel, but uh if you wanted software that could take an image and tell you whether or not the image contains a duck, you could train an ML model uh on a lot of images that do contain ducts, a lot that don't.
00:16:12.080 --> 00:16:15.840
And uh after sufficient training, you would have a model that you're pretty confident.
00:16:16.080 --> 00:16:22.000
If I show it an image it's never seen before, it would be able to correctly identify if there's a duck in that image.
00:16:22.159 --> 00:16:27.840
Um that's that's based on training data, and then it's it's that aspect of unseen data that makes it ML.
00:16:28.480 --> 00:16:39.840
That uses something called deep learning, which I won't go too far into, uh, except to say that deep learning, that's where your neural network comes in, and it's that technology that has enabled Gen AI.
00:16:39.919 --> 00:16:45.440
And I think Gen AI, when folks say AI, Gen AI is basically where they start thinking.
00:16:45.519 --> 00:16:53.360
So it's it's short for generative AI, and it's any AI that starts to create new content where content didn't exist before.
00:16:53.600 --> 00:17:08.480
And one of the big uses of that right now, if you think about ChatGPT or Gemini or whichever LLM models you're using, um, it's it's creating human speech or human language uh in order to interact with us from nothing.
00:17:08.559 --> 00:17:10.160
And so that's generative AI.
00:17:10.240 --> 00:17:18.640
And I think when folks think about AI, at least right now, a lot of them are thinking about generative AI, but it really is a big umbrella and it could mean a lot of things.
00:17:18.799 --> 00:17:19.680
Is that helpful?
00:17:20.000 --> 00:17:21.200
That is really helpful.
00:17:21.359 --> 00:17:33.599
And and yes, I think AI is at a point now where uh similar to the battery electric train conversation we had, you can relate to the problems as a consumer no differently than a large business, right?
00:17:33.680 --> 00:17:44.640
When you go to a if you're if you're deciding between an electric, fully electric vehicle versus a an ice uh internal combustion engine, which is classically what we've been driving, right?
00:17:45.039 --> 00:17:50.720
You sit down and you think, all right, well, okay, if I'm gonna go on a long road trip, what does that mean versus if I'm gonna use it around town, blah, blah, blah.
00:17:50.799 --> 00:17:52.160
You you you can kind of relate, right?
00:17:52.240 --> 00:17:55.920
It's no different than when we're thinking about that with battery electric trains.
00:17:56.160 --> 00:17:58.000
And right now, ChatGPT.
00:17:58.079 --> 00:18:09.839
Uh, my wife and I family have a trip upcoming, and believe it or not, ChatGPT has pretty much created an itinerary and it's told us when where we should go and when for the kids at what ages they are.
00:18:09.920 --> 00:18:14.000
It's it's phenomenal, but that's how consumers are are using it, right?
00:18:14.160 --> 00:18:19.759
Um, and Doran, you were famously talking about garbage in, garbage out.
00:18:19.920 --> 00:18:23.119
Uh, is this a good point where you'd want to chime in on this discussion?
00:18:23.519 --> 00:18:30.160
Yeah, I mean, um, you know, AI is a very big uh umbrella of a lot of uh things underneath it.
00:18:30.240 --> 00:18:32.960
But at the low level, the foundation is the data.
00:18:33.119 --> 00:18:37.519
Okay, and again, data is as you mentioned, garbage in, garbage out.
00:18:37.599 --> 00:18:46.799
If the data, the lowest point here, the the foundation, you provide uh low quality data, you'll get low quality AI in the end of the day.
00:18:47.039 --> 00:18:50.559
Uh, and how do we actually in our company create high quality?
00:18:50.640 --> 00:18:54.480
So, first of all, use very, very uh powerful sensors, okay?
00:18:54.640 --> 00:18:59.200
So high quality images with a lot of pixels and a lot of sensitivity.
00:19:00.000 --> 00:19:02.480
Uh, we use diverse inputs, okay.
00:19:02.559 --> 00:19:06.799
We're not using only visible cameras, we use uh thermal cameras as well.
00:19:06.960 --> 00:19:15.519
So when you have diverse inputs, you increase the quality of the data, the quantity of the data, we use a lot of data.
00:19:15.680 --> 00:19:38.240
Okay, uh we ran more than one million kilometers of rail already, collecting endless amounts of data out there in order to train our uh our different models and the diversity of conditions, we do it in daytime, nighttime, uh in Europe, in the US, in the Far East, uh you name it, in shunting yards, in main lines.
00:19:38.400 --> 00:19:56.799
So, the more you get high-quality data of image quality, diverse inputs, quantity, and diversity, you get a very, very good uh uh uh bucket of data in which you can train the machine learning and therefore provide a much higher level of uh performance of AI.
00:19:59.440 --> 00:20:04.720
I think this is a perfect time to jump in with some of the questions because I think the stage is set.
00:20:05.119 --> 00:20:11.680
Uh I think that uh there are a lot of questions about the data.
00:20:11.920 --> 00:20:17.839
And so that's why I want to chime in here because garbage in, garbage out, quality data, etc.
00:20:18.400 --> 00:20:27.200
Um, but we still live in a time where I firmly believe there never is enough data to be perfect.
00:20:27.440 --> 00:20:42.000
Uh in 2016, uh, the company I was running, uh, we had a monitoring company, we were using accelerometer data, and we were training neural networks to identify a certain type of uh uh data image or response.
00:20:42.319 --> 00:20:49.519
And man, you might think you've got that thing trained just right, and then here comes something else that's just a little bit different.
00:20:49.759 --> 00:21:00.559
And so that's one of the big questions we have here, and we are this time uh going to try to do my best to follow the number of questions that were voted with the highest number of votes.
00:21:00.640 --> 00:21:02.720
So this was the highest number of votes for real.
00:21:02.880 --> 00:21:10.240
I know that there was some folks in the audience that noticed I was not seeing all that when we were in the middle of our presentation.
00:21:10.480 --> 00:21:15.359
The questions were not floating up properly, but this question uh is for both of you.
00:21:15.519 --> 00:21:19.759
Uh, we know that we don't yet have 100% accurate AI models.
00:21:20.160 --> 00:21:23.119
What is the approach of your company to address these inaccuracies?
00:21:23.279 --> 00:21:25.119
False positives, false negatives.
00:21:25.279 --> 00:21:28.160
Um, do you keep the human in the loop for QA?
00:21:28.319 --> 00:21:32.240
This was pointed towards Jason, but I definitely think this is a question for everybody.
00:21:32.400 --> 00:21:33.599
So, Jason, you want to start?
00:21:33.680 --> 00:21:34.319
That'd be great for you.
00:21:34.559 --> 00:21:35.359
I'd be happy to, yeah.
00:21:35.440 --> 00:21:38.000
And I and it certainly is like a general purpose question.
00:21:38.079 --> 00:21:41.599
And it's it's also this really isn't specific to Rails.
00:21:41.839 --> 00:21:45.920
I think everyone, there's a lot of people asking themselves uh this question right now.
00:21:46.079 --> 00:21:54.400
And I think it's important to remember that uh AI itself um and and the models that we're building, these are these are tools.
00:21:54.559 --> 00:21:59.119
Uh there's other innovations that have been in similar spaces previously.
00:21:59.359 --> 00:22:02.319
And Uh and those things aren't always accurate either.
00:22:02.480 --> 00:22:04.240
So AI is not human, right?
00:22:04.400 --> 00:22:07.119
And also humans themselves aren't perfect.
00:22:07.359 --> 00:22:11.440
So how do we approach like potential inaccuracies?
00:22:11.599 --> 00:22:23.279
Well, the the first is kind of knowledge that you're as you're tuning a model, you're tuning it for specific uh outputs or specific uh predictions.
00:22:23.680 --> 00:22:32.720
And you can kind of you can in part of your training decide what aspects are important to you and which ones are least important to you.
00:22:32.960 --> 00:22:37.680
And so if you're thinking like, how am I going to predict everything?
00:22:37.759 --> 00:22:42.799
Uh the the the reality is you're just you're probably not, or at least not very well.
00:22:42.960 --> 00:22:47.920
And each one of those elements has a different kind of level of accuracy that's going to be associated with it.
00:22:48.000 --> 00:22:54.319
And tuning the model, you're going to optimize for for one or two things and not try to optimize for everything.
00:22:54.640 --> 00:22:57.519
So how do we and how do we address them?
00:22:57.680 --> 00:23:00.960
We do the same thing we would do with any other technology.
00:23:01.279 --> 00:23:07.680
We test the heck out of it, we test it with simulated data, we test it with real data.
00:23:07.839 --> 00:23:20.160
Um, we do like we're we're big believers in keeping the human in the loop until we come to a point where we feel really good about this model and its accuracy against the real world.
00:23:20.319 --> 00:23:28.079
And in some cases, that can be at a terminal in a system, somebody sees an output in the model and says, no, I don't think that's right, and they give it feedback.
00:23:28.240 --> 00:23:38.160
Um, in some ways, so when we talked about like crossing sight lines on the panel and uh predicting when vegetation will grow at what areas, um, it's it's visual inspections, right?
00:23:38.240 --> 00:23:50.000
It's having folks go out uh and do uh uh snapshots of what they're already doing now in order to inspect these crossings and give feedback into that model, you know, making sure that we um that we have it right.
00:23:50.240 --> 00:23:57.039
So it's it's tested a lot, um, have high standards and keep humans engaged, right?
00:23:57.119 --> 00:24:01.359
It's not like it's not a set and forget situation, and I don't think it ever should be.
00:24:01.920 --> 00:24:07.680
And I'm gonna, I know we got like 60 questions, so we're gonna try to bust through as many of these as possible.
00:24:07.839 --> 00:24:14.240
That was a great answer, and I think the reason I really wanted to start with that one is that's an overarching answer and question, right?
00:24:14.400 --> 00:24:20.240
Um, another question here was the rails, the rail signal industry historically is very risk-averse.
00:24:20.400 --> 00:24:32.880
And with the potential of AI integration into train control, how is the industry managing some of the faults that happen in AI data sets like data contamination, concept drift, model collapse, uh that sort of thing?
00:24:33.039 --> 00:24:34.960
That also ties in with another question.
00:24:35.039 --> 00:24:38.559
I'm trying to group some of these together, which was directly for Doran.
00:24:38.799 --> 00:24:41.440
Does the system interact with positive train control systems?
00:24:41.599 --> 00:24:43.119
So I would say those two are very similar.
00:24:43.200 --> 00:24:44.000
Do you, Doran?
00:24:44.079 --> 00:24:47.440
I'm gonna kick that one to you unless you want some help.
00:24:47.759 --> 00:24:49.279
Uh no, I think it's fine.
00:24:49.440 --> 00:24:53.279
Um, first, trying to address the first question again.
00:24:53.359 --> 00:25:04.640
I think by training these AI models with different types of uh I mentioned, for example, different types of uh uh cameras that we use, for example.
00:25:04.799 --> 00:25:11.039
Therefore, you you you you you use different types of inputs in order to improve the performance.
00:25:11.200 --> 00:25:18.400
So you reduce the contamination of these uh uh the these set data uh sets that you get.
00:25:18.799 --> 00:25:24.400
Um, as uh obviously Jason mentioned, there's always at this stage always a man in the loop.
00:25:24.480 --> 00:25:27.279
Okay, there's also always a human being in the loop.
00:25:27.440 --> 00:25:33.039
And by the way, uh again, we need to understand AI is not a revolution, it's an evolution.
00:25:33.200 --> 00:25:39.839
So you start from a certain point, it's far from perfect, and it keeps getting better and better.
00:25:40.079 --> 00:25:47.759
Through these uh uh you know uh steps, there's always a human being that tests all the time, but then again, this is not enough.
00:25:48.000 --> 00:25:55.119
So we're bringing in order to reduce the data contamination, we do a lot of testing, we put the human loop inside.
00:25:55.519 --> 00:25:59.440
Um uh we use different inputs.
00:25:59.519 --> 00:26:05.200
Uh, there's a a different angle to it as well, from a cyber perspective.
00:26:05.359 --> 00:26:10.799
Okay, you can contaminate data by you know tapping onto it and changing the data as well.
00:26:10.960 --> 00:26:24.640
So, this is another very important thing as you go into AI and high levels of use of AI is really increase the cybersecurity aspect as well from a contamination uh perspective.
00:26:24.960 --> 00:26:29.279
And you know, we give a lot of deep deal detail into these things as well.
00:26:29.359 --> 00:26:34.079
So we put a lot of measures to reduce uh uh these things.
00:26:34.160 --> 00:26:36.480
So I hope I you know I give a very 30,000.
00:26:37.119 --> 00:26:38.319
Yeah, Jason, go on.
00:26:38.400 --> 00:26:39.680
Uh please, if you want to add.
00:26:40.000 --> 00:26:43.279
Oh, sorry, I didn't want to interrupt that, but I did want to want to just chime in here.
00:26:43.359 --> 00:26:54.559
I'm I'm really glad, Doran, that you said um a couple times that it's like multiple dispersed data sets integrated together that really kind of level this stuff up, because I 100% agree with that.
