Latest / Agentic Conversations (formally mlops.community) / AI & Aliens: New Eyes on Ancient Questions // Richard Cloete // #288

Transcript

AI & Aliens: New Eyes on Ancient Questions // Richard Cloete // #288 · Agentic Conversations (formally mlops.community)

0:00
To hi, I'm I'm Doctor Richard Cludy.
0:03
I am a computer scientist at Harvard University and also
0:07
running my startup called Seeker Robotics.
0:10
And I take my coffee instant and black, no sugar, no milk.
0:16
Welcome back to a very, very special ML OPS community
0:21
podcast. This one is right out of left
0:25
field. Today we hit the E brake and
0:28
skid on by 2020 fives coming at me hot.
0:35
We're talking about UA PS and how Richard is using AI to try
0:39
and identify random things in the sky.
0:44
We also talked about his startup Seeker and how he is observing
0:49
things in the sea and his work with the Minor Planet Center on
0:58
the observatory and using AI and ML to sift through the mountains
1:05
of data that they are getting at the Ruben Vera Ruben
1:10
Observatory. That was the name of it.
1:13
So if you like this, you're interested in what Richard is
1:19
doing, I highly encourage you to check out the Galileo Project
1:24
that we mentioned a few times or check out what he's doing with
1:27
Seeker Robotics. Let's get into this episode and
1:33
let me know what y'all think of this because my mind was blown.
1:42
I got to know, man, like you're doing some AI that is not in the
1:46
traditional side of AII think or when people think AI these days,
1:52
they don't necessarily think about the AI that you're doing,
1:56
but you're doing such cool stuff.
1:58
And I want to know like what drew you into the world that you
2:04
are in right now? And maybe tell me a bit more
2:06
about what exactly you're doing and how you're using AI to do
2:11
it. OK, well, maybe we should start
2:13
with a bit about the Galileo project, an overview of, of what
2:18
that is all about, because that's where I'm doing the AI.
2:20
So it'll give some context to what we're doing.
2:23
So at the, at the Galileo project, you know, we, we're
2:25
building the sensor suite of, of, of sensors and the aim is to
2:30
essentially monitor as much of the electromagnetic spectrum as
2:33
possible. So we've got infrared cameras,
2:38
we've got an all sky visible camera monitoring the sky as
2:41
well. We've got acoustic sensors,
2:45
we've got like things like a magnetometer as well and a few
2:49
other sensors. And the the idea is essentially
2:54
see if we can detect anything in the sky that is unusual.
2:59
So we obviously expect to see things like birds and, and
3:02
airplanes and whatever else, you know, might fly through the sky.
3:08
But the main point is to look for UAP.
3:11
So I don't know if you're familiar with UAP and what's
3:15
really recently been going on in the in the media and in in
3:18
government as well. But essentially we know that
3:23
there's something going on and someone is not telling the
3:26
truth. We don't know whether it's the
3:27
government or the whistleblowers or whatever, but we don't care
3:30
as scientists, we just want to put instruments on the ground,
3:32
collect data and let the data speak for itself.
3:35
That's the purpose of really the Gallo project, to get some data
3:39
and you know, we'll see what happens from there.
3:42
And one of the, the core systems, the ones that I'm
3:46
working on is known as the Dalek.
3:48
It's a hemispherical array of, of about 8 cameras with a, a
3:53
camera on the top as well as Zenith.
3:55
And I'm using the video feeds from there to detect one,
4:01
looking rather at the the video feeds to see if we can detect
4:04
and track any, any objects flying across the sky.
4:09
So my task was really to build an AI model that could achieve
4:15
this. And when I started there, there
4:18
was no real data set out there to be able to work on data
4:23
coming from infrared cameras looking at the sky.
4:26
There's a few data sets out there you can go to hug and face
4:28
or whatever. And you can get tons of
4:30
different data sets for a different machine learning
4:32
purposes. But unfortunately a lot of them
4:37
have like for example, we want something like an airplane.
4:40
We want to be able to detect airplanes, a data set of
4:43
airplanes. But these airplanes might be on
4:46
a tarmac or they might be on a kids T-shirt or it might be some
4:52
cartoon airplane or something. And we don't need that kind of
4:55
that kind of data. We need data from cameras
4:58
pointing up at the sky with aircraft flying through them.
5:02
The same with birds. You know, birds might be two
5:04
birds on a branch on a coffee mug or something.
5:06
That's just, we not, not the kind of data we need.
5:11
And so we, we pretty soon realized that what we needed to
5:14
do was simulate our data. So we, we built a tool which we
5:19
call air synth. So air synthetic data basically.
5:25
And, and we used Blender 3D. So that's a 3D rendering
5:30
modeling application which you're probably familiar with.
5:34
And we use a bit of Python And grabbed a bunch of 3D models,
5:39
open source 3D models, and basically distribute them in the
5:44
scene, randomly distribute them in the scene within the view,
5:47
first room of the camera, and then render them out.
5:50
So you end up with images of airplanes or birds or a mixture,
5:57
balloons, blimps, drones, you know, all those kind of things
5:59
you might find in the sky. And you get a mixture of those
6:02
in these images at different orientations, different
6:04
distances from the camera so that they are different sizes
6:09
and, and, and the different liking conditions.
