Proceed, Pause, Escalate: Balancing AI Autonomy on the Factory Floor

WEBVTT
00:00:02.008 --> 00:00:07.087
Well, hello and welcome to Visions, a machine vision and automation solutions podcast.
00:00:07.627 --> 00:00:11.307
I'm your host, Jim Tatum, Senior Editor of Vision Systems Design,
00:00:11.580 --> 00:00:16.561
and Visions is an Endeavor business media production from your friends at Vision Systems Design.
00:00:17.205 --> 00:00:21.637
Here you'll find the latest on everything from end-user machine vision solutions
00:00:21.637 --> 00:00:26.656
to trends, developments, and perspectives on all things machine vision and imaging.
00:00:27.573 --> 00:00:30.667
Whether you've been working in the industry for a while or you're just starting
00:00:30.667 --> 00:00:34.867
to take a closer look at it, this podcast is designed to grow your knowledge
00:00:34.867 --> 00:00:38.956
and bring greater focus to your understanding of the imaging and machine vision industry.
00:00:39.630 --> 00:00:41.232
And now, on to our show.
00:00:42.637 --> 00:00:47.287
Well, hi everybody and welcome to Visions. Today we're going to delve into AI's
00:00:47.287 --> 00:00:49.183
expanding role in manufacturing.
00:00:49.901 --> 00:00:54.477
As AI takes on a bigger role in such applications as machine vision inspection,
00:00:54.477 --> 00:00:59.307
for example, one question keeps coming up when should the system make the call
00:00:59.307 --> 00:01:01.089
and when should a human step in,
00:01:01.867 --> 00:01:08.322
today we are pleased to welcome back Dijam Panigrahi co-founder and COO of GridRaster,
00:01:08.856 --> 00:01:14.434
a company that specializes in next generation spatial AI and extended reality technology,
00:01:15.143 --> 00:01:20.737
Dijon discusses a three-tiered approach to this balancing act between autonomous
00:01:20.737 --> 00:01:26.634
decision making inspection reliability regulatory compliance and operator trust,
00:01:27.231 --> 00:01:33.518
and why the future of ai on the factory floor may well depend as much on governance as technology,
00:01:34.264 --> 00:01:39.257
welcome back, Dijam, always a pleasure to have you here but if you could first off
00:01:39.257 --> 00:01:44.602
just kind of give me a brief overview describing the three-tier framework and how you came to it.
00:01:46.200 --> 00:01:49.358
Now, this has actually evolved.
00:01:50.618 --> 00:01:56.398
Over quite a bit of time, and we can see that struggle with our various customers
00:01:56.398 --> 00:02:02.034
that we are working, both on the industrial manufacturing and also defense, right?
00:02:03.238 --> 00:02:09.378
So it's always been a case, what is that you can allow an AI to make decisions
00:02:09.378 --> 00:02:10.838
and move ahead on its own?
00:02:10.838 --> 00:02:17.848
Make you know and what is that okay i want the human to be in the loop right
00:02:17.848 --> 00:02:19.888
many many many decisions where,
00:02:20.404 --> 00:02:26.262
you're looking at human ultimately kind of signing off or making the final decisions in many ways right,
00:02:26.813 --> 00:02:31.668
and there are there are certain processes which is off limits like you know
00:02:31.668 --> 00:02:35.118
ai is not at all trusted there right,
00:02:35.658 --> 00:02:39.298
and interestingly you could you could see some of this kind of play out very
00:02:39.298 --> 00:02:44.738
openly as well like you saw the recent disagreement between anthropic and,
00:02:45.258 --> 00:02:50.378
the DOD in many ways kind of falls into that bracket where anthropic i mean
00:02:50.378 --> 00:02:55.701
they on one side kind of agree what the i can be used for but they're somehow not comfortable,
00:02:56.316 --> 00:03:00.608
taking the ai which kind of does your maybe the automation of your supply chain
00:03:00.608 --> 00:03:05.858
and other stuff but when it goes to the making real battle decisions,
00:03:06.178 --> 00:03:08.908
you know, that's where they are not comfortable, right?