00:26:54.640 --> 00:27:09.039
And I actually think that's more the immediate thing that we need to work through in order to make the best use of this technology is is integrating and bringing in those multiple data sets uh together to make sure we have the best understanding of reality as possible.
00:27:09.279 --> 00:27:22.480
The other thing I I want to chime in on here, and it it relates to the last question too, is we need to be so it before we start on this, we need to understand what success is and how to measure it, right?
00:27:22.640 --> 00:27:52.480
And so we need to be continuously measuring what success looks like, and then taking every iteration of that model, uh, every future version and watching that those metrics come back so that we can continue to know that we're hitting the marks, so that we don't have this like concept drift sort of problem uh where if if we keep our eye on the ball in terms of where success is, then we can make sure that all future revisions are still hitting the mark.
00:27:52.559 --> 00:27:55.279
And then if they do start to drift, we can reel them in.
00:27:55.759 --> 00:27:59.039
So uh just yes, let's get to like a yes-no.
00:27:59.359 --> 00:28:09.119
Does the does the rail vision system integrate with PTC and or the signal system as of today?
00:28:09.440 --> 00:28:10.640
As of today, no.
00:28:10.880 --> 00:28:14.880
Uh we comp I know it's a yes and no question, just a very quick one.
00:28:14.960 --> 00:28:17.039
Uh we complement PTC.
00:28:17.119 --> 00:28:26.160
Okay, PTC is it's a signal system, it's a blind system, and the whole idea is to uh you know know where the other trains are and to avoid any collision.
00:28:26.400 --> 00:28:39.519
We live in a world we live in a world that there's much more on the rails, you know, there's things can fall, you know, trees, branches, a rock, uh you know, crossing people, crossing cars, uh you name it.
00:28:39.680 --> 00:28:41.920
Okay, so we complement it.
00:28:42.000 --> 00:28:53.759
Uh at this point of time, we do not uh uh uh you know uh talk to the PTC, but we hear uh from our clients that you know it's it's you know it's a good idea.
00:28:54.000 --> 00:28:57.519
So we uh thinking over internally how to do so.
00:28:57.680 --> 00:29:06.799
Uh so yes, in in the future I believe that yes, we will be part uh of the PTC and will be an additional layer to what the PTC provides today.
00:29:07.279 --> 00:29:10.079
Jason, the your system?
00:29:10.559 --> 00:29:17.519
So we we do have a module now that reads in PTC device logs for that unscheduled breaking event root cause analysis.
00:29:17.680 --> 00:29:23.599
Uh, we don't have any current integrations that interact, especially in a real-time aspect, with PTC.
00:29:24.240 --> 00:29:25.200
Okay, fair.
00:29:25.440 --> 00:29:29.039
And uh and this is actually gonna pile onto what was just said.
00:29:29.200 --> 00:29:43.920
I I enjoyed this question of uh when implementing AI in a safety critical use case where a single failure is unacceptable, how do you comfortably conclude that a model is ready to be moved from development to production?
00:29:44.160 --> 00:29:53.279
And and I think that ties directly into what Doran said about evolution, not revolution, as well as what you said, Jason, about data sets and testing.
00:29:53.440 --> 00:29:59.200
So are there certain thresholds that can be discussed or or are we still trying to figure that out?
00:30:00.079 --> 00:30:01.359
There's always thresholds.
00:30:01.519 --> 00:30:10.559
Actually, um uh we tend to think about ML and predictions as like a yes-no thing, as a as a binary thing.
00:30:10.720 --> 00:30:16.960
Um, in reality, a lot of these models are a sliding scale of confidence, right?
00:30:17.440 --> 00:30:33.680
And if we want to be very protective, we can be more um uh conservative on when we flip the switch and say, okay, I think this needs to be uh dealt with or it needs to be inspected, or you know, just depending on the model that's being used and how that's going on.
00:30:33.920 --> 00:30:41.680
And then the other thing is is we need to be thorough in our testing and our validation, including human validation.
00:30:41.839 --> 00:30:44.319
So it kind of ties into what I was saying a moment ago.
00:30:44.559 --> 00:30:48.960
We before we even start, we need to understand what success is, right?
00:30:49.039 --> 00:30:58.000
So we don't find success, we we identify what we believe it is, and we identify how to measure it, and then we apply that measurement to all the work that we do.
00:30:58.160 --> 00:31:04.799
And so when when does when is it good to move it from development to production when we're feeling good about that success success metric?
00:31:05.039 --> 00:31:26.799
And um, and then how do we mitigate against risks uh when it comes to you know, like no no failures allowed, um keeping humans in the loop for as long as we need to, and or or any other fail-safes that are already in the system to make sure that um uh that the net result is a positive one.
00:31:27.759 --> 00:31:32.720
There's been a lot of questions in here, uh, and I'll give this to you, Doran, to kick off if you'd like.
00:31:32.880 --> 00:31:37.039
Um, there's been a lot of questions just about data ownership, litigation.
00:31:37.200 --> 00:31:40.319
I know we covered this a little briefly in the panel.
00:31:40.559 --> 00:31:48.559
Um, I think there's a you know, what are the legal ramifications of uh missing something or who owns the data?
00:31:48.880 --> 00:31:51.359
Do you guys want to talk about that at 10,000 feet?
00:31:51.519 --> 00:31:58.720
I mean, I don't know if we're there yet as an industry, but I'll I'll I'll touch it if it's uh very very briefly.
00:31:58.880 --> 00:32:02.799
Uh from our aspect, we have two sets of uh products, okay?
00:32:02.960 --> 00:32:08.240
One that sits and provides real-time alerts to the to the engineer, to the driver.
00:32:08.480 --> 00:32:08.720
Okay.
00:32:09.359 --> 00:32:15.200
Uh, and basically he gets real-time alerts on a computer that sits within the cabin itself.
00:32:15.359 --> 00:32:19.680
Okay, it is recorded as well, like a V VR on the computer itself.
00:32:19.759 --> 00:32:25.599
So basically, the data is owned completely by the uh the operator.
00:32:26.319 --> 00:32:41.039
In addition, the other product we didn't talk about too much about our products, but we have a complementary product of uh uh uh operational web application, okay, that basic basically logs all the incidents we collected.
00:32:41.200 --> 00:32:44.640
So it shoots it out to uh uh a cloud.
00:32:44.880 --> 00:32:48.240
Now a cloud here it makes life easier as well.
00:32:48.319 --> 00:32:56.960
We can put it on any cloud, either our cloud and provide the inputs to the uh client, and if necessary, he can put it on his own cloud.
00:32:57.200 --> 00:33:05.119
So basically, on the two types of uh products that we have, it's completely owned by the operator or the customer himself.
00:33:05.279 --> 00:33:08.079
So we're not even touching the data.
00:33:09.359 --> 00:33:14.559
Yeah, that that matches our experience as well in in terms of like who owns the data.
00:33:14.720 --> 00:33:20.319
Um, for the most part, whomever collects it uh generally owns it and then maintains that ownership of it.
00:33:20.400 --> 00:33:30.240
Um will pull it into our pipelines and our system in order to derive the insights, and uh the insights are generally owned by um uh the party as well.
00:33:30.400 --> 00:33:35.680
It's the uh the models and the processing in the middle uh that we focus on for ownership.
00:33:36.160 --> 00:33:44.799
And then in terms of the other, I mean, I'm not a lawyer, not really prepared to give like legal uh legal advice around that sort of stuff.
00:33:44.960 --> 00:33:57.519
Um uh, but I I do think it's it's important to know uh if there's something that's potentially dangerous uh uh that we can do something about.
00:33:57.599 --> 00:33:58.640
I'll just say that.
00:33:59.200 --> 00:34:02.799
I think this question dovetails as well for Doran.
00:34:02.960 --> 00:34:12.079
Uh, what has been the regulatory acceptance from the different countries with enhancing or augmenting what the train engineer uh uh can see?
00:34:12.719 --> 00:34:21.760
And I would even go so far as to expand that based on the conversation we've been having around uh uh regulatory acceptance.
00:34:21.920 --> 00:34:28.960
Are you seeing different countries responding to this technology in a different way?
00:34:31.039 --> 00:34:34.079
Um let me start with the US.
00:34:34.400 --> 00:34:47.280
Okay, uh the US um slowly, slowly we hear more at the MTSB talking about uh uh you know uh recommendation and start using collision avoidance systems.
00:34:47.519 --> 00:34:49.360
Okay, it's not a regulation.
00:34:49.519 --> 00:34:56.239
If it would have been a regulation in the US, I presume life our life will be much easier if it's a regulation than either product.
00:34:56.480 --> 00:35:03.920
We're we're kind of at the point that we're actually educating the market uh with our uh technology.
00:35:04.159 --> 00:35:06.239
So regulation is not there yet.
00:35:06.400 --> 00:35:17.440
Uh, but we see not only in the US, uh, the more and more huge accidents worldwide that could be avoided, but not only the big ones, even the small ones, okay?
00:35:17.599 --> 00:35:23.440
The ones that you you know you hit a switch or you have a derailment, even if there's no one hurt.
00:35:23.599 --> 00:35:40.639
Okay, so I think slowly, slowly the rail industry is starting to understand the value of technology and how they can how AI, for example, can bring a lot of uh you know a lot of uh uh value to the customer.
00:35:40.800 --> 00:35:46.400
Uh but again, there's no regulations as of yet in comparison, for example, to the automotive industry.
00:35:46.559 --> 00:35:55.360
In the automotive industry, in the automotive industry today, when you buy a car, you have ADAS products on the car, similar to what we do today.
00:35:55.440 --> 00:35:58.400
This is part of what you have because it's already regulated.
00:35:58.719 --> 00:36:04.960
I do believe it will be uh and it started already because the NTSB are pushing it.
00:36:05.039 --> 00:36:14.559
I presume the next stage, hopefully the FRA will recommend it to the operators and slowly, slowly it will move hopefully towards regulating it because there's a lot of potential.
00:36:14.800 --> 00:36:21.119
Again, I mentioned earlier how we can help the the the uh uh the the engineer.
00:36:21.280 --> 00:36:38.960
I I can an engineer can see a few hundred meters, I can or a few hundred yards, I can extend what he can see to 1.2 miles, I can show him in pitch dark like it was daylight, I can uh show them switch positions where they are, and if they're faulty.
00:36:39.119 --> 00:36:41.039
I can do so many things here.
00:36:41.199 --> 00:36:45.519
So there's a lot of things that we can add uh and assist the driver.
00:36:45.679 --> 00:37:03.360
And I think the drive behind it from a regulation perspective is all the accidents and the value that the operators are starting to understand, the return on investment they can get with such a technology by reducing the operational expenses and all the accidents they have and everything.
00:37:05.519 --> 00:37:07.440
There's been several questions.
00:37:07.679 --> 00:37:23.440
Um, and this this this can go to both of you about the and we again we covered this briefly in the panel, uh, autonomy or uh basically the elimination of the engineers operating the trains.
00:37:23.760 --> 00:37:32.320
Um do you believe, do the two of you believe that we're headed towards using this technology to eliminate jobs, support jobs?
00:37:32.559 --> 00:37:36.239
Um let's hit the elephant in the room about uh jobs.
00:37:37.280 --> 00:37:39.039
Who wants to take that one?
00:37:39.679 --> 00:37:42.320
I'll I'll start uh if that's all right, Doran.
00:37:42.639 --> 00:37:54.880
So um first off, and you said you said this earlier, Walt, and we said it during the panel too, but all all innovations, all technology, these are tools for humans.
00:37:55.039 --> 00:38:02.960
There this is here to enable, to uplift, to enhance uh humans and move us forward.
00:38:03.280 --> 00:38:06.719
And um uh none of us can predict the future.
00:38:06.960 --> 00:38:12.239
Uh uh even our predictive models can't 100% predict the future.
00:38:12.480 --> 00:38:20.800
But I'll say that if we look into recent history, so I'm I'm 45, I just uh had a birthday a couple of weeks ago.
00:38:21.039 --> 00:38:37.440
I remember um I remember before the internet, I wasn't in the workforce, but I do remember a time before the internet, I remember before smartphones, I can I can remember what life was like before and after these other sorts of real disruptive technologies.
00:38:37.519 --> 00:38:43.280
And I feel like we've said the word disruptive so much in the last 10 years that like it's almost lost its meaning.
00:38:43.440 --> 00:38:52.960
Um, but in these situations, we didn't have a net loss of opportunity uh for people to work, for people to engage with the technology.
00:38:53.119 --> 00:38:56.079
What we saw was new industries, right?
00:38:56.400 --> 00:39:00.480
Uh uh new types of technology, new ways for people to work.
00:39:00.639 --> 00:39:04.880
Uh a blockchain would not exist if it weren't for the internet, right?
00:39:05.039 --> 00:39:09.039
Like there's there's brand new things that come out of these technologies.
00:39:09.280 --> 00:39:15.679
And is it the the only thing that's safe to assume, I think, is that things will change.