6:13
So we have a, we have lots of really good data sets that we
6:16
then use to train you're the V8 models.
6:19
Well, you're the V5 models in the previous paper and the
6:22
previous round, but now we've moved on to your V8 and your
6:26
V11, which has just come out as well.
6:29
And so we've trained several different flavors of the models
6:34
on, on, on these data sets. And then we use a bunch of
6:38
different tracking algorithms as well.
6:40
So we use things like just a regular sort algorithm, which is
6:43
simple online real time tracking byte sort.
6:47
And I think Viola comes with the true as well.
6:49
So we've tried a few of those and essentially what we're doing
6:52
is we're detecting objects in the in the video feeds and then
6:57
tracking them. So we're extracting the
6:59
trajectories. And then we're also working on
7:02
on algorithms to be able to look at these different trajectories
7:07
because that's what's really interesting.
7:08
I mean, you might have an object, but it might just be a
7:12
little white dot, you know, it might be far away.
7:16
We don't yet have distance. So we're working on
7:19
triangulation. But the the trajectories are
7:23
what's really interesting because that'll tell you whether
7:25
something is moving in a way that doesn't conform to typical
7:31
known man made or natural phenomena.
7:35
And so the idea is that we get alerts generated every time the
7:37
system identify as something unusual, but we're not at that
7:42
point yet. We've got, we've got you know
7:44
what the detection and tracking algorithms pretty nailed.
7:47
There's always room in machine learning obviously to improve on
7:51
your detection accuracy and that that's an ongoing project and I
7:55
suspect it will be for some time.
7:57
But we are able to robustly extract out detections and and
8:03
tracks, which is really cool to see.
8:06
And how many of these different apparatuses do you have around
8:09
and where are they? We have at the moment 3 sites
8:15
that we, I think it's three sites.
8:18
My memory is quite bad. So I, I think it's about 3 and
8:24
I, I can't really say where they are.
8:28
One of the big questions we had around the gallery product when
8:32
we, when we started deploying to different sites was like, or can
8:34
we go and tell people, you know, where we put in these things?
8:37
And the consensus was generally no.
8:39
And the reason for that is because there's so much like
8:44
stigma and sensitive issues around the UAPUFO subject.
8:49
We don't want to give away locations of our sensor
8:52
apparatus in case you get some people out there who say, all
8:55
right, I'm going to go over there and I'm going to try and
8:58
trick them, you know, and, you know, mess with our data
9:01
streams, mess with the cameras or just even mess vandalize the
9:04
instruments. So for the time being, those are
9:08
all kind of secret. Yeah.
9:11
Well, hopefully you had one in New Jersey over the last month
9:15
or two because I heard it was a pretty big hot spot there.
9:18
Yeah, it was. There was a lot going on over
9:21
there. I'm, I'm not entirely sure what
9:23
was going on, but we do have some portable systems, I'll just
9:28
say that. Oh cool, So you can go almost
9:32
storm chasing when there's things that are happening and
9:35
you get a hotspot. And what made you want to get
9:39
into this and recognize that you could a, combined these models
9:46
with the identification and basically put AI to use at this
9:55
task? Because I don't think I've heard
9:57
a use case quite like this in the last five years.
10:01
I've been interviewing people. And that's one of the reasons
10:03
that I wanted to get you on here is because it is so unique.
10:07
Yeah. So, I mean, there's so much
10:11
data, so many videos feeds that we, that we, we have to process
10:15
and it's just not feasible to have someone sitting there
10:17
monitoring 24/7. And on top of that as well, the,
10:23
it's very often the case that the object might be very small
10:26
and it's hard to discern exactly what it is.
10:28
And humans are really bad at classifying When you see things
10:31
in, in, in, in video feeds or in images, you know that people
10:35
make all sorts of mistakes. One person says it's a bird and
10:37
one says it's an airplane. So we don't want to rely on, on
10:42
humans to be able to process that data.
10:44
Instead, we just, you know, build the machine and a model to
10:47
do it for us. That way we get consistent
10:49
answers across 24/7, 365 days a year, and it's much faster as
10:56
well. Yeah.
10:58
So the obvious question is, have you found anything?
11:03
There's been, there's been several cases where we've seen
11:05
something that's interesting. But I must stress that at the
11:10
moment the the systems are not fully operational in the sense
11:13
that they out there continuously processing, processing data and
11:17
sending us alerts and detecting anomaly.
11:21
If we're not here to that stage, however, we we are in the
11:26
commissioning phase. So we're at the moment deploying
11:28
instruments to the ground, leaving out there collecting
11:31
data, analyzing that data. And the goal is really just to
11:34
see if the, the data is what we expect to see.
11:39
You know, if the systems are performing the way we expect
11:41
them to form and getting the baseline as well, right.
11:43
That's really important is we're trying to get a baseline of
11:46
what's normal for this area and that's different for every area.
11:49
1 area you might have more birds or more aircraft if it's make
11:52
sure an airport or something. And so just through like manual
11:57
review of the data and testing algorithms, yeah, we pulled out
12:01
some some things that look really interesting.
12:04
And unfortunately we haven't been able to explain.