00:03:08.908 --> 00:03:13.348
So, you know, we'll see more of this kind of play out over a time period.
00:03:13.348 --> 00:03:17.608
But fundamentally for us, right, because we have a vision systems,
00:03:17.608 --> 00:03:19.395
because we are in many ways.
00:03:20.421 --> 00:03:27.161
Bringing the ai from like the back office to actual real environment which consequences with the,
00:03:27.621 --> 00:03:34.646
robots operating with operator around with real implications of something not
00:03:34.948 --> 00:03:39.047
going right there is real implications there so,
00:03:39.841 --> 00:03:46.701
largely our play and i think where large chunk of our customers are more comfortable
00:03:46.701 --> 00:03:49.731
on the tire where you have the human in the loop
00:03:50.171 --> 00:03:53.961
you know there are certain key decisions where the human have to make a decision
00:03:53.961 --> 00:04:00.670
right but 80 of the work could be kind of done by ai but there is the human in the loop right,
00:04:01.421 --> 00:04:06.056
but many of the i completely understand that's many of the back offices stuff
00:04:06.230 --> 00:04:10.201
that is kind of happening like in terms of recording your data,
00:04:10.781 --> 00:04:14.741
you know, taking some recommendation out of it, and then, you know.
00:04:15.561 --> 00:04:17.581
It's more on the back office kind of a thing.
00:04:17.581 --> 00:04:22.031
We see a lot of the fully autonomous AI agents kind of operating.
00:04:22.031 --> 00:04:27.161
So that's the spectrum, which based on which, you know, this whole three-tier,
00:04:27.161 --> 00:04:29.151
and we understand why it is there, right?
00:04:29.151 --> 00:04:35.641
Okay. So to follow that path, say in a machine vision inspection environment, what,
00:04:36.397 --> 00:04:44.081
kind of decisions should AI be allowed to make autonomously and which should
00:04:44.081 --> 00:04:48.355
always require human aerospace specifically?
00:04:48.995 --> 00:04:49.785
Sure, sure.
00:04:50.761 --> 00:04:54.591
Actually, what has happened, like, I mean, when you look at,
00:04:54.591 --> 00:04:58.701
like, let's take the case of we are doing an inspection of a wing of an aircraft.
00:04:58.701 --> 00:05:01.121
Just take just the example of that, right?
00:05:01.901 --> 00:05:06.511
Or even a flight line when there is an inspection that is happening after that
00:05:06.511 --> 00:05:09.301
you're making a decision the aircraft is going to kind of take off.
00:05:10.518 --> 00:05:15.144
So the moment I say that I let the AI make all the decisions,
00:05:15.144 --> 00:05:20.221
in a way, I'm kind of confident that my AI will do it 100% right.
00:05:21.034 --> 00:05:29.154
So now the question is, do we believe that it has reached that level that it
00:05:29.154 --> 00:05:31.284
can get everything 100% right?
00:05:31.614 --> 00:05:35.714
Right because end of the day you are kind of,
00:05:36.474 --> 00:05:42.884
you're training your AI models to be able to make decisions but is there scenarios
00:05:42.884 --> 00:05:50.894
where the ai model may not be kind of exposed right and and this is where the the uniqueness of,
00:05:51.574 --> 00:05:56.944
in human being there to see certain deviations and you know flag certain things
00:05:56.944 --> 00:06:01.939
which maybe AI is not yet exposed to and may not be able to make those decisions.
00:06:02.862 --> 00:06:06.633
So when it comes to those scenarios, what, like, for example,
00:06:06.633 --> 00:06:10.523
like when I am kind of making a determination during the inspection that what
00:06:10.523 --> 00:06:13.844
kind of and repair that a particular defect requires,
00:06:14.343 --> 00:06:20.793
I may use a AI agent to go and basically I take a picture, give it in my age and just like, OK,
00:06:21.393 --> 00:06:24.418
I can measure correctly. Yes, I am confident it will do it.