00:39:15.840 --> 00:39:30.239
And we all need to be ready to adapt, to see what's possible and see what tools are out there and find ways for us to use the tools to enhance our work rather than allow the tools to replace our work.
00:39:30.400 --> 00:39:32.320
That's that's the way I feel about that.
00:39:33.360 --> 00:39:35.920
I I love that whole soundbite.
00:39:36.079 --> 00:39:37.199
That was fantastic.
00:39:37.440 --> 00:39:41.920
Um, yeah, I don't think anyone here is setting out to eliminate the jobs.
00:39:42.159 --> 00:39:48.559
Um, I think that both of you have been pretty uh upfront about the problem statements that you found.
00:39:48.719 --> 00:39:56.559
And frankly, both of your problem statements are around loss of life and eliminating loss of life and increasing safety.
00:39:56.800 --> 00:40:03.679
Um and it's not about, I think oftentimes people immediately think, well, uh, you're gonna eliminate this job.
00:40:03.840 --> 00:40:09.599
Well, what uh more often than not, it's it's not eliminated, that person might go somewhere else to do something slightly different.
00:40:09.840 --> 00:40:16.079
Because they have a skill set and they have uh knowledge that they're they're bringing with them.
00:40:16.400 --> 00:40:18.159
Doran, did you want to pile on that?
00:40:18.239 --> 00:40:20.639
Because I multiple people were asking that question.
00:40:21.199 --> 00:40:24.000
Yeah, again, I think I'm tapping to what Jason said.
00:40:24.079 --> 00:40:26.320
Again, we're not here to replace uh the human being.
00:40:26.400 --> 00:40:33.360
We we're here to improve the the performance or you know, making their lives easier, improving the performance.
00:40:33.519 --> 00:40:38.559
Now, again, in some areas, by the way, I mentioned that on the panel as well.
00:40:38.719 --> 00:40:55.440
We see this outside of the US, uh an HR problem, a human resource problem, being that they don't have enough people uh to fill in the gaps uh from, for example, uh drivers or engineers that need to be on the uh on the on the train.
00:40:56.079 --> 00:41:00.639
And I've seen this by the way in in other areas, not only in the rail industry.
00:41:00.800 --> 00:41:08.480
So we're moving forward towards uh the future whereby there won't be enough people to accommodate uh you know these jobs.
00:41:08.719 --> 00:41:15.119
So the areas in which where they are looking at autonomy are coming from a different perspective.
00:41:15.360 --> 00:41:20.800
They don't have enough people to, or they don't have the people to accommodate these jobs.
00:41:21.199 --> 00:41:24.239
But again, I think at this day today we're very far.
00:41:24.400 --> 00:41:27.599
I mentioned AI is an evolution, not a revolution.
00:41:27.679 --> 00:41:29.920
You know, we don't have an autonomous car.
00:41:30.079 --> 00:41:38.480
When I joined my previous company, they said, Oh, and that was almost 10 years ago, next year we'll have full of autonomous cars.
00:41:38.639 --> 00:41:41.599
10 years almost have passed, we're far from it.
00:41:41.840 --> 00:41:43.440
So it's not there.
00:41:43.679 --> 00:41:45.519
Uh, industry is talking a little bit.
00:41:45.599 --> 00:41:49.599
By the way, when you talk about autonomy, PTC is autonomy.
00:41:49.760 --> 00:42:01.440
Okay, it's a part of autonomy, so it's small building blocks that help increase operational efficiency, safety, but they're here to help, they're not here to replace.
00:42:01.679 --> 00:42:04.559
Okay, so yeah, and I I I agree.
00:42:04.639 --> 00:42:06.960
And in 20, I just wanted to pile on to what you said.
00:42:07.039 --> 00:42:13.199
I mean, 2015, 2016, I remember a lot of people saying that drones were gonna replace bridge inspectors.
00:42:13.360 --> 00:42:24.159
Like, there will be no more bridge inspectors because there's drones that are gonna live in little boxes that sit on the abutments and the piers, and every once in a while they wake up and they fly around the bridge and then they go back to sleep.
00:42:24.400 --> 00:42:25.360
I haven't seen that.
00:42:25.519 --> 00:42:26.400
Maybe you guys have.
00:42:26.559 --> 00:42:28.000
I don't know what movie it's in.
00:42:28.239 --> 00:42:41.679
Um, but what we have seen, you might start with a pie in the sky idea like that, but usually uh what I've seen historically is things come from that pie in the sky idea that are actually tangible.
00:42:41.840 --> 00:42:48.800
So, for example, drones, maybe they're not doing that, but the industry's done a really good job of figuring out what drones are good for, right?
00:42:48.880 --> 00:42:50.639
And they're being used as such.
00:42:50.880 --> 00:42:53.679
And I think it's similar with almost every technology.
00:42:53.840 --> 00:43:04.400
You've got to have somebody, some guy has to come up with some crazy idea that is at the top of the tree that gets the tree shaking to drop out some pretty cool stuff, right?
00:43:04.559 --> 00:43:12.639
Yeah, and I don't even know if that was a good example or analogy using the tree, but I think you've Jason's smiling, so I think he's following what I'm saying here.
00:43:12.719 --> 00:43:25.119
So I'm gonna uh absolutely and Jason, there was a specific question for you here, and as we want to kind of wrap this up, sure railroads and state highway departments share the responsibility for ensuring grade crossings are safe.
00:43:25.280 --> 00:43:30.320
What actions have been taken to reintroduce this technology to the highway community?
00:43:30.719 --> 00:43:40.719
Yeah, so we we've actually been reaching out to some um uh some state uh DOT folks trying to figure out how to best work that in.
00:43:40.880 --> 00:43:56.880
Um it's a very regulated uh space, and so uh uh getting in there and finding the right people who can you know enact change uh is uh a problematic.
00:43:57.280 --> 00:43:59.760
Is my connection blinking a little?
00:44:00.239 --> 00:44:00.559
Nope.
00:44:00.960 --> 00:44:01.440
Okay, great.
00:44:02.239 --> 00:44:02.400
Yeah.
00:44:02.480 --> 00:44:03.920
So we've been doing reach outs.
00:44:04.000 --> 00:44:12.639
We're we're still in the early stages of that, of trying to get that sort of collaboration um uh across the uh the industries.
00:44:12.719 --> 00:44:31.360
And that's crossings, it's it's it's kind of funny in that like those are in the physical world and in this model, um, where those interactions kind of lay uh between the the roads and the and the rail, but the question is absolutely right, they are a key um figure in the maintenance of those.
00:44:33.440 --> 00:44:35.440
I had a couple more questions I do want to get.
00:44:35.760 --> 00:44:38.400
I thought this one was interesting because it was just different.
00:44:38.639 --> 00:44:54.000
Um given the accelerating emergence of disruptive technologies and the pronounced generational disparities within contemporary workforces, how critical is the deliberate preparation of organizations for technological integration?
00:44:54.239 --> 00:45:00.559
And what is the role of strategic change communication in shaping successful adoption outcomes?
00:45:01.280 --> 00:45:02.880
Anyone want to jump on that one?
00:45:05.360 --> 00:45:08.880
I just spoke, so I was trying to give Doran a bit of space there.
00:45:09.199 --> 00:45:09.760
Yeah, it's okay.
00:45:10.159 --> 00:45:10.800
Let me try.
00:45:10.880 --> 00:45:12.400
Uh I hopefully I understood.
00:45:12.480 --> 00:45:25.519
I presume the the question behind here relies to the um the change that needs to be done within an organization uh to have a successful adoption.
00:45:25.679 --> 00:45:26.480
Am I understanding?
00:45:26.559 --> 00:45:28.079
Uh exactly.
00:45:28.320 --> 00:45:33.840
I think this is a very important uh piece here in the project.
00:45:34.719 --> 00:45:45.199
Uh first of all, when you come, and I presume this relates to JSON's technology as well, when you come with uh something very uh uh new, first of all, you need to educate the market.
00:45:45.679 --> 00:45:50.480
People in general are in fear of change, nobody wants to change.
00:45:50.719 --> 00:46:00.400
Uh and you're coming with something completely different that could change their operational methods, that could uh uh change the way they work.
00:46:00.719 --> 00:46:03.039
And in general, people don't like change.
00:46:03.599 --> 00:46:10.239
So when you come with a new technology, uh very unique technology, you know, and correct me if I'm wrong, Jason.
00:46:10.320 --> 00:46:11.679
I presume you feel the same as us.
00:46:11.760 --> 00:46:17.760
You know, we need to educate the market, we need to let them understand why this is helpful for them.
00:46:17.920 --> 00:46:27.760
So there is, I think there is a gap that you need to overcome uh in promoting such technologies and to treat the organ organization as well.
00:46:28.239 --> 00:46:31.920
Uh for them to to accept such uh unique technologies.
00:46:32.079 --> 00:46:34.000
So, yeah, I think it's a great question.
00:46:34.159 --> 00:46:35.679
I think it needs to be addressed.
00:46:35.760 --> 00:46:36.880
This is what we're trying to do.
00:46:36.960 --> 00:46:39.039
I presume Jason is doing the same.
00:46:39.360 --> 00:46:41.679
Uh but yeah, this is part of the puzzle.
00:46:42.320 --> 00:46:44.559
So I I there are a couple things I want to tack on.
00:46:44.800 --> 00:46:46.239
I'm glad you asked this question.
00:46:46.400 --> 00:46:51.440
Uh uh when I saw it initially, it it just it jumped out to me and it spoke to me.
00:46:51.519 --> 00:47:06.239
And and a lot of that I think is just I've come from a technology consulting background, and this is trying to help an organization prepare itself to adopt new things is just like that's like 90% of the work.
00:47:06.559 --> 00:47:09.199
And so uh messaging is important.
00:47:09.360 --> 00:47:12.000
Uh messaging is one piece of the puzzle.
00:47:12.320 --> 00:47:29.920
Building a culture of folks at every level who spend a little bit of time looking at the horizon and seeing what's coming, and then trying kind of tagging on to what I was saying before about looking at a tool and saying, how can that tool enable me rather than how might that tool replace me?
00:47:30.159 --> 00:47:31.840
Leaning forward on that sort of thing.
00:47:32.000 --> 00:47:34.159
You want to have that sort of culture.
00:47:34.320 --> 00:47:48.239
Uh, anytime you feel that fear of, oh no, oh no, this thing might you know replace me or it might be bad, or I'm just, or it's it's change and I don't like change, that fear is a signal that you're trying to put a genie back in a bottle.
00:47:48.320 --> 00:47:52.880
And I think we all know that that isn't something that that you can actually do and that can actually work.
00:47:53.119 --> 00:48:04.159
What you're doing is you're delaying the inevitable, and then you're making yourself less prepared to adopt the disruption uh once everyone else has.
00:48:05.199 --> 00:48:09.119
Yeah, and I think that's a great way to uh close this out, frankly.
00:48:09.280 --> 00:48:15.280
I mean, I think that was a really good um 30,000-foot view just of just change, right?
00:48:15.679 --> 00:48:16.719
That's all we're talking about here.
00:48:16.880 --> 00:48:18.239
That's the whole point of this panel.
00:48:18.320 --> 00:48:23.039
That's the whole point of this podcast, is uh it's not gonna stop.
00:48:23.119 --> 00:48:27.519
And and Jason, you know, I'm 46, so similar to you.
00:48:28.079 --> 00:48:35.199
It's it's interesting your comments earlier because yeah, we are at an age where you know we grew up without the internet.
00:48:35.599 --> 00:48:40.320
Uh, we grew up without, you know, we you know, my house was full of vinyl records.
00:48:40.400 --> 00:48:42.559
Uh, and then if you just think about that, right?
00:48:42.639 --> 00:48:55.519
We watched this, it went from vinyl, 8-track was in there somewhere, but it went from vinyl to cassettes to CDs, right, to digital downloads, to now people don't even have a wall of music in their home anymore.
00:48:55.599 --> 00:48:59.280
You know, that used storage of music was a big deal.
00:48:59.519 --> 00:49:02.400
Uh now that like my kids will never see that.
00:49:02.559 --> 00:49:16.079
Um, and and you watched people above us and below us generationally grow up with email versus people like my father who had to at a late stage of life adopt to emailing, becoming right, right?
00:49:16.320 --> 00:49:23.840
But it's it's how you accept it and move on, and how you change and use it as a tool that's how you're gonna be successful.
00:49:24.239 --> 00:49:33.519
I think just to add in, I think the new generation, you know, our kids, you know, I think they accept much quick quicker quicker change compared to us.
00:49:33.760 --> 00:49:37.840
Okay, again, you said we were I'm the oldest here, I'm 52.
00:49:37.920 --> 00:49:42.239
So I so I yeah, so I'm I'm older.
00:49:42.400 --> 00:49:55.760
Uh but no, again, I see I see my my kids, I see how quickly expect they they they adopt change, and I think you know, as you know, as we go down the road, you know, younger generation will accept change uh much quicker.
00:49:55.920 --> 00:50:04.320
Uh so as I presume in a few years' time, we'll probably see, you know, change might be that it will move at a quicker pace.
00:50:05.519 --> 00:50:06.159
Yeah.
