Latest / Elon Musk Podcast / NASA will use AI in current and future spaceflight missions to the Moon, Mars and beyond.

Transcript

NASA will use AI in current and future spaceflight missions to the Moon, Mars and beyond. · Elon Musk Podcast

0:01
Hey, everybody. Welcome back to the Elon Musk
0:04
Podcast. This is a show where we discuss
0:07
the critical crossroads that shape SpaceX, Tesla X, The
0:11
Boring Company, and Neuralink. I'm your host, Will Walden.
0:20
Good afternoon. Thank you for joining us for
0:22
NASA's first artificial intelligence Town Hall, I'm
0:25
Melissa Howell. Today's all about offering more
0:27
insight and how NASA is using AI to build on its foundation of
0:31
innovative technology and exploration and the role AI will
0:35
continue to play here at the agency as we look to the future.
0:38
After the discussion, we'll take questions from the audience
0:41
right here in the room and those joining us online at conference
0:44
dot IO. To kick off today, Town Hall,
0:46
I'd like to welcome Administrator Bill Nelson for
0:48
opening remarks. Hey everybody, When a new
1:00
generation of tools reaches NASA's doorstep, we don't just
1:09
plug in and suddenly go into a whole new bunch of things.
1:18
We study them, we iterate them, and we advance them for the
1:24
benefit of all. We work with partners across
1:29
industry, across universities, across our government, across
1:35
the world to find ways to improve them.
1:41
And so we do what NASA does best.
1:47
The technology that we have today is astounding.
1:53
Take for example, the phone in your pocket or even the phone
1:58
that you're using to watch this town hall.
2:02
It has over a hundred 100,000 times the processing power of
2:08
the Apollo Guidance computer. The technology, of course, that
2:14
took us in Apollo 11 and put Neil and Buzz on the surface of
2:19
the Moon. And we're talking about just the
2:23
phone that we use to send emails.
2:28
So imagine what we can do today as NASA dives into the next
2:34
generation of artificial intelligence.
2:38
We have used AI safely, sustainably, successfully for
2:44
decades. Yet today this technology is
2:49
transforming before our eyes. There is new promise in
2:55
artificial intelligence, for example, machine learning tools
3:01
like neural networks, deep learning, generative AI, and
3:07
modeling. And as this technology grows, so
3:13
too does our capacity to use it, to test it, to refine it, and to
3:19
integrate it into our work to try to benefit all of humankind.
3:28
When used right, AI accelerates the pace of discovery, and it
3:36
can support our missions. It can drive our research.
3:39
It can analyze our data. It can support our spacecraft,
3:45
aircraft, science, and a lot more.
3:49
And it can open new possibilities in our ability to
3:54
land on celestial bodies and to navigate to them, to peer into
4:01
the vast corners of the cosmos and even in the search for life.
4:07
And remember, that's a statutory requirement of NASA searching
4:11
for life. That's why we're digging on Mars
4:13
right now. That's why we're looking for
4:17
exoplanets. And it goes on and on.
4:22
That's why we're sampling that asteroid sample from Bennu.
4:29
Well, AI can make our work more efficient, but that's only if we
4:36
approach these new tools in the right way, with the same pillars
4:42
that have defined us since the beginning.
4:46
Safety, transparency, and reliability.
4:53
Those three things matter not only to NASA, but to the
4:59
president and to the vice president, and they are focused
5:05
on making sure that when our government uses AI, including
5:15
these emerging technologies, that we do so safely, securely,
5:22
transparently, and responsibly. So consistent with the
5:28
president's executive order on artificial intelligence that he
5:33
wrote last October, we recently announced our new chief AI
5:41
officer. You're going to get to meet
5:45
David. And I thank David and all of our
5:49
panelists who are going to share with you today about where we're
5:54
going with AI. When new tools arrive at our
6:00
doorstep, we test them, we improve them, and we deploy them
6:05
to hopefully better serve humankind.