00:06:24.674 --> 00:06:28.763
It can localize in the XYZ coordinate where it is and all of that.
00:06:28.763 --> 00:06:33.813
What's the concentration of the various defects that are there and make a decision
00:06:34.133 --> 00:06:39.302
on, you know, maybe classifying them to be what defect type that we need to do.
00:06:39.825 --> 00:06:44.393
And then basically allowing let's suppose a robot or somebody needs to act on
00:06:44.393 --> 00:06:49.143
that and activating the robot to repair those kind of defects the moment you
00:06:49.143 --> 00:06:54.378
bring in that translation till providing that to a robot or,
00:06:54.854 --> 00:06:57.930
even providing the judgment to the human in,
00:06:58.633 --> 00:07:01.833
in some way there is needs to be a verification that happens.
00:07:02.853 --> 00:07:07.473
By the human and then there's another scenario and and that's why like If you're
00:07:07.473 --> 00:07:12.785
aware, like FAA, even today, doesn't allow a fully autonomous AI inspect. They do not agree.
00:07:13.653 --> 00:07:16.673
It is still from even, I mean, there are many elements to it,
00:07:17.253 --> 00:07:19.433
why the AI governance is this way.
00:07:19.433 --> 00:07:22.653
And there is a very good reason for it. I mean, for example,
00:07:22.993 --> 00:07:28.423
from a regulatory angle, they haven't seen where AI could be all-encompassing
00:07:28.423 --> 00:07:31.267
and be able to do it all. No, they haven't seen that.
00:07:31.435 --> 00:07:36.853
And so finally, even for your aircraft to take off, the human has to sign off
00:07:36.853 --> 00:07:39.461
on that, which means that you could,
00:07:40.024 --> 00:07:45.593
in many ways, you're using AI as a supporting tool, like, but that support is
00:07:45.593 --> 00:07:49.843
a big support, you know, is maybe does your 95% of the job.
00:07:49.843 --> 00:07:55.634
But finally, that 5% still has to be done by the human to get this confidence, right?
00:07:56.993 --> 00:08:01.353
So you have a regulatory element that you can play into the role of the governance
00:08:01.353 --> 00:08:06.023
of what is feasible, not feasible, and particularly challenging becomes,
00:08:06.023 --> 00:08:07.451
I mean, inspection is still challenging.
00:08:08.131 --> 00:08:12.219
It's a very clear cut, and I don't know if you're tracking what's happening
00:08:12.219 --> 00:08:16.619
in different parts of the world, but if you look at, I think, in general,
00:08:17.299 --> 00:08:21.929
in Europe, they recently put some deadlines around, I think it was maybe here
00:08:21.929 --> 00:08:24.878
sometime in August or something, like in 2026, I think it is August,
00:08:25.517 --> 00:08:31.159
that, okay, before you can have these AI agents kind of play out in those operational
00:08:31.159 --> 00:08:32.930
manufacturing environments,
00:08:33.919 --> 00:08:38.259
there has to be a foolproof study with the risk assessment.
00:08:38.259 --> 00:08:42.569
And all of that to be done before you can use it. In fact, inspection has been
00:08:42.569 --> 00:08:46.348
kind of called out as one of the use cases which is considered a high-risk.
00:08:46.684 --> 00:08:51.229
Inspection, quality control, and all of that, and they have to prove that with
00:08:51.229 --> 00:08:56.419
the data to be able to use the agentic AIs in a good way.
00:08:58.359 --> 00:09:03.035
So, and for us, and I would touch upon one other thing is in general,
00:09:03.035 --> 00:09:07.570
if you look at the industrial or manufacturing environments,
00:09:08.011 --> 00:09:11.535
which is very different from like a software and all like, and,
00:09:12.195 --> 00:09:16.305
it's something that we have been able to better understand over the last two,
00:09:16.305 --> 00:09:21.432
three years, as we kind of push this AI more onto the factory floors,
00:09:22.041 --> 00:09:25.595
is you see how the systems and build out.