00:50:06.400 --> 00:50:10.079
And just to be clear, HR has not been notified during the productions this five days.
00:50:11.280 --> 00:50:15.599
I love how we all disclosed our ages, and I end up being the youngest one in the room.
00:50:15.760 --> 00:50:18.880
Uh that yeah, two positive things from my point.
00:50:19.440 --> 00:50:20.320
Congrats to you.
00:50:20.559 --> 00:50:20.960
All right.
00:50:21.119 --> 00:50:22.159
Well, here's the thing.
00:50:22.320 --> 00:50:25.840
Um, I think the way they close this out, there's a lot of questions here we didn't get to.
00:50:25.920 --> 00:50:34.880
There was no way we're gonna get to all of them, but there were a lot of very specific product questions to both of you that I didn't want to get to on this podcast.
00:50:34.960 --> 00:50:41.039
I wanted to stick away stay away from overly commercialization uh of the of the questions.
00:50:41.199 --> 00:50:50.880
And so if people want to ask questions or they have a question that didn't get answered, I believe, Doran, uh, they can find you at railvision.
00:50:51.199 --> 00:50:52.559
Is it railvision.io?
00:50:53.039 --> 00:50:56.239
Yeah, it's uh yeah, it's the one at railvision.io.
00:50:57.519 --> 00:50:58.320
Perfect.
00:50:58.480 --> 00:51:02.559
And that's d-o-r-o-n at railvision.io.
00:51:03.199 --> 00:51:04.800
Uh, or you can find him on the website.
00:51:04.960 --> 00:51:06.079
I think he's there as well.
00:51:06.159 --> 00:51:07.280
And then Jason, how about you?
00:51:07.440 --> 00:51:10.719
Your contact information or easy way to get a hold of you directly?
00:51:10.960 --> 00:51:18.639
Yeah, so um first off, uh, X-Track is a product and it's a X-Track X-T R A C K at uh.ai.
00:51:19.440 --> 00:51:22.480
Um, you can reach me through email there.
00:51:22.559 --> 00:51:31.519
Um, I don't I wish ish I wish it was just a first name, but it's s chin r j at xtract.ai or objectcomputing.com.
00:51:32.000 --> 00:51:36.639
The easiest way to get me is really on LinkedIn um or or something like that.
00:51:36.800 --> 00:51:41.280
So if if you've uh if you can locate me on LinkedIn, my uh messages are always open.
00:51:41.440 --> 00:51:45.119
You can just send me a message and I'd be happy to connect and chat about whatever you like.
00:51:45.440 --> 00:51:45.920
Cool.
00:51:46.159 --> 00:51:49.440
Well, thank you both for taking the time to be on the panel.
00:51:49.599 --> 00:51:55.119
Uh Doran wins the longest commute award to get to the to Indianapolis for that panel.
00:51:55.280 --> 00:51:56.480
I appreciate that, Doran.
00:51:56.639 --> 00:51:58.559
Yeah, uh appreciate it.
00:51:58.719 --> 00:51:59.440
It was really fun.
00:51:59.519 --> 00:52:00.159
I enjoyed it.
00:52:00.239 --> 00:52:00.559
Who knows?
00:52:00.639 --> 00:52:05.760
Maybe we'll have you guys back in a different way, shape, or form in the future because this isn't going away.
00:52:05.920 --> 00:52:11.199
Uh, and if I learned anything from this particular panel, is it's the wild, wild west.
00:52:11.360 --> 00:52:12.880
And it's pretty exciting.
00:52:13.039 --> 00:52:15.119
So there's lots of fun stuff to do.
00:52:15.440 --> 00:52:16.079
Absolutely.
00:52:16.239 --> 00:52:17.199
Appreciate it, Jason.
00:52:17.360 --> 00:52:17.840
Doran.
00:52:18.079 --> 00:52:18.639
Thank you very much.
00:52:18.800 --> 00:52:19.119
Thanks.
00:52:19.280 --> 00:52:19.519
Thanks.
00:52:19.679 --> 00:52:20.480
Thank you.
00:52:21.840 --> 00:52:25.920
Thank you for rolling with Arima on today's episode of Platform Chats.
00:52:26.079 --> 00:52:35.119
For further information about Arima, please visit arema.org or contact us at info at arema.org.
00:00:00.160 --> 00:00:10.000
Uh first off, and you said this earlier, Walt, and we said it during the panel too, but all all innovations, all technology, these are tools for humans.
00:00:10.640 --> 00:00:18.079
This is here to enable, to uplift, to enhance uh humans and move us forward.
00:00:18.320 --> 00:00:21.839
And um uh none of us can predict the future.
00:00:22.000 --> 00:00:27.280
Uh uh, even our predictive models can't 100% predict the future.
00:00:27.519 --> 00:00:35.920
But I'll say that if we look into recent history, so I'm I'm 45, I just uh had a birthday a couple of weeks ago.
00:00:36.079 --> 00:00:52.560
I remember um I remember before the internet, I wasn't in the workforce, but I do remember a time before the internet, I remember before smartphones, I can I can remember what life was like before and after these other sorts of real disruptive technologies.
00:00:52.640 --> 00:00:58.399
And I feel like we've said the word disruptive so much in the last 10 years that like it's almost lost its meaning.
00:00:58.560 --> 00:01:08.079
Um but in these situations, we didn't have a net loss of opportunity uh for people to work, for people to engage with the technology.
00:01:08.159 --> 00:01:11.200
What we saw was new industries, right?
00:01:11.519 --> 00:01:15.599
Uh uh new types of technology, new ways for people to work.
00:01:15.760 --> 00:01:20.000
Uh a blockchain would not exist if it weren't for the internet, right?
00:01:20.159 --> 00:01:24.159
Like there's there's brand new things that come out of these technologies.
00:01:24.400 --> 00:01:30.799
And is it the the only thing that's safe to assume, I think, is that things will change.
00:01:30.959 --> 00:01:45.359
And we all need to be ready to adapt, to see what's possible and see what tools are out there, and find ways for us to use the tools to enhance our work rather than allow the tools to replace our work.
00:01:45.519 --> 00:01:47.439
That's that's the way I feel about that.
00:01:52.640 --> 00:02:00.640
You may know the American Railway Engineering and Maintenance of Way Association, or AREMA, as the quote, keepers of the manuals.
00:02:00.719 --> 00:02:05.840
You may know them as the quote, people behind the largest annual railroad conference in North America.
00:02:06.079 --> 00:02:08.639
Heck, you may not know about AREMA at all.
00:02:08.800 --> 00:02:19.840
This podcast is designed, no pun intended, to change your view of who AREMA is and how AREMA has changed the trajectory of many railway careers over its 100-plus year history.
00:02:20.080 --> 00:02:29.759
Welcome to Platform Chats with your host, Walt Lesser, where he takes a moment to discuss the impacts AREMA has had on the very people who are proud to be called members.
00:02:29.919 --> 00:02:53.439
Are you ready to roll with AREMA?com.
00:02:53.759 --> 00:02:57.840
Hey everybody, welcome back to another episode of Platform Chats.
00:02:57.919 --> 00:03:06.639
I'm your host, Walt Blesser, and wildly enough, we are on episode number 47, 46, 48, something like that.
00:03:06.719 --> 00:03:12.400
I know there's a ghost episode somewhere in my list that never quite got recorded, but we'll get to that one.
00:03:12.560 --> 00:03:15.120
Uh sometimes people have to go do other things.
00:03:15.360 --> 00:03:20.080
Um first time to do this at Platform Chats.
00:03:20.400 --> 00:03:30.479
We have decided to host a podcast as a follow-up to our emerging technologies panel that we hosted in Arima, at Arima in Indy.
00:03:31.039 --> 00:03:39.199
As many of you know, uh the Emerging Technologies panel is actually the birth child of Platform Chats.
00:03:39.360 --> 00:03:55.520
So Platform Chats started in five or so years ago as a podcast to reach out to a whole new host of members uh as well as non-members to potentially get interest in joining Arima.
00:03:56.000 --> 00:04:07.680
And after a couple years of that, there was a request to do one live, which that then turned into the genesis of the Emerging Technologies panel where we could do just a live panel discussion.
00:04:07.840 --> 00:04:09.360
So similar but different.
00:04:09.439 --> 00:04:14.080
Uh, and our first one was in 2022, I believe, uh, in Denver.
00:04:14.400 --> 00:04:18.800
And this that was on a Wednesday morning, and we had a we had a decent sized room.
00:04:18.879 --> 00:04:36.399
Uh, it wasn't the world's largest room, but uh we then have a lot of requests to Arima that hey, people liked it, they liked interacting with the panel, they liked asking live questions and getting live answers instead of just being talked to, like many of our presentations that we do.
00:04:36.560 --> 00:04:38.480
Uh very communication 1.0.
00:04:38.560 --> 00:04:40.399
We're trying to do more of a 2.0.
00:04:40.639 --> 00:04:45.600
And uh believe it or not, this year we got moved to the uh to the Monday session.
00:04:45.839 --> 00:04:52.240
We were given a full hour and we were given a very large room, and it was it was an honor to be able to moderate this panel.
00:04:52.319 --> 00:05:00.560
I'm I'm just really, really cool thing uh that we can do this, and uh that the reception uh has been so positive.
00:05:00.639 --> 00:05:08.720
So thank you all for your feedback and for pushing Arima headquarters to uh to do things that you guys want to hear about.
00:05:08.879 --> 00:05:11.040
And so this year uh we did AI.
00:05:11.199 --> 00:05:30.800
Uh we have done in the past, we've done battery electric uh locomotives, we've done hydrogen power, uh, we've done all kinds of interesting stuff that the goal being listeners and those that are in the audience can think about it uh and not have the answers.
00:05:31.120 --> 00:05:33.199
And frankly, the people on the stage don't have all the answers.
00:05:33.279 --> 00:05:34.000
And that's the point.
00:05:34.079 --> 00:05:34.800
That's the whole point.
00:05:34.879 --> 00:05:45.040
We like to play in a space where uh no one has all the answers because we very much want to uh get people excited about trying to figure things out.
00:05:45.120 --> 00:05:59.839
Um, there was a really cool advertisement during Sunday Night Football last night that I was watching with uh something about how this is a horrible time to be a problem, uh, because there's a lot of individuals out there that are trying to quote unquote find their problem to solve it.
00:06:00.000 --> 00:06:11.920
And uh that's inspiring to people like my kids, uh, and uh hopefully a lot of people that are even listening to this to figure out what that problem is and is there a way to apply technologies of today to solve it.
00:06:12.079 --> 00:06:21.120
Um and so with that, I did bring back our two panelists, uh, both Jason and Doran are here from um Railvision and Xtrack Object Computing.
00:06:21.279 --> 00:06:27.120
I'm gonna have them do a brief uh introduction for those of you that did not sit through the panel.
00:06:27.199 --> 00:06:43.439
Um we are not gonna do the full 20-minute PowerPoints, but they'll each do a brief introduction as to kind of who they are in the world and why they're in the roles that they're in and heck, why they're even interested in trying to apply AI to um to the rail industry.
00:06:43.680 --> 00:06:50.720
And then what we're gonna do is I've got a oh man, we've got so many questions uh that were submitted through the app.
00:06:50.879 --> 00:06:53.439
And we're gonna use that as the basis of today's podcast.
00:06:53.600 --> 00:06:57.839
Again, this is very different from uh what we've done in the past.
00:06:59.600 --> 00:07:01.439
And uh we'll start with Jason.
00:07:01.680 --> 00:07:06.399
Jason, if you want to go ahead and do an introduction, that'd be great.
00:07:06.959 --> 00:07:07.600
Absolutely.
00:07:07.759 --> 00:07:14.560
Thank you so much, Walt, for inviting me here today to talk to you, and then also for the invitation uh on the panel.
00:07:14.720 --> 00:07:20.160
That was an amazing experience for me, and so I'm just I'm so thankful for all of that.
00:07:20.319 --> 00:07:22.560
Um, so yeah, my name is Jason Schindler.
00:07:22.879 --> 00:07:31.600
I've been uh uh working at Object Computing for about 13 years, and I have about 20 years in the technology and innovation industry.
00:07:31.759 --> 00:07:39.600
Um, object computing is a technology consultancy company, and that's really where my work has been for the last 20 years.
00:07:39.680 --> 00:07:50.480
So I've spent 20 years helping organizations of all shapes and sizes uh identify and adopt the emerging technologies that's gonna help move their business forward.
00:07:50.639 --> 00:07:52.560
I've done that in a number of industries.
00:07:52.720 --> 00:08:01.360
Um, for the last five years, uh I've been working with the Xtrack platform, which I think we're gonna talk about a little bit more in a moment, perhaps.
00:08:01.439 --> 00:08:05.279
Um, but the Xtrack platform is a platform specific to Rail.
00:08:05.519 --> 00:08:12.240
It brings AI and ML insights into like operational and safety efficiencies within Rail.
00:08:12.480 --> 00:08:14.720
And that's that's why I'm here today.
00:08:17.680 --> 00:08:18.639
Outstanding.