12:07
We can't say that they are AAP and we can't say that they are a
12:11
bird or an airplane just because it's not enough data.
12:13
And in a lot of these cases, we've only got like one video,
12:17
you know, so even though there's lots of cameras recording, it
12:21
might be that at this particular time those cameras are down or
12:26
our acoustic system wasn't working or some, there was some
12:30
other issue. And, and so we don't have the,
12:33
the kind of data that we need to make a proper verification,
12:37
which is data from multiple different instruments all
12:40
corroborate in the exact same event at the same time.
12:46
And a key component of all this that's missing, which is
12:49
actively under development is the ability to triangulate.
12:53
Because if we can triangulate objects, we can get through a
12:56
distance. And then from that we can
12:58
extract things like the speed, velocity and so on and so forth.
13:02
And that's, that's something we're working really hard
13:05
towards. And we've got some, we've made
13:06
some really good progress in there as well.
13:09
But until, until we get that, that information, it's difficult
13:12
to even say. We can't say what, how big the
13:14
object is. So it could be something which
13:16
is really far away and huge, or it could be something small and
13:20
close. But yeah, watch the space
13:24
because there's definitely some, some interesting things coming
13:27
out. We just need, we just need to
13:29
have the right process in place to make some verifications
13:31
before we can talk more about them.
13:34
And it feels like you're, you're doing a lot of stuff out there
13:38
on the edge devoid of communication with any, any
13:43
cloud services. And so it's just gathering this
13:46
data and then someone or somehow you're batch downloading it and
13:52
then able to process it. Yeah.
13:54
So everything is running at the edge and we're using, we have a
14:00
stalling system set up there as well, but we also have 5G.
14:05
So we have the ability to egress data automatically, but not for
14:11
like live video feeds and things continuously.
14:15
So instead of what we do is restore the data locally and
14:18
then someone will go there and collect the data for like the
14:20
previous week or the previous two weeks or whatever.
14:23
They'll bring it back to the Center for Astrophysics, and
14:26
then they'll hook it up to our systems there and then transfer
14:30
the data to our cluster at Harvard where we can then do the
14:35
analysis. How do you separate the the
14:43
wheat from the shaft where I'm sure you get a lot of really
14:49
interesting types of folks that want to come and do stuff here
14:53
with like just saying aliens makes I think a certain type of
14:59
person be really excited and another type of person be turned
15:03
off. But you're doing it.
15:05
It seems like I mean, just talking to you for the last 20
15:08
minutes, you've got 2 feet on the ground.
15:10
You don't look like a few of my friends grown up that wore alien
15:14
shirts. And, and so how has it been
15:20
being in almost like in the scientific community trying to
15:24
do stuff in this field and staying on the straight and
15:29
narrow, if we can call it that? Yeah, You know, there's so much
15:33
stigma around the the UFOUAP subject.
15:36
It's been going on for decades already.
15:39
And if you see a new segment, you'll likely hear the The
15:43
X-Files music playing in the background or something like
15:45
that, because there's no need for this stuff.
15:48
So we're, we're taking a completely scientific, pragmatic
15:52
approach here. We're not, you know, we're not
15:54
trying to make any assumptions about what we might or might not
15:56
find. We're just saying, let's get the
15:59
instruments out there, put them on the ground, collect the data,
16:01
and then we can talk about the data.
16:02
We'll open source that data and we can argue about what it might
16:05
be or might not be. Let other scientists interpret
16:08
that data. And so all my conversations that
16:11
I've had with where would be with peers or whoever have have
16:14
been on the same, the same line of thought and we've not really
16:19
had any pushback against that. There's some people who of
16:23
course have a little. Bit of a.
16:26
You know, these guys are crazy mindset, but I mean, So what you
16:32
know, if you if you don't look, you're not going to find
16:34
anything. So let's get out there and, and
16:36
have a look at least at least we can say we tried.
16:39
And when it comes to bringing people onto the project, of
16:43
course, we, we, we do interviews as well.
16:45
So we, we sit down with the person and, and chat with them
16:48
to just make sure that, you know, they're not on the other
16:52
side. We, there are, there are the UFO
16:56
communities, a bit of a mixed bag.
16:58
And, and we bring in generally we bring in technical people
17:03
because that's where that's where the, the real work is.
17:08
And there's still a lot, still a lot to do.
17:09
Then we, we always show it on, on people.
17:11
It's a completely volunteer based project.
17:14
So we always looking for additional additional help on
17:17
that front. That's excellent.
17:19
OK, so that's UAP stuff, which I find absolutely fascinating and
17:24
I imagine it keeps you up at night and you love working on it
17:29
and that's why you're doing it. There's also other really cool
17:33
things that you're doing that I want to get into, one being
17:37
around like observatories and how you are seeing AI being used
17:42
with telescopes and being used like with images from deep space
17:48
and all that. Can you go into some of that?
17:51
I work very closely as well with the Minor Planet Center.
17:54
And one of the big tasks that we're we're looking at now is
17:59
there's going to be a new telescope coming online very
18:01
soon called the Vera Rubin Telescope or the LST.