00:09:25.595 --> 00:09:31.555
Because from an effective AI governance point of view, it's not that we fit
00:09:31.555 --> 00:09:35.855
in something outside, but you are trying to build it into the overall system
00:09:35.855 --> 00:09:38.585
to be able to really be meaningful.
00:09:38.585 --> 00:09:41.305
We define the guardrails and those things have to be defined.
00:09:41.305 --> 00:09:46.495
There are very, so many system interplays happen and you may have different
00:09:46.495 --> 00:09:48.085
agents trying to do different things.
00:09:48.085 --> 00:09:52.265
And all of a sudden you have so many agents that you're trying to kind of manage
00:09:52.265 --> 00:09:56.315
and it just becomes way too much. I don't think so. We have handled those many
00:09:56.715 --> 00:09:58.106
with that degree of confidence.
00:09:58.681 --> 00:10:02.775
But one of the biggest play that we have seen in those environments has been
00:10:02.775 --> 00:10:08.645
that is usually, like if I have to broadly put, there are one that are operationally
00:10:08.645 --> 00:10:11.077
kind of focused technology elements.
00:10:11.431 --> 00:10:14.715
And there is one which is more what they would normally call IT,
00:10:14.715 --> 00:10:19.152
the ERPs and other systems that are there in all these places, right?
00:10:19.814 --> 00:10:25.463
So there is an interplay that happens, right, between the operational setup and the ERP system.
00:10:26.368 --> 00:10:29.567
Right now, if you see, and if you actually dig a little deeper.
00:10:30.124 --> 00:10:35.073
All the system operational systems, okay, the way the IT is integrated with
00:10:35.073 --> 00:10:38.210
the operational system is IT is almost like a reporting tool.
00:10:38.657 --> 00:10:42.843
There is not a bidirectional operational exchange that really kind of happening
00:10:42.843 --> 00:10:44.457
in those environments, right?
00:10:44.886 --> 00:10:50.293
In order to really allow the agentic AIs to make decisions and all that,
00:10:50.293 --> 00:10:55.193
even if you define the guardrails, then many of those data points that comes
00:10:55.193 --> 00:10:58.912
from the ERP systems or the IT infrastructure overall
00:10:59.301 --> 00:11:04.033
needs to be taken into consideration when you are making some decisions to automate.
00:11:04.033 --> 00:11:07.223
And I don't think so those infrastructure exist today, right?
00:11:07.391 --> 00:11:12.803
So there is an infrastructural level kind of work that needs to be done to be
00:11:12.803 --> 00:11:18.023
getting prepared for like better, even the AI governance, right?
00:11:18.363 --> 00:11:22.083
It's a very interesting intrigue. I mean, there are different pressures that
00:11:22.083 --> 00:11:26.543
are going on, but we absolutely see the immense value of it.
00:11:26.543 --> 00:11:31.333
For example, for us, for example, like you want to kind of plan a robot to kind
00:11:31.333 --> 00:11:33.803
of do a repair of a certain object.
00:11:33.803 --> 00:11:38.273
Right okay and you know and it's a mobile robot you're trying to allow them
00:11:38.273 --> 00:11:43.073
to do a path planning of it right and we have now reached a reach a scenario
00:11:43.073 --> 00:11:46.102
where i can just put on a headset without any coding no path
00:11:46.485 --> 00:11:49.451
planning anything needed you just wear a headset walk around it
00:11:49.863 --> 00:11:54.623
and you know you say okay this is what it is and this is my target okay and
00:11:54.623 --> 00:11:59.377
you can let the robot then figure out everything else on its own right we come to those stages.
00:11:59.906 --> 00:12:04.358
But there is inherent risks that are there. That's why we are forced to have
00:12:04.671 --> 00:12:08.017
human in the loop making some of the final decisions around that.
00:12:08.899 --> 00:12:13.671
Will, over time, I think it's also a question. I think the speed at which AI
00:12:13.671 --> 00:12:18.701
is going, I don't think so. We are able to kind of assimilate and keep up with it.