00:08:18.879 --> 00:08:23.120
And um we'll get to your products here in a second, Jason.
00:08:23.199 --> 00:08:26.079
So thank you for uh for the quick introduction.
00:08:26.319 --> 00:08:30.079
Uh Doran is our other guest who is with Rail Vision.
00:08:30.319 --> 00:08:32.159
Doran's based in uh Israel.
00:08:32.320 --> 00:08:33.519
Doran, are you on there?
00:08:33.840 --> 00:08:34.639
I hope.
00:08:34.960 --> 00:08:36.320
Can you hear me, guys?
00:08:36.639 --> 00:08:37.679
Absolutely.
00:08:38.000 --> 00:08:38.639
Good.
00:08:39.759 --> 00:08:44.879
So um again, thank you as well for joining, uh, for joining this uh podcast.
00:08:44.960 --> 00:08:48.080
And again, I had a great time with you guys uh at the panel.
00:08:48.240 --> 00:08:52.080
It was very interesting, and I'm really happy on the opportunity to continue this on the panel.
00:08:52.240 --> 00:08:56.960
I'm Doron Coadier, I'm the VP business development here at RailVision.
00:08:57.360 --> 00:09:02.159
Um actually, I came from the world of AI uh in my previous roles.
00:09:02.320 --> 00:09:06.320
I used to be working in the defense industry, the automotive, and now the rails.
00:09:06.639 --> 00:09:16.159
So I saw it in different areas uh in different perspectives, and you know, I truly believe that AI has a lot to bring to the railway industry.
00:09:16.399 --> 00:09:19.120
Uh, we're gonna talk in a minute as well about RailVision.
00:09:19.279 --> 00:09:30.320
RailVision basically provides uh very unique sensors to the rail industry in the field of uh uh operational efficiency, safety, uh, but we'll talk about it in a second.
00:09:30.399 --> 00:09:34.879
So I'm very very proud here to join uh this podcast.
00:09:36.000 --> 00:09:41.120
So if we stay on the theme of uh find your problem, which I really enjoyed that theme.
00:09:41.200 --> 00:09:43.679
I I gotta figure out who's behind that advertisement.
00:09:43.840 --> 00:09:52.480
Um so again, entrepreneurs, startups, etc., they all start with a problem statement, right?
00:09:52.639 --> 00:09:59.600
We all look at something in the industry and say, I think we can do this better or more efficient or safer.
00:10:00.080 --> 00:10:01.759
Uh I'm gonna go to you, Jason.
00:10:01.840 --> 00:10:06.720
What was your problem statement for what you're doing uh at Xtrack?
00:10:07.519 --> 00:10:18.080
It was really about finding ways to bring more modern uh data insight tools to the rail industry.
00:10:18.320 --> 00:10:54.879
And we were um we we started off with a single problem, which was identifying the root cause of unscheduled breaking events by consuming PTC device logs and building an ML uh model to read the logs, identify the root cause, and then we have a human in the loop check at the end where if an individual feels that that uh diagnosis is wrong, uh they can provide reasons for what it should be, and then we retrain the model based on that information, and then the next version of it would include those things as well.
00:10:55.039 --> 00:10:58.799
So that's that's where we got started with that very specific use case.
00:10:58.960 --> 00:11:12.720
And that was very helpful because it gave us the opportunity to work directly with folks in the rail industry instead of kind of a lot of a lot of people when they're building tools, they're also imagining problems instead of to your point finding them.
00:11:12.799 --> 00:11:21.840
Uh so they just they imagine a problem and then they go solve that, and then they find out there's no customers, uh, right, because it's not actually a problem, or they didn't imagine it correctly, or whatever.
00:11:22.080 --> 00:11:36.480
Um, a strong belief of ours is we need to partner with the people that have the problems, be able to confirm them, know how to best solve them, and how to best put our uh solution out in order to where it can be consumed and impactful.
00:11:37.360 --> 00:11:38.720
Outstanding, yes.
00:11:38.960 --> 00:11:42.639
And I I I wholeheartedly agree with that uh approach.
00:11:42.720 --> 00:11:49.919
Um, it's coming up with a problem that doesn't exist could lead you to uh not a very long-lasting startup.
00:11:50.720 --> 00:11:51.840
A lot of work, at least, yeah.
00:11:52.080 --> 00:11:52.399
Yeah.
00:11:52.639 --> 00:11:54.320
How about you at RailVision?
00:11:54.480 --> 00:11:59.840
What did you uh how did you envision the issue in the rail industry and how you were gonna apply AI?
00:12:00.480 --> 00:12:04.799
Yeah, well again, RailVision was founded to solve a very big problem.
00:12:04.960 --> 00:12:10.240
There's a huge amount of accidents worldwide uh in the rail industry.
00:12:10.480 --> 00:12:22.559
Just to give you some figure facts, in 2024 there were almost a thousand rail fatalities, almost 7,000 non-fatal rail injuries, more than a thousand derailments.
00:12:22.799 --> 00:12:26.720
So it's a big problem, not only obviously in the US, worldwide.
00:12:26.960 --> 00:12:41.200
And uh I even I even mentioned, I think, in the on the on the panel itself, that the the AAR president at the time, a few years ago, said that every three hours in the US, a person or a vehicle is struck by train.
00:12:41.360 --> 00:12:43.679
So this is what we're trying to solve.
00:12:43.919 --> 00:12:52.240
And but not only we're trying to solve it, we understand that one of the big challenges is the human cause behind this problem.
00:12:52.720 --> 00:12:58.639
Uh, either the limited eyesight of a human being, uh, the limited human resources.
00:12:58.720 --> 00:13:15.039
In some areas, they're not enough, for example, train engineers or drivers, so they work longer hours, so you know, putting more load on these uh on these individuals, harsh environment, you know, snow, fog, rain, loss of attentions.
00:13:15.200 --> 00:13:28.399
So the many, many problems coming from the human cause, and what we're trying, we're trying to assist uh and to come with the you know, we're bringing AI in order to solve this uh big problem.
00:13:31.679 --> 00:13:38.240
So it's and again, it feels very much like a tool in the toolbox, though, right?
00:13:38.399 --> 00:13:48.240
We're not trying to replace folks, we're trying hard to uh help give them necessary tools to make their jobs easier, better, faster.
00:13:48.399 --> 00:13:49.759
Is that a true statement?
00:13:50.960 --> 00:13:53.919
Yes, this is okay for more effective.
00:13:54.000 --> 00:13:55.360
Yes, it's it's a true statement.
00:13:55.519 --> 00:14:00.799
We want to actually, you know, human beings are limited in what uh the their abilities.
00:14:00.879 --> 00:14:03.360
We're trying to, you know, increase their abilities, okay?
00:14:03.519 --> 00:14:12.000
See beyond what a human person can do in order to assist them to do it in more safer, operationally uh more efficient.
00:14:12.639 --> 00:14:16.000
Uh yes, so it's it's quite a true true statement here.
00:14:16.720 --> 00:14:25.759
And Jason, you did a really good job at the panel setting the stage for before because before we go any further, there's a lot of questions here about AI.
00:14:25.919 --> 00:14:26.240
Okay.
00:14:26.480 --> 00:14:29.679
Um, and when we use that term, it's very, very broad.
00:14:30.000 --> 00:14:35.919
And when I think of AI, I think of anything from machine learning, I think of uh neural networks.
00:14:36.159 --> 00:14:42.080
Uh when people hear that term heck, I think some people just start thinking about data centers.
00:14:42.399 --> 00:14:42.960
Right.
00:14:43.360 --> 00:14:47.759
So uh and video chips and et cetera, et cetera, and and semiconductors.
00:14:47.840 --> 00:15:07.039
And so when you when we use the term AI, could you just maybe give us a 20,000-foot overview of your view of what AI is if you say you're at the cocktail party having a conversation with someone who knows nothing about it, and then let's drill down into how we're using it or applying it uh here.
00:15:07.200 --> 00:15:08.000
That'd be great.
00:15:08.399 --> 00:15:08.639
Sure.
00:15:08.799 --> 00:15:09.679
Yeah, happy to.
00:15:09.919 --> 00:15:16.399
So um AI itself is a broad, like it's it's a umbrella uh field of study, right?
00:15:16.480 --> 00:15:24.480
So it's been around since the 50s, and it's really focused on designing systems which can mimic certain aspects of human intelligence.
00:15:24.639 --> 00:15:29.279
And that like AI, artificial intelligence, that that's really as an umbrella term what it is.
00:15:29.440 --> 00:15:33.440
Um it's it's in our vernacular a lot right now.
00:15:33.600 --> 00:15:37.279
It's a very buzzword thing uh for good reason.
00:15:37.519 --> 00:15:55.600
Um uh but ML or machine learning is a specific kind of AI, and it uses uh like statistical models and algorithms to like uh predict elements of future unseen data based on training data that it's already received.
00:15:55.759 --> 00:16:11.919
Uh so an example, um, I don't recall if this was the one I used at the panel, but uh if you wanted software that could take an image and tell you whether or not the image contains a duck, you could train an ML model uh on a lot of images that do contain ducts, a lot that don't.
00:16:12.080 --> 00:16:15.840
And uh after sufficient training, you would have a model that you're pretty confident.
00:16:16.080 --> 00:16:22.000
If I show it an image it's never seen before, it would be able to correctly identify if there's a duck in that image.
00:16:22.159 --> 00:16:27.840
Um that's that's based on training data, and then it's it's that aspect of unseen data that makes it ML.
00:16:28.480 --> 00:16:39.840
That uses something called deep learning, which I won't go too far into, uh, except to say that deep learning, that's where your neural network comes in, and it's that technology that has enabled Gen AI.
00:16:39.919 --> 00:16:45.440
And I think Gen AI, when folks say AI, Gen AI is basically where they start thinking.
00:16:45.519 --> 00:16:53.360
So it's it's short for generative AI, and it's any AI that starts to create new content where content didn't exist before.
00:16:53.600 --> 00:17:08.480
And one of the big uses of that right now, if you think about ChatGPT or Gemini or whichever LLM models you're using, um, it's it's creating human speech or human language uh in order to interact with us from nothing.
00:17:08.559 --> 00:17:10.160
And so that's generative AI.
00:17:10.240 --> 00:17:18.640
And I think when folks think about AI, at least right now, a lot of them are thinking about generative AI, but it really is a big umbrella and it could mean a lot of things.
00:17:18.799 --> 00:17:19.680
Is that helpful?
00:17:20.000 --> 00:17:21.200
That is really helpful.
00:17:21.359 --> 00:17:33.599
And and yes, I think AI is at a point now where uh similar to the battery electric train conversation we had, you can relate to the problems as a consumer no differently than a large business, right?
00:17:33.680 --> 00:17:44.640
When you go to a if you're if you're deciding between an electric, fully electric vehicle versus a an ice uh internal combustion engine, which is classically what we've been driving, right?
00:17:45.039 --> 00:17:50.720
You sit down and you think, all right, well, okay, if I'm gonna go on a long road trip, what does that mean versus if I'm gonna use it around town, blah, blah, blah.
00:17:50.799 --> 00:17:52.160
You you you can kind of relate, right?
00:17:52.240 --> 00:17:55.920
It's no different than when we're thinking about that with battery electric trains.
00:17:56.160 --> 00:17:58.000
And right now, ChatGPT.
00:17:58.079 --> 00:18:09.839
Uh, my wife and I family have a trip upcoming, and believe it or not, ChatGPT has pretty much created an itinerary and it's told us when where we should go and when for the kids at what ages they are.
00:18:09.920 --> 00:18:14.000
It's it's phenomenal, but that's how consumers are are using it, right?
00:18:14.160 --> 00:18:19.759
Um, and Doran, you were famously talking about garbage in, garbage out.
00:18:19.920 --> 00:18:23.119
Uh, is this a good point where you'd want to chime in on this discussion?
00:18:23.519 --> 00:18:30.160
Yeah, I mean, um, you know, AI is a very big uh umbrella of a lot of uh things underneath it.
00:18:30.240 --> 00:18:32.960
But at the low level, the foundation is the data.
00:18:33.119 --> 00:18:37.519
Okay, and again, data is as you mentioned, garbage in, garbage out.
00:18:37.599 --> 00:18:46.799
If the data, the lowest point here, the the foundation, you provide uh low quality data, you'll get low quality AI in the end of the day.
00:18:47.039 --> 00:18:50.559
Uh, and how do we actually in our company create high quality?
00:18:50.640 --> 00:18:54.480
So, first of all, use very, very uh powerful sensors, okay?
00:18:54.640 --> 00:18:59.200
So high quality images with a lot of pixels and a lot of sensitivity.
00:19:00.000 --> 00:19:02.480
Uh, we use diverse inputs, okay.
00:19:02.559 --> 00:19:06.799
We're not using only visible cameras, we use uh thermal cameras as well.
00:19:06.960 --> 00:19:15.519
So when you have diverse inputs, you increase the quality of the data, the quantity of the data, we use a lot of data.