18:06
And this telescope will collect a lot more data than previous
18:12
generation telescopes. It'll see a lot more of the sky
18:15
and we'll see much further. So you can, you can imagine
18:19
going to the back into your backyard at night with a
18:21
flashlight and you can only see so far.
18:25
And then you come along with a bigger flashlight and you, you
18:28
can suddenly see much further. And there's a whole lot of new
18:30
objects, new things in, in at the back of your garden.
18:34
And it's the same kind of thing with this new telescope coming
18:36
out online, we'll be seeing so much further.
18:38
There's so many new objects and, and not just different
18:42
varieties, but so many more of them.
18:47
And so one of the, one of the things that kind of recently
18:52
picked up a little bit of traction, instead of
18:54
interstellar objects, we've had a MUA, MUA, which was the first
18:59
interstellar object to enter our solar system in 2017.
19:04
Well, the first one detected, let's say.
19:07
And, and so working with, with the, the data from from these,
19:12
these telescopes, I'm building models to be able to identify
19:16
which ones will, let's say, flag which ones might be interstellar
19:20
objects. And then interstellar object is
19:24
just something that comes into our solar system for a little
19:27
bit and then it leaves. Yeah, that's right.
19:30
Could be. It could be anything.
19:31
It could be a rock. It could be a a spaceship.
19:33
You notice, you know, it could be anything.
19:36
So keeping it up with the theme, yeah, I like it.
19:39
This telescope that you're talking about, the Vera Rubin,
19:42
it's in outer space. I know it's it's ground based.
19:46
OK. Yeah, I think it's in Chile.
19:50
But it's more powerful because just because of the technology.
19:53
Yeah, it's got a a much bigger aperture as well.
19:57
It can just see much further into space than than regular
20:00
ground basin systems. There's also the one that is in
20:05
space, right? The James Webb or the Hubble?
20:08
Yeah, yeah. And that one is fascinating
20:13
because it it gets all of the ozone.
20:15
I had thought that the ozone was a bit of a pain in the butt when
20:20
it comes to the telescopes because it can muddy up the
20:24
vision or the image. Yeah, honestly, I'm not a a big
20:28
telescope buff. I just work with the data.
20:31
So I can't really describe much about the the technical aspects
20:34
of the telescopes, but there is definitely atmospheric
20:38
disturbance which which plays a role and after fact that in but
20:42
the, the, the Reuben telescope, it'll image the entire, the
20:47
southern sky like two or three times a night, I can't remember.
20:52
So that's in contrast to something like the James Webb or
20:56
the Hubble, which is much more like narrow field, narrow field
20:59
of view. OK.
21:01
Yeah. And I see the difference there.
21:04
And so the idea is like you're going to be creating different
21:10
models that are going to help you identify when things come
21:14
interstellar. Is there other models that
21:17
you're creating? Yeah, So it's not just actually
21:19
interstellar objects, it's we're trying to classify as many
21:22
different types of objects as we as we can.
21:24
So data typically gets submitted to the minor planet center.
21:31
And this data can come from other territories.
21:33
It can come from amateur astronomers who have A and it
21:36
goes through some one of their, one of their pipelines and, and
21:40
then essentially before it hits onto their onto their website,
21:46
we want to be able to intercept that data.
21:48
So we're working with the minor plans to be able to look at this
21:51
data and identify whether one of the objects is an interstellar
21:54
object, whether it's the main delta, Jupiter, Trojan, Mars
21:59
crosser, whatever, whatever, you know.
22:01
So we have tons of different classes of objects out there.
22:04
And just by using machine and models, we can kind of separate
22:08
the different types of classes. There are already tools out
22:12
there which, which work towards this.
22:14
And there's, there's a, there's a good one actually called
22:17
Digest 2, which is a software that I've been working with as
22:23
well. It kind of operates as a binary
22:25
classifier. And what happens is people
22:28
submit an object to the minor planet center and it produces
22:32
what's called a Digest 2 score. And the score is ranges from
22:36
zero to 100. And if it goes over 65, it's,
22:41
it's classified as something like interesting, basically a
22:44
near Earth object, OK. And, and it goes onto your onto
22:47
the website and into the database.
22:49
Anything else like a main Belter or Mars crosser, we don't really
22:52
care about those because a minor planet set is mostly focused on
22:57
threats to Earth, near Earth objects.
23:00
So sometimes an an interstellar object can be a near Earth
23:07
object, a mua. MUA was classified as a near
23:09
Earth object. It was detected and only
23:11
discovered really because it was detected as a near Earth object.
23:14
It's it was coming towards Earth and they think, hey, what's
23:17
that? Let's take a look.
23:19
And people started investigating more and then they thought, OK,
23:22
there's some unusual properties about this.
23:24
And that's how it all kind of blew up and anchor all the media
23:28
attention it did. But there's just so much data
23:32
coming in there. You can't, you can't have my
23:34
guys sitting there manually processing this data and looking
23:37
to see if this part anything unusual.