00:12:19.101 --> 00:12:23.941
And I think in general, like when you have humans working in those environments,
00:12:23.941 --> 00:12:26.581
firstly, you have to develop trust with the system.
00:12:26.941 --> 00:12:31.452
And usually the trust with the system comes over a time period of use.
00:12:31.870 --> 00:12:36.421
And I'm sure as a matter of time, we'll begin to kind of develop those kind of trust.
00:12:36.792 --> 00:12:42.401
And I'm sure the AI outcomes would, and which AI outcomes are already validating
00:12:42.401 --> 00:12:45.386
anyway, but just that we need to get comfortable. Okay.
00:12:46.175 --> 00:12:51.081
It looks like you've at one point said that too much oversight can be as damaging
00:12:51.081 --> 00:12:52.887
as too little oversight. Yeah.
00:12:53.456 --> 00:12:58.821
That being the case, How does a manufacturer avoid alert fatigue when AI-driven
00:12:58.821 --> 00:13:02.772
inspection systems are generating review requests on the production line?
00:13:04.621 --> 00:13:05.245
How do you deal with that?
00:13:06.029 --> 00:13:09.282
Yeah, I think that's a real problem that we see too.
00:13:09.602 --> 00:13:14.652
I think in many of the systems, actually, we have become the bottlenecks of
00:13:14.652 --> 00:13:19.320
the humans or the operators that they are there, you know, kind of in the bottlenecks.
00:13:19.710 --> 00:13:23.919
We are not able to keep up with what AI is able to provide us.
00:13:24.418 --> 00:13:30.472
I think in that, like, the way we see it, right, we believe that there are certain
00:13:30.472 --> 00:13:32.752
things I don't think so humans should be reviewing it.
00:13:32.752 --> 00:13:39.782
I think AI has gotten that good at that, like this whole three-tire thing that I'm talking about.
00:13:39.782 --> 00:13:44.932
Now, the three-tire approach has been kind of designed from angle right now,
00:13:44.932 --> 00:13:50.087
which is, okay, what we have conceived that, okay, this is the reason, right?
00:13:50.610 --> 00:13:55.223
But if you break it down, even within that, a human in the loop kind of processes
00:13:55.619 --> 00:14:00.256
in which we are still having at many of the places the human to make decisions,
00:14:00.726 --> 00:14:04.942
I think over a time period, you will see that some of those processes,
00:14:04.942 --> 00:14:08.022
not all, maybe some of them, we will develop,
00:14:08.502 --> 00:14:14.596
that kind of trust in the system to say that part doesn't need any more human intervention.
00:14:15.107 --> 00:14:20.502
I think we will reach that stage where there is a harmony between how AI agents
00:14:20.502 --> 00:14:23.204
are kind of working and the human in the loop.
00:14:24.002 --> 00:14:28.622
But it will happen over a time period when we have these data points proving
00:14:28.622 --> 00:14:35.122
that out over a really significant data that, okay, proves certain things doesn't
00:14:35.122 --> 00:14:36.882
still need the human intervention.
00:14:36.882 --> 00:14:41.322
And that's where I believe the pressure, also the fatigue that you talk about
00:14:41.322 --> 00:14:46.882
from an operator angle will significantly reduce. But there is indeed a very
00:14:47.282 --> 00:14:49.374
palpable friction that exists today.
00:14:50.084 --> 00:14:55.848
Okay. Ultimately, he should be responsible for setting escalation thresholds
00:14:55.848 --> 00:14:59.258
in an AI-enabled vision system. Yes, we do.
00:14:59.798 --> 00:15:06.978
And the thing is, again, many of those escalation that we have to kind of design
00:15:06.978 --> 00:15:12.948
into the system are still kind of in works, like who is going to respond to it.
00:15:12.948 --> 00:15:17.120
As I said, there is a preparedness that meets on the organizational side as well.