6:12
There's a lot of risk with AI because if it's employed in ways
6:19
that are not for the betterment of humankind, then it could be
6:26
disastrous. But the way we're doing it is
6:30
the way that we employed technology for the very first
6:35
step on the moon 55 years ago. And that's how, using every tool
6:41
at our disposal, we're going to make leap after giant leap in
6:46
the decades to come. And that's going to be how we
6:51
lead. And so I want to bring up our
6:56
deputy administrator, Colonel Pam Melroy, a person you all
7:03
know whose resume far exceeds her as a space shuttle commander
7:10
and an Air Force test pilot, and on top of that, a great team
7:16
member and as we refer to ourselves as a crew, a great
7:21
crew member. Come on up, Pam.
7:31
Thank you so much, Sir. It's a not just an honor, it's a
7:36
pleasure to serve. So thank you.
7:38
I'm really excited about today. You can probably guess you, you
7:42
guys all know I'm really a nerd at heart.
7:44
So moments like this are very exciting for me.
7:47
So I just want to say thank you to everybody for joining us for
7:50
this town hall. Of course, we're gathering today
7:52
to discuss one of the most transformative and exciting
7:57
frontiers and technology, artificial intelligence.
8:01
At its core, AI is an advanced form of statistics and
8:06
probability that creates the capability for a computer system
8:10
to perform complex tasks that have traditionally required
8:14
human intelligence to integrate, from reasoning to decision
8:18
making and even cool things like creative endeavors and art.
8:24
At NASA, Unsurprisingly science and technology Organization,
8:28
this is not just a buzzword. We have already harnessed the
8:32
power of AI tools to benefit humanity by safely supporting
8:37
our missions and our research projects, analyzing data to
8:42
reveal underlying trends and patterns, and developing systems
8:47
that are capable of supporting spacecraft and aircraft
8:51
autonomously. Unsurprisingly, in a poll across
8:55
the federal government, NASA had more use cases than any other
8:59
federal agency. I can tell you, I want to give
9:03
kudos to our Associate Administrator for Science,
9:06
Doctor Nikki Fox. Five years ago, I heard her give
9:10
a talk about the use of AI in heliophysics data.
9:14
That blew my mind. It was awesome.
9:16
So we're already doing this, and we're doing a lot of it.
9:20
But as we push those boundaries and we continue to find exciting
9:24
new ways to do use AI, we have to recognize the importance of
9:29
responsible usage like we do with any other disruptive
9:32
technology. So as was said most recently by
9:36
Spider Man, with great power comes great responsibility.
9:40
So we're totally committed to that.
9:43
We have to ensure that what we are doing in our AI initiatives
9:47
are guided by robust governance and protective measures.
9:52
So our goal of course is to harness the benefits of AI for
9:55
the betterment of humanity. But we have to do the other
9:58
side, which is to safeguard against the potential risks,
10:02
which are unfortunately all too real from unintended bias and
10:07
data collection or things like data compression, which ends up
10:11
resulting in less than accurate answers.
10:14
The consequences can range from funny, think auto correct on
10:19
your phone to very serious impacting human health and
10:24
safety. The administrator mentioned the
10:28
president's executive order. So as hard as we're leaning in
10:32
on the technology, we must be steadfast and lean in to
10:37
formalize the processes and the protocols for AI usage to make
10:42
sure that we're taking advantage safely of AI innovations across
10:47
the agency. So having that governance
10:49
structure is critically important.
10:52
It provides us procedures and guidance from the agency.
10:55
And actually, when you put that kind of structure in, it can
10:59
actually empower you, our workforce, and unleash your
11:03
innovation in a way that we all feel comfortable.
11:06
We're doing it the right way. Our artificial Intelligence
11:11
Working Group has been championed by three powerhouse
11:15
leaders at the agency, our Chief Technologist, AC Terrania, Chief
11:20
Scientist Kate Calvin, and our CIO Jeff Seaton.
11:25
This working group has been looking at this executive order
11:28
and developing recommendations in response to the directive
11:32
that are very focused on NASA, on our mission, how to best
11:36
enable our mission. The other exciting step that we
11:40
took is we named David Salvagnini as our chief
11:43
artificial intelligence officer, and I think his phone started
11:47
ringing off the hook right away, maybe even before we announced.