00:19:15.680 --> 00:19:38.240
Okay, uh we ran more than one million kilometers of rail already, collecting endless amounts of data out there in order to train our uh our different models and the diversity of conditions, we do it in daytime, nighttime, uh in Europe, in the US, in the Far East, uh you name it, in shunting yards, in main lines.
00:19:38.400 --> 00:19:56.799
So, the more you get high-quality data of image quality, diverse inputs, quantity, and diversity, you get a very, very good uh uh uh bucket of data in which you can train the machine learning and therefore provide a much higher level of uh performance of AI.
00:19:59.440 --> 00:20:04.720
I think this is a perfect time to jump in with some of the questions because I think the stage is set.
00:20:05.119 --> 00:20:11.680
Uh I think that uh there are a lot of questions about the data.
00:20:11.920 --> 00:20:17.839
And so that's why I want to chime in here because garbage in, garbage out, quality data, etc.
00:20:18.400 --> 00:20:27.200
Um, but we still live in a time where I firmly believe there never is enough data to be perfect.
00:20:27.440 --> 00:20:42.000
Uh in 2016, uh, the company I was running, uh, we had a monitoring company, we were using accelerometer data, and we were training neural networks to identify a certain type of uh uh data image or response.
00:20:42.319 --> 00:20:49.519
And man, you might think you've got that thing trained just right, and then here comes something else that's just a little bit different.
00:20:49.759 --> 00:21:00.559
And so that's one of the big questions we have here, and we are this time uh going to try to do my best to follow the number of questions that were voted with the highest number of votes.
00:21:00.640 --> 00:21:02.720
So this was the highest number of votes for real.
00:21:02.880 --> 00:21:10.240
I know that there was some folks in the audience that noticed I was not seeing all that when we were in the middle of our presentation.
00:21:10.480 --> 00:21:15.359
The questions were not floating up properly, but this question uh is for both of you.
00:21:15.519 --> 00:21:19.759
Uh, we know that we don't yet have 100% accurate AI models.
00:21:20.160 --> 00:21:23.119
What is the approach of your company to address these inaccuracies?
00:21:23.279 --> 00:21:25.119
False positives, false negatives.
00:21:25.279 --> 00:21:28.160
Um, do you keep the human in the loop for QA?
00:21:28.319 --> 00:21:32.240
This was pointed towards Jason, but I definitely think this is a question for everybody.
00:21:32.400 --> 00:21:33.599
So, Jason, you want to start?
00:21:33.680 --> 00:21:34.319
That'd be great for you.
00:21:34.559 --> 00:21:35.359
I'd be happy to, yeah.
00:21:35.440 --> 00:21:38.000
And I and it certainly is like a general purpose question.
00:21:38.079 --> 00:21:41.599
And it's it's also this really isn't specific to Rails.
00:21:41.839 --> 00:21:45.920
I think everyone, there's a lot of people asking themselves uh this question right now.
00:21:46.079 --> 00:21:54.400
And I think it's important to remember that uh AI itself um and and the models that we're building, these are these are tools.
00:21:54.559 --> 00:21:59.119
Uh there's other innovations that have been in similar spaces previously.
00:21:59.359 --> 00:22:02.319
And Uh and those things aren't always accurate either.
00:22:02.480 --> 00:22:04.240
So AI is not human, right?
00:22:04.400 --> 00:22:07.119
And also humans themselves aren't perfect.
00:22:07.359 --> 00:22:11.440
So how do we approach like potential inaccuracies?
00:22:11.599 --> 00:22:23.279
Well, the the first is kind of knowledge that you're as you're tuning a model, you're tuning it for specific uh outputs or specific uh predictions.
00:22:23.680 --> 00:22:32.720
And you can kind of you can in part of your training decide what aspects are important to you and which ones are least important to you.
00:22:32.960 --> 00:22:37.680
And so if you're thinking like, how am I going to predict everything?
00:22:37.759 --> 00:22:42.799
Uh the the the reality is you're just you're probably not, or at least not very well.
00:22:42.960 --> 00:22:47.920
And each one of those elements has a different kind of level of accuracy that's going to be associated with it.
00:22:48.000 --> 00:22:54.319
And tuning the model, you're going to optimize for for one or two things and not try to optimize for everything.
00:22:54.640 --> 00:22:57.519
So how do we and how do we address them?
00:22:57.680 --> 00:23:00.960
We do the same thing we would do with any other technology.
00:23:01.279 --> 00:23:07.680
We test the heck out of it, we test it with simulated data, we test it with real data.
00:23:07.839 --> 00:23:20.160
Um, we do like we're we're big believers in keeping the human in the loop until we come to a point where we feel really good about this model and its accuracy against the real world.
00:23:20.319 --> 00:23:28.079
And in some cases, that can be at a terminal in a system, somebody sees an output in the model and says, no, I don't think that's right, and they give it feedback.
00:23:28.240 --> 00:23:38.160
Um, in some ways, so when we talked about like crossing sight lines on the panel and uh predicting when vegetation will grow at what areas, um, it's it's visual inspections, right?
00:23:38.240 --> 00:23:50.000
It's having folks go out uh and do uh uh snapshots of what they're already doing now in order to inspect these crossings and give feedback into that model, you know, making sure that we um that we have it right.
00:23:50.240 --> 00:23:57.039
So it's it's tested a lot, um, have high standards and keep humans engaged, right?
00:23:57.119 --> 00:24:01.359
It's not like it's not a set and forget situation, and I don't think it ever should be.
00:24:01.920 --> 00:24:07.680
And I'm gonna, I know we got like 60 questions, so we're gonna try to bust through as many of these as possible.
00:24:07.839 --> 00:24:14.240
That was a great answer, and I think the reason I really wanted to start with that one is that's an overarching answer and question, right?
00:24:14.400 --> 00:24:20.240
Um, another question here was the rails, the rail signal industry historically is very risk-averse.
00:24:20.400 --> 00:24:32.880
And with the potential of AI integration into train control, how is the industry managing some of the faults that happen in AI data sets like data contamination, concept drift, model collapse, uh that sort of thing?
00:24:33.039 --> 00:24:34.960
That also ties in with another question.
00:24:35.039 --> 00:24:38.559
I'm trying to group some of these together, which was directly for Doran.
00:24:38.799 --> 00:24:41.440
Does the system interact with positive train control systems?
00:24:41.599 --> 00:24:43.119
So I would say those two are very similar.
00:24:43.200 --> 00:24:44.000
Do you, Doran?
00:24:44.079 --> 00:24:47.440
I'm gonna kick that one to you unless you want some help.
00:24:47.759 --> 00:24:49.279
Uh no, I think it's fine.
00:24:49.440 --> 00:24:53.279
Um, first, trying to address the first question again.
00:24:53.359 --> 00:25:04.640
I think by training these AI models with different types of uh I mentioned, for example, different types of uh uh cameras that we use, for example.
00:25:04.799 --> 00:25:11.039
Therefore, you you you you you use different types of inputs in order to improve the performance.
00:25:11.200 --> 00:25:18.400
So you reduce the contamination of these uh uh the these set data uh sets that you get.
00:25:18.799 --> 00:25:24.400
Um, as uh obviously Jason mentioned, there's always at this stage always a man in the loop.
00:25:24.480 --> 00:25:27.279
Okay, there's also always a human being in the loop.
00:25:27.440 --> 00:25:33.039
And by the way, uh again, we need to understand AI is not a revolution, it's an evolution.
00:25:33.200 --> 00:25:39.839
So you start from a certain point, it's far from perfect, and it keeps getting better and better.
00:25:40.079 --> 00:25:47.759
Through these uh uh you know uh steps, there's always a human being that tests all the time, but then again, this is not enough.
00:25:48.000 --> 00:25:55.119
So we're bringing in order to reduce the data contamination, we do a lot of testing, we put the human loop inside.
00:25:55.519 --> 00:25:59.440
Um uh we use different inputs.
00:25:59.519 --> 00:26:05.200
Uh, there's a a different angle to it as well, from a cyber perspective.
00:26:05.359 --> 00:26:10.799
Okay, you can contaminate data by you know tapping onto it and changing the data as well.
00:26:10.960 --> 00:26:24.640
So, this is another very important thing as you go into AI and high levels of use of AI is really increase the cybersecurity aspect as well from a contamination uh perspective.
00:26:24.960 --> 00:26:29.279
And you know, we give a lot of deep deal detail into these things as well.
00:26:29.359 --> 00:26:34.079
So we put a lot of measures to reduce uh uh these things.
00:26:34.160 --> 00:26:36.480
So I hope I you know I give a very 30,000.
00:26:37.119 --> 00:26:38.319
Yeah, Jason, go on.
00:26:38.400 --> 00:26:39.680
Uh please, if you want to add.
00:26:40.000 --> 00:26:43.279
Oh, sorry, I didn't want to interrupt that, but I did want to want to just chime in here.
00:26:43.359 --> 00:26:54.559
I'm I'm really glad, Doran, that you said um a couple times that it's like multiple dispersed data sets integrated together that really kind of level this stuff up, because I 100% agree with that.
00:26:54.640 --> 00:27:09.039
And I actually think that's more the immediate thing that we need to work through in order to make the best use of this technology is is integrating and bringing in those multiple data sets uh together to make sure we have the best understanding of reality as possible.
00:27:09.279 --> 00:27:22.480
The other thing I I want to chime in on here, and it it relates to the last question too, is we need to be so it before we start on this, we need to understand what success is and how to measure it, right?
00:27:22.640 --> 00:27:52.480
And so we need to be continuously measuring what success looks like, and then taking every iteration of that model, uh, every future version and watching that those metrics come back so that we can continue to know that we're hitting the marks, so that we don't have this like concept drift sort of problem uh where if if we keep our eye on the ball in terms of where success is, then we can make sure that all future revisions are still hitting the mark.
00:27:52.559 --> 00:27:55.279
And then if they do start to drift, we can reel them in.
00:27:55.759 --> 00:27:59.039
So uh just yes, let's get to like a yes-no.
00:27:59.359 --> 00:28:09.119
Does the does the rail vision system integrate with PTC and or the signal system as of today?
00:28:09.440 --> 00:28:10.640
As of today, no.
00:28:10.880 --> 00:28:14.880
Uh we comp I know it's a yes and no question, just a very quick one.
00:28:14.960 --> 00:28:17.039
Uh we complement PTC.
00:28:17.119 --> 00:28:26.160
Okay, PTC is it's a signal system, it's a blind system, and the whole idea is to uh you know know where the other trains are and to avoid any collision.
00:28:26.400 --> 00:28:39.519
We live in a world we live in a world that there's much more on the rails, you know, there's things can fall, you know, trees, branches, a rock, uh you know, crossing people, crossing cars, uh you name it.
00:28:39.680 --> 00:28:41.920
Okay, so we complement it.
00:28:42.000 --> 00:28:53.759
Uh at this point of time, we do not uh uh uh you know uh talk to the PTC, but we hear uh from our clients that you know it's it's you know it's a good idea.
00:28:54.000 --> 00:28:57.519
So we uh thinking over internally how to do so.
00:28:57.680 --> 00:29:06.799
Uh so yes, in in the future I believe that yes, we will be part uh of the PTC and will be an additional layer to what the PTC provides today.
00:29:07.279 --> 00:29:10.079
Jason, the your system?
00:29:10.559 --> 00:29:17.519
So we we do have a module now that reads in PTC device logs for that unscheduled breaking event root cause analysis.
00:29:17.680 --> 00:29:23.599
Uh, we don't have any current integrations that interact, especially in a real-time aspect, with PTC.
00:29:24.240 --> 00:29:25.200
Okay, fair.
00:29:25.440 --> 00:29:29.039
And uh and this is actually gonna pile onto what was just said.
00:29:29.200 --> 00:29:43.920
I I enjoyed this question of uh when implementing AI in a safety critical use case where a single failure is unacceptable, how do you comfortably conclude that a model is ready to be moved from development to production?
00:29:44.160 --> 00:29:53.279
And and I think that ties directly into what Doran said about evolution, not revolution, as well as what you said, Jason, about data sets and testing.
00:29:53.440 --> 00:29:59.200
So are there certain thresholds that can be discussed or or are we still trying to figure that out?
00:30:00.079 --> 00:30:01.359
There's always thresholds.
00:30:01.519 --> 00:30:10.559
Actually, um uh we tend to think about ML and predictions as like a yes-no thing, as a as a binary thing.
00:30:10.720 --> 00:30:16.960
Um, in reality, a lot of these models are a sliding scale of confidence, right?
00:30:17.440 --> 00:30:33.680
And if we want to be very protective, we can be more um uh conservative on when we flip the switch and say, okay, I think this needs to be uh dealt with or it needs to be inspected, or you know, just depending on the model that's being used and how that's going on.
00:30:33.920 --> 00:30:41.680
And then the other thing is is we need to be thorough in our testing and our validation, including human validation.
00:30:41.839 --> 00:30:44.319
So it kind of ties into what I was saying a moment ago.
00:30:44.559 --> 00:30:48.960
We before we even start, we need to understand what success is, right?
00:30:49.039 --> 00:30:58.000
So we don't find success, we we identify what we believe it is, and we identify how to measure it, and then we apply that measurement to all the work that we do.