00:15:17.607 --> 00:15:22.328
And when you're working within the regulatory framework and the legal framework,
00:15:22.328 --> 00:15:26.068
I think we did not talk about the legal aspect of it. Even today,
00:15:26.068 --> 00:15:28.817
how everything is defined, right?
00:15:29.130 --> 00:15:33.268
Ultimately, the legal consequences kind of lies with whoever,
00:15:33.268 --> 00:15:35.440
the manufacturer or whoever it is, right? Right.
00:15:35.818 --> 00:15:41.738
If with any of this AI elements not working out properly, I mean,
00:15:41.738 --> 00:15:45.068
the legal consequences still kind of stays with the manufacturer.
00:15:45.068 --> 00:15:49.034
I think there is somewhere I think those are the frameworks that needs to work that out.
00:15:49.475 --> 00:15:54.122
And I can understand where, you know, even with all the right intent,
00:15:54.249 --> 00:16:00.635
where is the compliance and the legal frameworks and all kind of playing a role into what can you do
00:16:00.974 --> 00:16:05.625
or what you cannot, even if the AI systems are ready to do it. But I don't think so.
00:16:06.258 --> 00:16:09.868
Overall, how we run the business with all the compliance and frameworks,
00:16:09.868 --> 00:16:14.228
sorry, compliance, regulatory elements and legal frameworks and all of those
00:16:14.228 --> 00:16:16.262
things. I don't think so. Those have evolved yet.
00:16:16.755 --> 00:16:22.408
We would need all of that to evolve through the process. And there is obviously
00:16:22.408 --> 00:16:27.738
efforts going on. There is an immense pressure from the business side of the things, right?
00:16:28.541 --> 00:16:34.438
Which is happening, but in a way you're kind of managing or balancing the intent
00:16:34.438 --> 00:16:39.249
of a business to generate profit versus making sure things are from a safety angle.
00:16:40.169 --> 00:16:45.296
Well, to move on further with that, as Agenic AI becomes more common in factory
00:16:45.296 --> 00:16:48.526
automation, I don't know exactly where it is yet,
00:16:49.310 --> 00:16:53.776
but do you see machine vision systems evolving from defection detection tools
00:16:53.776 --> 00:16:59.746
into systems that can recommend or initiate corrective actions on the line? It is. It is.
00:17:01.086 --> 00:17:04.776
Yes, we are there, as I said. Many of the things we are kind of,
00:17:04.776 --> 00:17:08.756
yeah, many of this thing we are already like where I was talking of the robotic
00:17:08.756 --> 00:17:13.146
path planning or talking about a robot going and, you know, we're using those
00:17:13.146 --> 00:17:14.626
agenti guys to do make the
00:17:15.066 --> 00:17:18.436
inspection of different places and just kind of move along.
00:17:19.266 --> 00:17:23.136
But considering, I mean, one of the considerations that is taken in the space
00:17:23.136 --> 00:17:28.796
that we work in, aerospace and defense, is the value of the asset itself is so high, right?
00:17:28.796 --> 00:17:34.366
You know, they just cannot, they're just not comfortable leaving it to any autonomous things doing it.
00:17:34.366 --> 00:17:39.226
But I think when the same things are extended to maybe some structures and all,
00:17:39.226 --> 00:17:44.837
which may not be that expensive, I think they would be willing to maybe take that direction.
00:17:45.133 --> 00:17:50.656
But yeah, all those things are kind of playing out So to get to the point where
00:17:50.656 --> 00:17:53.876
you have all the guardrails and all the things you're talking about in place
00:17:53.876 --> 00:17:55.463
What kind of timeline are we looking at?
00:17:55.602 --> 00:17:57.901
How long do you think before that'll actually end?
00:17:59.562 --> 00:18:03.068
Let me first, you know, see the position here.
00:18:03.068 --> 00:18:07.818
Right now, if you look at, like, I think I was going through this report from
00:18:07.818 --> 00:18:09.878
Deloitte, I think they had an extended report.