23:38
So while this tool that we have at the moment, Digus 2 works
23:42
really well, it's there's a, there's a, there's a lot of like
23:46
false positives, right? So in many cases what happens is
23:51
an NEO is missed. So it'll, it'll have a low Digus
23:55
2 score, for example. And as a result, it'll really
23:59
get kind of like kicked out of the system and people won't pay
24:01
attention to it. And that could be for a variety
24:05
of reasons. The problem is that we need to,
24:07
we need to detect as many near as obvious as possible because
24:10
because they might be a threat to Earth, could be on a
24:12
collision course with Earth. The, the things I'm working on
24:17
at the minor Planet Center as well is improving their system
24:20
to be able to, with better accuracy and more confidence, be
24:24
able to predict whether the object is a near Earth object or
24:27
not. And we should have a paper
24:29
coming out probably a couple of months, I would say on that.
24:34
But yeah, machine learning is playing a big role in this
24:37
because, you know, there's just so much data and the models
24:41
today are are really good at at differentiating between
24:45
different classes of of objects, especially in this kind of data
24:48
that we're working with. And are you doing the same thing
24:51
where it is trying to simulate different visual aspects of it
24:57
or Or is this more of like tabular data?
25:00
That's right. Yeah, Yeah, it's, it's more,
25:03
it's just purely a numerical at this point.
25:06
We're not looking at any images, though we do, we would like to
25:09
at some point start incorporating that image that,
25:12
that data as well, because there's a lot you can get from
25:15
the image data you can get like like hers and and things like
25:18
that. So yeah, in the future we hope,
25:20
we hope to be able to do that. Nice.
25:23
Well, I also want to talk about the ocean exploration stuff
25:29
you're doing and the seeker I think is, is what it's called,
25:34
right? Robotics is super cool.
25:37
So first of all, like it makes complete sense to me why you
25:44
would be very excited and interested in exploring the
25:49
outer space and then also wanting to explore the ocean
25:53
because there is so little that we know about it.
25:56
So tell me a bit about Seeker Robotics.
25:59
Wow. OK, so Seeker Robotics is a
26:05
company that I recently started. It's follows the same kind of
26:10
idea as a Galileo project, right?
26:12
So the Galileo project, we're looking at the sky.
26:14
We're trying to detect objects that are moving through the sky
26:17
and track them and classify them.
26:19
But what I, what I realized is that there's, there's a lot of
26:25
there's a lot of situations, events that happen around
26:28
UAPUFOS at sea. OK, that's Constantine in, in,
26:33
in, in the stories as well. Something coming out of the
26:35
water, something going into the water.
26:37
The three Navy videos that were released by the, the Pentagon, I
26:41
think in 2017, it was all related to the ocean as well.
26:45
You constantly hear reports of these objects buzzing nuclear
26:50
submarines or aircraft carriers, nuclear power aircraft carriers.
26:55
And, and, and So what I wanted to do was maybe move some of
26:59
the, the technologies that we developed at the LA project over
27:03
to the ocean. Have a look what's going on
27:04
there. There's so much out there and
27:06
it's just not being monitored, right?
27:07
Like 70% of the the planet is covered with, with ocean and
27:11
there's so few assets out there actually doing proper robust
27:15
monitoring. Of course there's things like
27:17
boys and there's this manned research vessels out there and
27:21
there's the military and fishing and whatever, but nobody's
27:24
really sitting there paying attention to consistently
27:27
monitoring the sky, the surface and the, the water below.
27:31
And so the idea was to take the knowledge that I'd, that I'd
27:35
gained from, through the Galileo project and, and, and build a
27:38
network of these, they're called Usvs, uncrewed surface vehicles.
27:44
And you can, you can Google what they look like online.
27:46
They typically like a a surfboard kind of shaped object,
27:50
but the idea is essentially it's a floating observatory, OK.
27:55
And we want to have a network of these things so that they are
27:58
constantly monitoring these different domains and sending
28:01
back data in real time. So sending back alerts every
28:05
time something is detected. It could be schools of fish, it
28:09
could be a submarine, it could be another boat on the ocean, it
28:12
could be an airplane over a head or a bird or whatever.
28:16
And, and so Seeker is a full proper company, but the idea
28:23
here is that we can generate a lot of really important data for
28:28
marine biologists, for the Coast Guard, for situational awareness
28:32
with 500 defense, for atmosphere and climate research, for
28:37
oceanography. There's just so much data that
28:38
we can collect out there that we can use these these data as a
28:42
means to fund the research and development of these platforms,
28:48
which we can then develop anomaly detection algorithms on
28:53
top of that would allow us to maybe look and see if we can
28:56
find anything unusual. So it's a kind of A2 pronged
28:59
approach to to the subject because I've been working with
29:03
RV now for 2 1/2 years and he's done an incredible job.
29:10
But a common theme in all of this is that there's really no
29:14
funding that comes along because there's still, as I mentioned,
29:19
there's so much stigma around the subject.
29:21
There's no NSF grants out there. There's no, there's no funding
29:25
tracks essentially to help scientists do their work, then
29:29
do the work that they need to do on the UFO subject.
29:33
So Abby has been doing an incredible job raising funding
29:36
all through donations. OK, the, the entire Gallo Gallup
29:40
part is basically all funded through donations.
29:43
But long term that's not really sustainable because eventually
29:48
at some point, you know we're going to run out of money and
29:51
there's going to be nobody to step in there and and donate
29:53
some extra cash to us. So we need a sustainable revenue
29:57
stream. And so that's where I started
29:58
thinking a bit more broadly about how we can actually build
30:01
something out there that will generate a revenue stream that
30:04
would also allow the continuation of UAP research.