00:30:58.160 --> 00:31:04.799
And so when when does when is it good to move it from development to production when we're feeling good about that success success metric?
00:31:05.039 --> 00:31:26.799
And um, and then how do we mitigate against risks uh when it comes to you know, like no no failures allowed, um keeping humans in the loop for as long as we need to, and or or any other fail-safes that are already in the system to make sure that um uh that the net result is a positive one.
00:31:27.759 --> 00:31:32.720
There's been a lot of questions in here, uh, and I'll give this to you, Doran, to kick off if you'd like.
00:31:32.880 --> 00:31:37.039
Um, there's been a lot of questions just about data ownership, litigation.
00:31:37.200 --> 00:31:40.319
I know we covered this a little briefly in the panel.
00:31:40.559 --> 00:31:48.559
Um, I think there's a you know, what are the legal ramifications of uh missing something or who owns the data?
00:31:48.880 --> 00:31:51.359
Do you guys want to talk about that at 10,000 feet?
00:31:51.519 --> 00:31:58.720
I mean, I don't know if we're there yet as an industry, but I'll I'll I'll touch it if it's uh very very briefly.
00:31:58.880 --> 00:32:02.799
Uh from our aspect, we have two sets of uh products, okay?
00:32:02.960 --> 00:32:08.240
One that sits and provides real-time alerts to the to the engineer, to the driver.
00:32:08.480 --> 00:32:08.720
Okay.
00:32:09.359 --> 00:32:15.200
Uh, and basically he gets real-time alerts on a computer that sits within the cabin itself.
00:32:15.359 --> 00:32:19.680
Okay, it is recorded as well, like a V VR on the computer itself.
00:32:19.759 --> 00:32:25.599
So basically, the data is owned completely by the uh the operator.
00:32:26.319 --> 00:32:41.039
In addition, the other product we didn't talk about too much about our products, but we have a complementary product of uh uh uh operational web application, okay, that basic basically logs all the incidents we collected.
00:32:41.200 --> 00:32:44.640
So it shoots it out to uh uh a cloud.
00:32:44.880 --> 00:32:48.240
Now a cloud here it makes life easier as well.
00:32:48.319 --> 00:32:56.960
We can put it on any cloud, either our cloud and provide the inputs to the uh client, and if necessary, he can put it on his own cloud.
00:32:57.200 --> 00:33:05.119
So basically, on the two types of uh products that we have, it's completely owned by the operator or the customer himself.
00:33:05.279 --> 00:33:08.079
So we're not even touching the data.
00:33:09.359 --> 00:33:14.559
Yeah, that that matches our experience as well in in terms of like who owns the data.
00:33:14.720 --> 00:33:20.319
Um, for the most part, whomever collects it uh generally owns it and then maintains that ownership of it.
00:33:20.400 --> 00:33:30.240
Um will pull it into our pipelines and our system in order to derive the insights, and uh the insights are generally owned by um uh the party as well.
00:33:30.400 --> 00:33:35.680
It's the uh the models and the processing in the middle uh that we focus on for ownership.
00:33:36.160 --> 00:33:44.799
And then in terms of the other, I mean, I'm not a lawyer, not really prepared to give like legal uh legal advice around that sort of stuff.
00:33:44.960 --> 00:33:57.519
Um uh, but I I do think it's it's important to know uh if there's something that's potentially dangerous uh uh that we can do something about.
00:33:57.599 --> 00:33:58.640
I'll just say that.
00:33:59.200 --> 00:34:02.799
I think this question dovetails as well for Doran.
00:34:02.960 --> 00:34:12.079
Uh, what has been the regulatory acceptance from the different countries with enhancing or augmenting what the train engineer uh uh can see?
00:34:12.719 --> 00:34:21.760
And I would even go so far as to expand that based on the conversation we've been having around uh uh regulatory acceptance.
00:34:21.920 --> 00:34:28.960
Are you seeing different countries responding to this technology in a different way?
00:34:31.039 --> 00:34:34.079
Um let me start with the US.
00:34:34.400 --> 00:34:47.280
Okay, uh the US um slowly, slowly we hear more at the MTSB talking about uh uh you know uh recommendation and start using collision avoidance systems.
00:34:47.519 --> 00:34:49.360
Okay, it's not a regulation.
00:34:49.519 --> 00:34:56.239
If it would have been a regulation in the US, I presume life our life will be much easier if it's a regulation than either product.
00:34:56.480 --> 00:35:03.920
We're we're kind of at the point that we're actually educating the market uh with our uh technology.
00:35:04.159 --> 00:35:06.239
So regulation is not there yet.
00:35:06.400 --> 00:35:17.440
Uh, but we see not only in the US, uh, the more and more huge accidents worldwide that could be avoided, but not only the big ones, even the small ones, okay?
00:35:17.599 --> 00:35:23.440
The ones that you you know you hit a switch or you have a derailment, even if there's no one hurt.
00:35:23.599 --> 00:35:40.639
Okay, so I think slowly, slowly the rail industry is starting to understand the value of technology and how they can how AI, for example, can bring a lot of uh you know a lot of uh uh value to the customer.
00:35:40.800 --> 00:35:46.400
Uh but again, there's no regulations as of yet in comparison, for example, to the automotive industry.
00:35:46.559 --> 00:35:55.360
In the automotive industry, in the automotive industry today, when you buy a car, you have ADAS products on the car, similar to what we do today.
00:35:55.440 --> 00:35:58.400
This is part of what you have because it's already regulated.
00:35:58.719 --> 00:36:04.960
I do believe it will be uh and it started already because the NTSB are pushing it.
00:36:05.039 --> 00:36:14.559
I presume the next stage, hopefully the FRA will recommend it to the operators and slowly, slowly it will move hopefully towards regulating it because there's a lot of potential.
00:36:14.800 --> 00:36:21.119
Again, I mentioned earlier how we can help the the the uh uh the the engineer.
00:36:21.280 --> 00:36:38.960
I I can an engineer can see a few hundred meters, I can or a few hundred yards, I can extend what he can see to 1.2 miles, I can show him in pitch dark like it was daylight, I can uh show them switch positions where they are, and if they're faulty.
00:36:39.119 --> 00:36:41.039
I can do so many things here.
00:36:41.199 --> 00:36:45.519
So there's a lot of things that we can add uh and assist the driver.
00:36:45.679 --> 00:37:03.360
And I think the drive behind it from a regulation perspective is all the accidents and the value that the operators are starting to understand, the return on investment they can get with such a technology by reducing the operational expenses and all the accidents they have and everything.
00:37:05.519 --> 00:37:07.440
There's been several questions.
00:37:07.679 --> 00:37:23.440
Um, and this this this can go to both of you about the and we again we covered this briefly in the panel, uh, autonomy or uh basically the elimination of the engineers operating the trains.
00:37:23.760 --> 00:37:32.320
Um do you believe, do the two of you believe that we're headed towards using this technology to eliminate jobs, support jobs?
00:37:32.559 --> 00:37:36.239
Um let's hit the elephant in the room about uh jobs.
00:37:37.280 --> 00:37:39.039
Who wants to take that one?
00:37:39.679 --> 00:37:42.320
I'll I'll start uh if that's all right, Doran.
00:37:42.639 --> 00:37:54.880
So um first off, and you said you said this earlier, Walt, and we said it during the panel too, but all all innovations, all technology, these are tools for humans.
00:37:55.039 --> 00:38:02.960
There this is here to enable, to uplift, to enhance uh humans and move us forward.
00:38:03.280 --> 00:38:06.719
And um uh none of us can predict the future.
00:38:06.960 --> 00:38:12.239
Uh uh even our predictive models can't 100% predict the future.
00:38:12.480 --> 00:38:20.800
But I'll say that if we look into recent history, so I'm I'm 45, I just uh had a birthday a couple of weeks ago.
00:38:21.039 --> 00:38:37.440
I remember um I remember before the internet, I wasn't in the workforce, but I do remember a time before the internet, I remember before smartphones, I can I can remember what life was like before and after these other sorts of real disruptive technologies.
00:38:37.519 --> 00:38:43.280
And I feel like we've said the word disruptive so much in the last 10 years that like it's almost lost its meaning.
00:38:43.440 --> 00:38:52.960
Um, but in these situations, we didn't have a net loss of opportunity uh for people to work, for people to engage with the technology.
00:38:53.119 --> 00:38:56.079
What we saw was new industries, right?
00:38:56.400 --> 00:39:00.480
Uh uh new types of technology, new ways for people to work.
00:39:00.639 --> 00:39:04.880
Uh a blockchain would not exist if it weren't for the internet, right?
00:39:05.039 --> 00:39:09.039
Like there's there's brand new things that come out of these technologies.
00:39:09.280 --> 00:39:15.679
And is it the the only thing that's safe to assume, I think, is that things will change.
00:39:15.840 --> 00:39:30.239
And we all need to be ready to adapt, to see what's possible and see what tools are out there and find ways for us to use the tools to enhance our work rather than allow the tools to replace our work.
00:39:30.400 --> 00:39:32.320
That's that's the way I feel about that.
00:39:33.360 --> 00:39:35.920
I I love that whole soundbite.
00:39:36.079 --> 00:39:37.199
That was fantastic.
00:39:37.440 --> 00:39:41.920
Um, yeah, I don't think anyone here is setting out to eliminate the jobs.
00:39:42.159 --> 00:39:48.559
Um, I think that both of you have been pretty uh upfront about the problem statements that you found.
00:39:48.719 --> 00:39:56.559
And frankly, both of your problem statements are around loss of life and eliminating loss of life and increasing safety.
00:39:56.800 --> 00:40:03.679
Um and it's not about, I think oftentimes people immediately think, well, uh, you're gonna eliminate this job.
00:40:03.840 --> 00:40:09.599
Well, what uh more often than not, it's it's not eliminated, that person might go somewhere else to do something slightly different.
00:40:09.840 --> 00:40:16.079
Because they have a skill set and they have uh knowledge that they're they're bringing with them.
00:40:16.400 --> 00:40:18.159
Doran, did you want to pile on that?
00:40:18.239 --> 00:40:20.639
Because I multiple people were asking that question.
00:40:21.199 --> 00:40:24.000
Yeah, again, I think I'm tapping to what Jason said.
00:40:24.079 --> 00:40:26.320
Again, we're not here to replace uh the human being.
00:40:26.400 --> 00:40:33.360
We we're here to improve the the performance or you know, making their lives easier, improving the performance.
00:40:33.519 --> 00:40:38.559
Now, again, in some areas, by the way, I mentioned that on the panel as well.
00:40:38.719 --> 00:40:55.440
We see this outside of the US, uh an HR problem, a human resource problem, being that they don't have enough people uh to fill in the gaps uh from, for example, uh drivers or engineers that need to be on the uh on the on the train.
00:40:56.079 --> 00:41:00.639
And I've seen this by the way in in other areas, not only in the rail industry.
00:41:00.800 --> 00:41:08.480
So we're moving forward towards uh the future whereby there won't be enough people to accommodate uh you know these jobs.
00:41:08.719 --> 00:41:15.119
So the areas in which where they are looking at autonomy are coming from a different perspective.
00:41:15.360 --> 00:41:20.800
They don't have enough people to, or they don't have the people to accommodate these jobs.
00:41:21.199 --> 00:41:24.239
But again, I think at this day today we're very far.
00:41:24.400 --> 00:41:27.599
I mentioned AI is an evolution, not a revolution.
00:41:27.679 --> 00:41:29.920
You know, we don't have an autonomous car.
00:41:30.079 --> 00:41:38.480
When I joined my previous company, they said, Oh, and that was almost 10 years ago, next year we'll have full of autonomous cars.
00:41:38.639 --> 00:41:41.599
10 years almost have passed, we're far from it.
00:41:41.840 --> 00:41:43.440
So it's not there.
00:41:43.679 --> 00:41:45.519
Uh, industry is talking a little bit.
00:41:45.599 --> 00:41:49.599
By the way, when you talk about autonomy, PTC is autonomy.
00:41:49.760 --> 00:42:01.440
Okay, it's a part of autonomy, so it's small building blocks that help increase operational efficiency, safety, but they're here to help, they're not here to replace.
00:42:01.679 --> 00:42:04.559
Okay, so yeah, and I I I agree.
00:42:04.639 --> 00:42:06.960
And in 20, I just wanted to pile on to what you said.
00:42:07.039 --> 00:42:13.199
I mean, 2015, 2016, I remember a lot of people saying that drones were gonna replace bridge inspectors.
00:42:13.360 --> 00:42:24.159
Like, there will be no more bridge inspectors because there's drones that are gonna live in little boxes that sit on the abutments and the piers, and every once in a while they wake up and they fly around the bridge and then they go back to sleep.
00:42:24.400 --> 00:42:25.360
I haven't seen that.
00:42:25.519 --> 00:42:26.400
Maybe you guys have.
00:42:26.559 --> 00:42:28.000
I don't know what movie it's in.