00:18:09.878 --> 00:18:15.428
They were basically saying, okay, they had a study around that in terms of like how many,
00:18:15.988 --> 00:18:19.898
from an industry point of view, how many kind of adopted already the agentic
00:18:19.898 --> 00:18:25.048
AI and those autonomous pathways they're kind of taking and compare that with.
00:18:25.608 --> 00:18:27.308
The AI governance elements.
00:18:27.308 --> 00:18:32.249
So it seems like there is a significant gap, which is very understandable,
00:18:32.708 --> 00:18:39.710
that like I think 74% of the enterprises have adopted AI, like AI and all of that.
00:18:40.047 --> 00:18:44.192
But when it comes to the governance angle, there is only 21%, right?
00:18:44.343 --> 00:18:49.538
From where they feel, okay, they have something from maturity level that they
00:18:49.538 --> 00:18:52.515
can do those governance, make those governance happen, right?
00:18:52.840 --> 00:18:55.808
So you can already see the pressure is already there, right?
00:18:55.808 --> 00:19:00.938
You know, the advancements that is happening on the AI is going so much faster
00:19:00.938 --> 00:19:05.808
and there is indeed a business need, right? Ultimately, it's all driven by the business need, right?
00:19:06.268 --> 00:19:10.168
And so there's a lot of catch up that needs to be done from the governance side.
00:19:10.168 --> 00:19:12.670
I think that pressure is kind of happening.
00:19:13.094 --> 00:19:16.718
And at different levels, I think people have chosen different pathways.
00:19:16.718 --> 00:19:20.898
Like I think in US, we are seeing even it's driven from the government that
00:19:20.898 --> 00:19:25.148
we need to have the speed from you angle. They are saying, oh,
00:19:25.148 --> 00:19:30.468
no, we have to be very certain about things before we bring in all these agentic AI payments.
00:19:31.008 --> 00:19:34.638
And then I think there was some really good stuff happening in Singapore.
00:19:34.638 --> 00:19:38.060
Right. I think they have some what is that called IMDA or something like that,
00:19:38.565 --> 00:19:43.621
where the agency is driving. OK, they know that nothing can be achieved.
00:19:44.280 --> 00:19:47.687
To the perfection right from the beginning. So they have defined a model where
00:19:47.943 --> 00:19:51.716
they're going to iteratively kind of evolve this governance model,
00:19:51.948 --> 00:19:55.408
knowing very well that not everything is addressed right from the beginning.
00:19:55.960 --> 00:20:01.694
So very, very interesting how different approaches being taken at different places.
00:20:02.194 --> 00:20:05.513
We will see what outcome, which one does the best.
00:20:06.052 --> 00:20:11.334
Yeah, I actually cannot even say which one, which sides do you take, right? Yeah.
00:20:12.592 --> 00:20:18.204
Okay. How does the Proceed, Pause, Escalate framework help determine when the
00:20:18.204 --> 00:20:21.178
vision system should trust its own judgment? Yes.
00:20:22.785 --> 00:20:25.646
I think, let me give an example from an inspection standpoint, right?
00:20:26.122 --> 00:20:32.016
So there are many inspections, okay? So where, I'll say, based on what we are
00:20:32.016 --> 00:20:36.186
doing, we believe the vision system in many things you can trust today.
00:20:36.186 --> 00:20:42.237
But again, I know where some of those compliance or regulation thing comes as well.
00:20:42.805 --> 00:20:46.396
But for example, many of the defects, which is kind of innocuous,
00:20:46.396 --> 00:20:50.806
right? Or innocuous in the sense, maybe it doesn't have a life-threatening implication.
00:20:50.806 --> 00:20:52.210
I'm coming from that angle, right?
00:20:52.500 --> 00:20:56.826
But those inspections still need to be done. For example, you have an aircraft
00:20:56.826 --> 00:21:01.656
cabin where you are going to use your vision system to just use a headset, scan.
00:21:02.716 --> 00:21:07.766
The cabin and just see if everything on the seat is fine or if there is some
00:21:07.766 --> 00:21:09.296
tear or something needs to be done.