30:07
And then also some of that funding could be funneled back
30:12
to, I could, you know, secret. If it gets to a point where it's
30:15
generating revenue, we could start sponsoring postdocs at the
30:20
Galileo project or other scientific projects, or we could
30:23
purchase equipment, donate equipment for them or like the
30:26
Sole Foundation, another project based out of Stanford, we could
30:31
sponsor them as well. So the more companies out there
30:35
that are UAP friendly, I think the better.
30:39
And so the surfboard, it has all the same sensors that you have
30:45
with the the Galileo project ones that you're mentioning
30:49
before when you're putting them on the ground, like all these
30:51
different cameras and acoustics, infrared and all that.
30:55
Yeah. So it'll, it'll have very
30:56
similar sensors. Of course the Gallo is not is
30:59
not monitoring the ocean. So we will have additional
31:03
things like like hydrophones and conductivity sensors and
31:07
salinity and and other sensors. But the idea is is basically the
31:12
same to just monitor, detect and track and classify in multiple
31:16
domains. And how are you going to keep?
31:20
Or are you going to keep it in one spot?
31:22
Or are you gonna just let it float about?
31:24
Well, that's a challenge. The, the ocean is a powerful
31:28
place. So we're looking to use
31:31
something like the wave glider. The wave glider is a technology
31:35
which allows the object to move just through the motion of the
31:38
waves. So I don't think we'll be able
31:42
to keep it perfectly stationary. So the idea is to essentially,
31:45
if you imagine dividing the ocean into just a grid and then
31:48
each one of these Usbs will be responsible for monitoring one
31:51
of these grid cells. And and to do that we would
31:55
basically have the USB just moving in circles within that
31:59
that grid. Wow.
32:01
And so this is this hasn't gone out yet.
32:05
You haven't been collecting data yet on anything.
32:08
It's still no, it's yeah, it's still very much in the, in the
32:11
research and development stage. OK.
32:13
We have, we have some funding to to build our prototype and, and
32:19
we well on the way there, we actually looking to harness the
32:22
wave energy because a big problem with existing systems
32:27
out there is that they can't stay out for, for and and
32:30
indefinite amounts of time just because they run out of energy.
32:34
Yeah. So we're looking to be able to
32:35
harness the the motion of the waves to to generate electricity
32:40
and keep us out there for much longer.
32:42
Man, that is so cool. I am.
32:44
I'm a huge fan of all of this that you're doing and with
32:49
Seeker. Is there any plan to also try
32:53
and go under the water? So yeah, that that's a good
32:56
question to start off with us, we'll be monitoring just the
33:01
ocean surface. Let's say enough in our MVP,
33:05
we'll be monitoring maybe the 1st 10 to 20 meters of the, the
33:10
ocean surface and then this guy's above.
33:13
But as we, as we, as our system evolves and as we, we get more
33:18
funding, we, we do plan to to eventually build larger Usvs,
33:22
ones that are capable of mapping the sea floor, but also ones
33:26
that serve as a mothership next, call it and allow the the to
33:35
drop probes down and then retrieve them.
33:38
So if you, one of the big things that, that has a lot of research
33:45
going on is the, the water column, measuring the water
33:49
column. And, and so we'd be able to like
33:52
drop probes down, collect samples, maybe go to the ocean
33:56
floor, rise back up and dock and, and, and, and offer that
34:01
data. But it could also serve as a
34:03
communications hub. So we want to add things like
34:05
acoustic modems so that underwater assets, whether it be
34:10
another ROV or underwater observatories or any other
34:17
probes that are out there underwater can, can relay the
34:21
information back to to our our seeker USB.
34:24
And then we can send that over back to the servers on land.
34:28
Because typically these systems that are currently deployed to
34:31
the ocean, they have to surface if they want to broadcast.
34:34
So they can go out there collectively, they're
34:36
underwater, but then in order to send the data they have to
34:38
surface, and if there's a secret USB within range, they won't
34:43
have to do that anymore. Yeah.
34:46
It's like a 5G network for function.
34:49
Yeah, Yeah. It is weird to me that we
34:55
haven't done more in the space. And especially knowing, I think
35:00
everybody's probably said it at some point where like, oh, we
35:04
know more about space than we do about our own oceans.
35:08
And so the idea that we don't just set up different
35:13
observatories in our ocean is kind of weird to me.
35:19
Yeah. Yeah, it's, that's absolutely
35:21
true. True.
35:21
I mean, this is so much open space out there and nobody's out
35:25
there, nobody's watching it. Nobody has any idea what's going
35:28
on. You know, you've but that case
35:31
with these, with the drones in New Jersey, you mentioned
35:33
earlier, those are all coming apparently from the ocean.
35:38
Where are they coming from? If they're, if they're a foreign
35:40
tech, you know how they get in there, It means that there's not
35:42
enough observation going on, not enough monitoring to be able to
35:46
understand where these objects are coming from and where
35:50
they're going when they go back into the water.
35:52
There's also issues around the Coast Guard.
35:54
They don't have a lot of US VS out there.