00:42:28.239 --> 00:42:41.679
Um, but what we have seen, you might start with a pie in the sky idea like that, but usually uh what I've seen historically is things come from that pie in the sky idea that are actually tangible.
00:42:41.840 --> 00:42:48.800
So, for example, drones, maybe they're not doing that, but the industry's done a really good job of figuring out what drones are good for, right?
00:42:48.880 --> 00:42:50.639
And they're being used as such.
00:42:50.880 --> 00:42:53.679
And I think it's similar with almost every technology.
00:42:53.840 --> 00:43:04.400
You've got to have somebody, some guy has to come up with some crazy idea that is at the top of the tree that gets the tree shaking to drop out some pretty cool stuff, right?
00:43:04.559 --> 00:43:12.639
Yeah, and I don't even know if that was a good example or analogy using the tree, but I think you've Jason's smiling, so I think he's following what I'm saying here.
00:43:12.719 --> 00:43:25.119
So I'm gonna uh absolutely and Jason, there was a specific question for you here, and as we want to kind of wrap this up, sure railroads and state highway departments share the responsibility for ensuring grade crossings are safe.
00:43:25.280 --> 00:43:30.320
What actions have been taken to reintroduce this technology to the highway community?
00:43:30.719 --> 00:43:40.719
Yeah, so we we've actually been reaching out to some um uh some state uh DOT folks trying to figure out how to best work that in.
00:43:40.880 --> 00:43:56.880
Um it's a very regulated uh space, and so uh uh getting in there and finding the right people who can you know enact change uh is uh a problematic.
00:43:57.280 --> 00:43:59.760
Is my connection blinking a little?
00:44:00.239 --> 00:44:00.559
Nope.
00:44:00.960 --> 00:44:01.440
Okay, great.
00:44:02.239 --> 00:44:02.400
Yeah.
00:44:02.480 --> 00:44:03.920
So we've been doing reach outs.
00:44:04.000 --> 00:44:12.639
We're we're still in the early stages of that, of trying to get that sort of collaboration um uh across the uh the industries.
00:44:12.719 --> 00:44:31.360
And that's crossings, it's it's it's kind of funny in that like those are in the physical world and in this model, um, where those interactions kind of lay uh between the the roads and the and the rail, but the question is absolutely right, they are a key um figure in the maintenance of those.
00:44:33.440 --> 00:44:35.440
I had a couple more questions I do want to get.
00:44:35.760 --> 00:44:38.400
I thought this one was interesting because it was just different.
00:44:38.639 --> 00:44:54.000
Um given the accelerating emergence of disruptive technologies and the pronounced generational disparities within contemporary workforces, how critical is the deliberate preparation of organizations for technological integration?
00:44:54.239 --> 00:45:00.559
And what is the role of strategic change communication in shaping successful adoption outcomes?
00:45:01.280 --> 00:45:02.880
Anyone want to jump on that one?
00:45:05.360 --> 00:45:08.880
I just spoke, so I was trying to give Doran a bit of space there.
00:45:09.199 --> 00:45:09.760
Yeah, it's okay.
00:45:10.159 --> 00:45:10.800
Let me try.
00:45:10.880 --> 00:45:12.400
Uh I hopefully I understood.
00:45:12.480 --> 00:45:25.519
I presume the the question behind here relies to the um the change that needs to be done within an organization uh to have a successful adoption.
00:45:25.679 --> 00:45:26.480
Am I understanding?
00:45:26.559 --> 00:45:28.079
Uh exactly.
00:45:28.320 --> 00:45:33.840
I think this is a very important uh piece here in the project.
00:45:34.719 --> 00:45:45.199
Uh first of all, when you come, and I presume this relates to JSON's technology as well, when you come with uh something very uh uh new, first of all, you need to educate the market.
00:45:45.679 --> 00:45:50.480
People in general are in fear of change, nobody wants to change.
00:45:50.719 --> 00:46:00.400
Uh and you're coming with something completely different that could change their operational methods, that could uh uh change the way they work.
00:46:00.719 --> 00:46:03.039
And in general, people don't like change.
00:46:03.599 --> 00:46:10.239
So when you come with a new technology, uh very unique technology, you know, and correct me if I'm wrong, Jason.
00:46:10.320 --> 00:46:11.679
I presume you feel the same as us.
00:46:11.760 --> 00:46:17.760
You know, we need to educate the market, we need to let them understand why this is helpful for them.
00:46:17.920 --> 00:46:27.760
So there is, I think there is a gap that you need to overcome uh in promoting such technologies and to treat the organ organization as well.
00:46:28.239 --> 00:46:31.920
Uh for them to to accept such uh unique technologies.
00:46:32.079 --> 00:46:34.000
So, yeah, I think it's a great question.
00:46:34.159 --> 00:46:35.679
I think it needs to be addressed.
00:46:35.760 --> 00:46:36.880
This is what we're trying to do.
00:46:36.960 --> 00:46:39.039
I presume Jason is doing the same.
00:46:39.360 --> 00:46:41.679
Uh but yeah, this is part of the puzzle.
00:46:42.320 --> 00:46:44.559
So I I there are a couple things I want to tack on.
00:46:44.800 --> 00:46:46.239
I'm glad you asked this question.
00:46:46.400 --> 00:46:51.440
Uh uh when I saw it initially, it it just it jumped out to me and it spoke to me.
00:46:51.519 --> 00:47:06.239
And and a lot of that I think is just I've come from a technology consulting background, and this is trying to help an organization prepare itself to adopt new things is just like that's like 90% of the work.
00:47:06.559 --> 00:47:09.199
And so uh messaging is important.
00:47:09.360 --> 00:47:12.000
Uh messaging is one piece of the puzzle.
00:47:12.320 --> 00:47:29.920
Building a culture of folks at every level who spend a little bit of time looking at the horizon and seeing what's coming, and then trying kind of tagging on to what I was saying before about looking at a tool and saying, how can that tool enable me rather than how might that tool replace me?
00:47:30.159 --> 00:47:31.840
Leaning forward on that sort of thing.
00:47:32.000 --> 00:47:34.159
You want to have that sort of culture.
00:47:34.320 --> 00:47:48.239
Uh, anytime you feel that fear of, oh no, oh no, this thing might you know replace me or it might be bad, or I'm just, or it's it's change and I don't like change, that fear is a signal that you're trying to put a genie back in a bottle.
00:47:48.320 --> 00:47:52.880
And I think we all know that that isn't something that that you can actually do and that can actually work.
00:47:53.119 --> 00:48:04.159
What you're doing is you're delaying the inevitable, and then you're making yourself less prepared to adopt the disruption uh once everyone else has.
00:48:05.199 --> 00:48:09.119
Yeah, and I think that's a great way to uh close this out, frankly.
00:48:09.280 --> 00:48:15.280
I mean, I think that was a really good um 30,000-foot view just of just change, right?
00:48:15.679 --> 00:48:16.719
That's all we're talking about here.
00:48:16.880 --> 00:48:18.239
That's the whole point of this panel.
00:48:18.320 --> 00:48:23.039
That's the whole point of this podcast, is uh it's not gonna stop.
00:48:23.119 --> 00:48:27.519
And and Jason, you know, I'm 46, so similar to you.
00:48:28.079 --> 00:48:35.199
It's it's interesting your comments earlier because yeah, we are at an age where you know we grew up without the internet.
00:48:35.599 --> 00:48:40.320
Uh, we grew up without, you know, we you know, my house was full of vinyl records.
00:48:40.400 --> 00:48:42.559
Uh, and then if you just think about that, right?
00:48:42.639 --> 00:48:55.519
We watched this, it went from vinyl, 8-track was in there somewhere, but it went from vinyl to cassettes to CDs, right, to digital downloads, to now people don't even have a wall of music in their home anymore.
00:48:55.599 --> 00:48:59.280
You know, that used storage of music was a big deal.
00:48:59.519 --> 00:49:02.400
Uh now that like my kids will never see that.
00:49:02.559 --> 00:49:16.079
Um, and and you watched people above us and below us generationally grow up with email versus people like my father who had to at a late stage of life adopt to emailing, becoming right, right?
00:49:16.320 --> 00:49:23.840
But it's it's how you accept it and move on, and how you change and use it as a tool that's how you're gonna be successful.
00:49:24.239 --> 00:49:33.519
I think just to add in, I think the new generation, you know, our kids, you know, I think they accept much quick quicker quicker change compared to us.
00:49:33.760 --> 00:49:37.840
Okay, again, you said we were I'm the oldest here, I'm 52.
00:49:37.920 --> 00:49:42.239
So I so I yeah, so I'm I'm older.
00:49:42.400 --> 00:49:55.760
Uh but no, again, I see I see my my kids, I see how quickly expect they they they adopt change, and I think you know, as you know, as we go down the road, you know, younger generation will accept change uh much quicker.
00:49:55.920 --> 00:50:04.320
Uh so as I presume in a few years' time, we'll probably see, you know, change might be that it will move at a quicker pace.
00:50:05.519 --> 00:50:06.159
Yeah.
00:50:06.400 --> 00:50:10.079
And just to be clear, HR has not been notified during the productions this five days.
00:50:11.280 --> 00:50:15.599
I love how we all disclosed our ages, and I end up being the youngest one in the room.
00:50:15.760 --> 00:50:18.880
Uh that yeah, two positive things from my point.
00:50:19.440 --> 00:50:20.320
Congrats to you.
00:50:20.559 --> 00:50:20.960
All right.
00:50:21.119 --> 00:50:22.159
Well, here's the thing.
00:50:22.320 --> 00:50:25.840
Um, I think the way they close this out, there's a lot of questions here we didn't get to.
00:50:25.920 --> 00:50:34.880
There was no way we're gonna get to all of them, but there were a lot of very specific product questions to both of you that I didn't want to get to on this podcast.
00:50:34.960 --> 00:50:41.039
I wanted to stick away stay away from overly commercialization uh of the of the questions.
00:50:41.199 --> 00:50:50.880
And so if people want to ask questions or they have a question that didn't get answered, I believe, Doran, uh, they can find you at railvision.
00:50:51.199 --> 00:50:52.559
Is it railvision.io?
00:50:53.039 --> 00:50:56.239
Yeah, it's uh yeah, it's the one at railvision.io.
00:50:57.519 --> 00:50:58.320
Perfect.
00:50:58.480 --> 00:51:02.559
And that's d-o-r-o-n at railvision.io.
00:51:03.199 --> 00:51:04.800
Uh, or you can find him on the website.
00:51:04.960 --> 00:51:06.079
I think he's there as well.
00:51:06.159 --> 00:51:07.280
And then Jason, how about you?
00:51:07.440 --> 00:51:10.719
Your contact information or easy way to get a hold of you directly?
00:51:10.960 --> 00:51:18.639
Yeah, so um first off, uh, X-Track is a product and it's a X-Track X-T R A C K at uh.ai.
00:51:19.440 --> 00:51:22.480
Um, you can reach me through email there.
00:51:22.559 --> 00:51:31.519
Um, I don't I wish ish I wish it was just a first name, but it's s chin r j at xtract.ai or objectcomputing.com.
00:51:32.000 --> 00:51:36.639
The easiest way to get me is really on LinkedIn um or or something like that.
00:51:36.800 --> 00:51:41.280
So if if you've uh if you can locate me on LinkedIn, my uh messages are always open.
00:51:41.440 --> 00:51:45.119
You can just send me a message and I'd be happy to connect and chat about whatever you like.
00:51:45.440 --> 00:51:45.920
Cool.
00:51:46.159 --> 00:51:49.440
Well, thank you both for taking the time to be on the panel.
00:51:49.599 --> 00:51:55.119
Uh Doran wins the longest commute award to get to the to Indianapolis for that panel.
00:51:55.280 --> 00:51:56.480
I appreciate that, Doran.
00:51:56.639 --> 00:51:58.559
Yeah, uh appreciate it.
00:51:58.719 --> 00:51:59.440
It was really fun.
00:51:59.519 --> 00:52:00.159
I enjoyed it.
00:52:00.239 --> 00:52:00.559
Who knows?
00:52:00.639 --> 00:52:05.760
Maybe we'll have you guys back in a different way, shape, or form in the future because this isn't going away.
00:52:05.920 --> 00:52:11.199
Uh, and if I learned anything from this particular panel, is it's the wild, wild west.
00:52:11.360 --> 00:52:12.880
And it's pretty exciting.
00:52:13.039 --> 00:52:15.119
So there's lots of fun stuff to do.
00:52:15.440 --> 00:52:16.079
Absolutely.
00:52:16.239 --> 00:52:17.199
Appreciate it, Jason.
00:52:17.360 --> 00:52:17.840
Doran.
00:52:18.079 --> 00:52:18.639
Thank you very much.
00:52:18.800 --> 00:52:19.119
Thanks.
00:52:19.280 --> 00:52:19.519
Thanks.
00:52:19.679 --> 00:52:20.480
Thank you.
00:52:21.840 --> 00:52:25.920
Thank you for rolling with Arima on today's episode of Platform Chats.
00:52:26.079 --> 00:52:35.119
For further information about Arima, please visit arema.org or contact us at info at arema.org.