00:21:09.296 --> 00:21:13.606
Now the vision systems can do all of those things. right that's something even
00:21:13.606 --> 00:21:17.706
if you are able to completely offload it to the vision systems and say okay
00:21:17.706 --> 00:21:21.153
i'll take that 100 decision of it and move ahead right,
00:21:21.670 --> 00:21:25.766
so what would happen if even if something is missed is maybe you have a dissatisfied
00:21:25.766 --> 00:21:27.961
customer that's not an end all right,
00:21:28.616 --> 00:21:33.536
but when i come when i do like the rivets and all like what are there on the
00:21:33.536 --> 00:21:38.573
aircraft if i try to detect if there is a rivet which is missing or a bolt that is missing but
00:21:39.014 --> 00:21:42.436
is it something that you can completely depend on the AI, yeah,
00:21:42.436 --> 00:21:44.296
it would do a very good job.
00:21:44.296 --> 00:21:49.296
Like, you know, we can say to the confidence that almost 99 point,
00:21:49.296 --> 00:21:54.446
maybe 9%, but I will never be able to say 100% that it would capture everything.
00:21:54.446 --> 00:22:00.251
It has consequences. So that's where you want to always have the human in that load.
00:22:01.080 --> 00:22:08.781
Okay. There will probably not be anytime soon an utterly autonomous situation in that. Yeah.
00:22:09.077 --> 00:22:14.366
I think what would change is when you are doing the inspection of the autonomous
00:22:14.366 --> 00:22:18.591
system where humans are not flying, right?
00:22:19.396 --> 00:22:22.456
Because there may be a different conversation.
00:22:23.691 --> 00:22:27.041
Okay. Do you have anything you want to add to what we've discussed?
00:22:27.425 --> 00:22:28.995
The only thing I would say is
00:22:28.995 --> 00:22:33.045
there's a real pressure coming from the business side of the value of AI.
00:22:33.155 --> 00:22:37.887
That's been kind of proven. I think as we get more data points, you will see.
00:22:39.375 --> 00:22:43.775
If the operational issues that exist, so many of them, and when you start using
00:22:43.775 --> 00:22:50.933
HDI, you say, okay, I got maybe 70% of my operational issue resolved this year just using AI.
00:22:51.426 --> 00:22:57.643
It just builds those cases so strong right and it can do it and we we strongly believe it can do that,
00:22:58.230 --> 00:23:02.815
but i completely understand when you're balancing the the business requirements
00:23:02.815 --> 00:23:10.081
which is driven from a profit angle versus safety which is driven from the humane safety angle,
00:23:10.654 --> 00:23:14.635
i understand that and i think somewhere those balance needs to be kept and i
00:23:14.635 --> 00:23:21.170
think it's a journey that we have to make, but we'll see more and more agentic AI systems evolve.
00:23:21.646 --> 00:23:25.695
We get confidence with all guardrails that we're defining. The infrastructure
00:23:25.695 --> 00:23:29.575
is getting more ready to use more data points to, in real time,
00:23:29.575 --> 00:23:32.774
make those decisions, because that's where the fully autonomous things will happen.
00:23:33.117 --> 00:23:36.126
And we can then trust the system decisions much better.
00:23:38.129 --> 00:23:42.205
Well, that's a wrap for this episode of Visions, produced by Endeavor Business
00:23:42.205 --> 00:23:44.729
Media, a division of Endeavor B2B.
00:23:45.234 --> 00:23:48.235
Thanks very much for tuning in. If you enjoyed today's show,
00:23:48.235 --> 00:23:52.045
be sure to subscribe to the podcast and share this episode with a colleague
00:23:52.045 --> 00:23:52.984
who would find it helpful.
00:23:53.646 --> 00:23:59.535
Until our next episode, you can find us at vision-systems.com or on LinkedIn,
00:23:59.535 --> 00:24:04.176
Facebook, or X for more insights, updates, and breaking news to keep you in the know.
00:24:04.948 --> 00:24:08.640
Thanks for tuning in. Until next time, stay focused on your visions.