35:58
So there's a lot of situational awareness gaps.
36:01
A lot of the, the drugs, drug smugglers, surprisingly
36:05
sophisticated. They actually have underwater
36:07
drones that can, that can avoid, you know, manned vessels which
36:11
are sitting there looking to see if they can see anything on the
36:13
water. And these for intense purposes.
36:17
Unmanned submarines are smuggling drugs and we should be
36:22
able to detect those. And the other piece that is
36:27
fascinating is how the data that you collect, there's going to be
36:33
demand for that data. I can imagine that.
36:36
And it's not just going to be for folks like yourself that are
36:39
trying to find anomalies in what's going on, but also just
36:44
like anyone who wants to try to figure out, oh, is, is X
36:51
happening with the ocean? Or is why?
36:55
There's going to be so many questions that you now can add a
36:59
little bit more color to. Yeah, that's right.
37:02
And you know, at the moment the typical scenario is someone will
37:08
come up with a research plan or I want to measure the the water
37:11
temperature, fuss and touch location and there we go.
37:16
And then develop a proposal around there that submit that to
37:19
get the money. But then I'd have to maybe
37:21
either go and purchase the equipment and manually, you
37:25
know, deploy it and retrieve it and maintain it themselves, or
37:28
they'd have to hire a company to go out and do it.
37:31
There's a lot of costs and a lot of time overheads involved in
37:34
all of that. And and So what we're going to
37:36
offer is essentially a US view as a service.
37:39
So these things will be out there on the ocean and you say,
37:43
oh, we need this data. Maybe the data is already
37:45
available. You can just access through a
37:47
subscription plan or purchase this the data package from us.
37:50
Or you could take control of the fleet of US views and go and
37:54
conduct a mission, plan it all from the comfort of your office
37:57
or your home and don't have to worry about the deployment, the
38:00
retrieval and maintenance or, or, or any of that.
38:04
And that as well. The, the, the key thing is that
38:07
the CKUSBS will be out there constantly.
38:11
So it'll be a persistent coverage of data, which is
38:15
typical with do with typical missions.
38:17
You go out there, you collect your temperature data and then
38:19
you come back and then we have no idea what the temperature is
38:22
next week, you know? So there's just, there's just so
38:25
much, so little observation on the ocean.
38:30
Yeah. And are you not afraid that the
38:34
the traffic in the ocean is going to be a problem?
38:39
No shipping routes unknown. You don't have, you know, ships
38:42
just going off wherever they want.
38:45
They want to take the optimal path when they're transporting
38:49
cargo. And that's something maybe we
38:51
can help improve on as well with the data that we collect.
38:54
There will be cases, of course, where we might have to cross a
38:57
shipping lane or something like that, but these systems will all
39:01
have on them what's called, I think it's AIS, or automatic
39:06
identification system. So all marine vessels need to
39:09
have the system where they broadcast their position and
39:14
some other metadata like what they are and things like that,
39:17
and also be able to receive. So all the secret drones will
39:20
also have on them obstacle avoidance abilities, of course,
39:24
because we'll have the ability to analyze the the camera fits
39:28
anyway into object detection and tracking.
39:31
So I don't think there's going to be an issues.
39:34
I think there's more probably an issue going to be with maybe
39:36
sharks. Yeah, especially if it's shaped
39:40
like a surfboard. Then yeah, I also be a little
39:44
bit different shape. It's going to be more like a
39:47
like a catamarang kind of shape. But yeah, sharks, you know, if
39:51
there's some something dangling from it, who knows?
39:55
And we, we spoken with, we spoken with, with, with another
40:02
Research Institute and they operate some of these, these
40:06
Usbs. And they brought one back off
40:09
the ocean once, and they found a shark tooth embedded in the side
40:13
of it. Oh, wow, yeah, I imagine not
40:17
just sharks, but a lot of different creatures are going to
40:20
want to go up to it, get to know it and poke it a bit.
40:24
Yeah, I figure out what it is, try and dunk it or take it home
40:28
with them. And.
40:30
Yeah. And we and we plan to have a
40:31
live video feeds as well. So eventually, I mean, at the
40:34
moment, you know, there's stalling out there, which is
40:37
what's going to be transmitting our data.
40:38
But in the, in the near future, I think Elon Musk has mentioned
40:42
that there's going to be a next version of stalling and it'll
40:47
have capabilities of 10 times more bandwidth than current
40:50
systems. And, and from that point, we'll
40:52
probably be able to offer a live video feed directly from
40:55
anywhere in the ocean, essentially.
40:57
And so to make some good entertainment, being able to
40:59
like go to a website and click on play button and, and, and get
41:03
a live feed of the Arctic. Or maybe you get a notification
41:07
that some shark is attacking for your, your equipment.
41:10
You can get a live camera feed in there.
41:12
Or just the school of fish swimming by, like you said.
41:14
If it's able to tell you all of that and it can identify it like
41:19
you were saying, then that would be incredible to look at.
41:22
And one for the next Planet Earth series, yeah.
41:27
Yeah, lots of data on there. You've been getting all kinds of
41:33
data for a while now. What are some of the most
41:37
unexpected things you've seen? Yeah, so the, the the most
41:41
unexpected 1 was OK and and a picture picture camera pointed
41:48
30° off the horizon up at the you know at the sky.
41:52
There's a there's a tree coming like that down the the, the the
41:58
right hand, the left hand side of the of the of the frame and
42:03
an object comes in it moves with like a constant speed stops sits
42:09
there for a minute dead still. I can I can zoom in with the
42:12
pixels. I can't see any wings flapping.
42:14
I can't see any rotation of propellers or anything.
42:18
Sits there and then moves back up, goes to the top, sits there
42:23
again for like another minute and then just flies straight up
42:26
and disappears. Of course, we analyzed all the
42:31
different cameras that we had available to see if we could see
42:33
that object and other cameras and we couldn't.
42:37
We looked at the audio and we couldn't really.
42:41
We couldn't hear anything for that time period that kind of
42:46
correlated with what we were seeing.
42:49
We know that there are red tailed Hawks in that area, so it
42:54
is possible that that was just a red tailed hawk.
42:56
The event could at around noon if I'm not mistaken, and that's
43:00
roughly coincides with the time that these birds like to go out
43:04
hunting and they can have a for extended periods of time.
43:08
So in all likelihood it probably was a red tailed hawk, but I've
43:14
never seen anything behave like that on a video feed.
43:18
And because we couldn't see any like, you know, you zoom in on
43:24
the pixels and you can't see any kind of this distortion.
43:26
There's nothing moving. It's just static and it's round.
43:29
So if it. Was what I was going to add.
43:33
We thought maybe drone as well, but you know with drones you
43:38
have to register where, where you're flying and the
43:42
instrumentation that we have there is not far from an airport
43:47
within 5 miles, let's say of an airport.
43:50
And it's unlikely that there'll be anyone flying any drone
43:54
around there also because this is nothing out there, just
43:56
trees. Have you seen any other like
44:00
that that got you that basically showed, hey, this stopped and
44:06
then it moved in a way that I'm not used to seeing it moving?
44:12
Yeah, there's a, there's a few that that that paved like this.
44:16
But usually what happens is the objects are we don't see
44:20
anything, like I haven't seen anything yet, just like zipping
44:22
off or doing anything crazy, any kind of crazy maneuvers.
44:28
But we do find things sitting, hovering and then moving away or
44:33
sometimes like disappearing and I'm not entirely sure what.
44:37
Yeah, like it gets smaller and disappears.
44:40
So we don't have that. Maybe it's flying away from us,
44:42
or maybe the light goes off. Unfortunately, we don't need to
44:45
have the triangulation. So until we can get that
44:48
triangulation, we can't determine the object's size and
44:50
its distance. How?
44:53
How is the triangulation gonna work?
44:55
Well, we're gonna have to have like two or three other cameras,
44:59
like separated by some distance. And then you can, you can busy
45:04
triangulate and figure out where that object is with, with with
45:07
respect to those two cameras. It's a parallax effect.
45:11
Man, this is fascinating stuff and this is so cool to see that
45:15
you get to work on this every day and get up and think about
45:23
and look at, look at the data, think about how you're going to
45:26
capture more data on this. And, and really I'm excited
45:30
about Seeker. Once that gets up and online,
45:33
let me know because it's super cool project.
45:36
Really think that you're doing a great job with that and I
45:40
appreciate you coming on here. Sure.
45:42
Thank you so much. Yeah, there's, you know, I know
45:44
there's a lot of interest out there with the UAP and also
45:47
machine learnings and just general kind of technology.
45:50
So if anyone is out there and interested and feel free to get
45:54
in touch because we're always looking for extra, extra hands
45:57
so. Nice, yeah.
45:59
Get involved. And do you see other areas where
46:04
besides the Galileo project, where you can leverage AI and
46:09
machine learning to help answer some of these bigger questions?
46:15
Yeah. So I think 11 cool idea would be
46:20
to use the the actual images coming from telescopes and do
46:25
some sort of machine learning on those.
46:27
I don't think much is actually being done in that space at the
46:30
moment. The images are typically kind of
46:33
grainy and black and white, but as telescopes improve, you know,
46:39
the data fidelity will improve as well.
46:40
So there might be scope for maybe doing classification on
46:45
different types of objects in space and also looking for
46:47
anomalies in that way or, or, or looking to see if they maybe
46:52
they, they, they move to traverse the sky and then stop.
46:56
Because that's another thing astronomers are, they're
46:58
building these, these algorithms out there to, to be able to see
47:02
if something is, is moving across the sky, but they're not
47:04
taking consideration that it might stop.
47:06
Like if it stops, they'll lose it.
47:08
Or if it takes a right angle turn, they'll, they'll lose it.
47:11
They've got algorithms designed and built to, to detect what
47:16
they know is out there. You know, so more work on kind
47:22
of anomaly detection algorithms, I think is, is going to be
47:25
fascinating. It's something that's solely
47:26
needed because anomalies is really what drives progress,
47:30
right? I mean, you see something
47:32
unusual and you think, oh, what's that?
47:33
And you and you investigate and it's only then that you
47:35
actually, you learn something. So more work on anomaly
47:40
detection all around I think is is the way to go both in in
47:44
space and on Earth and everywhere else.