11:51
This is really just to under score the commitment that having
11:54
a focus area and taking this aspect of governance seriously
11:59
is really important to the leadership team and it's
12:02
important to all of us. So just so exciting when you
12:07
think about the future and all the interesting use cases that
12:11
we have for AIA. Pivotal aspect of our approach
12:15
involves how AI can enhance collaboration within our
12:18
workforce. That's a really exciting
12:21
opportunity. Communities of practice, working
12:24
groups and other avenues. We aim to foster a culture of
12:27
knowledge sharing about AI and collective learning.
12:32
And thank you. We have seen a significant
12:34
increase in the number of AI trained employees.
12:38
Yes, we do have formal training for this and this number is
12:41
poised to grow. And I encourage those of you who
12:44
are listening who have not taken this training to look into it.
12:48
This is all part of this commitment to responsible AI.
12:51
And it goes actually beyond our internal operations.
12:55
We're looking at partnerships with regards to AI leaders in
12:58
the private sector and in academia, recognizing that this
13:03
is incredibly important to collaborate to drive cutting
13:06
edge developments that impact the largest problems of our
13:11
time. So our participation in
13:13
initiatives like the National Science Foundation's National
13:17
Artificial Intelligence Research Resource Pilot is just one great
13:21
example of how we're dedicated to also leveraging our partners
13:25
and collaboration to advance AI on a much broader scale,
13:30
impacting not just NASA's mission, but the whole country.
13:34
AI is going to help us in so many areas, analyzing
13:38
heliophysics and Earth science imagery, probing the depths of
13:41
space with our telescopes for new insights, having them work
13:45
together to schedule communications on our networks.
13:49
We all know our communications networks are so loaded with
13:53
fascinating information, but we can barely squeeze it all in.
13:57
AI can help with that and just all kinds of other exciting
14:02
things, helping crews in the future on the way to Mars.
14:06
And really just think about, we don't even know yet what new
14:10
insights we're gonna get by using these new techniques to
14:14
look at old data in new ways. Again, establishing a very
14:20
robust governance structure and empowering you.
14:24
We can harness that full potential to fuel our own
14:27
missions. Dr. Innovation and continue to
14:30
change the world. So as I close, I just want to
14:33
emphasize it is a powerful, ingenious and very exciting
14:37
tool. But if we don't manage it
14:39
responsibly, we're going to open ourselves up to a world of risk
14:44
that jeopardizes our credibility and our mission.
14:48
Fortunately, this is something we are very good at at NASA.
14:53
We challenge ourselves to manage risk effectively, to push
14:57
boundaries and be innovative, to boldly explore as a team and do
15:02
it with the best possible risk management practices.
15:06
That's exactly what we're going to do with AI.
15:09
So I'm excited for you to hear today from the panel who really
15:13
knows what's going on, Jeff, AC, Kate, and Dave about how we're
15:17
using AI at NASA, how the field is changing, and how you can
15:21
learn more and educate yourself more and hopefully take
15:25
advantage of this technology. So thanks for joining us today.
15:38
Thank you, Deputy Administrator Melroy and Administrator Nelson.
15:42
I hope you all are ready for what is going to be a great
15:44
discussion. We're going to go ahead and
15:46
welcome our panelists to the stage.
15:47
Joining us today for the discussion, we have Doctor Kate
15:49
Calvin, NASA's chief scientist, AC Sharnia.
15:53
He serves as the agency's chief technologist.
15:56
And then we also have Dave Salganini, NASA's first
16:00
artificial intelligence officer, and our Chief Information
16:03
Officer, Jeff Seaton. Go ahead and give them a round
16:06
of applause for joining us today.
16:15
Thank you all for being here. We're looking forward to having
16:17
what will be an exciting and important conversation around
16:20
artificial intelligence. I want to start by giving you a
16:23
chance to talk a little bit about yourselves and your role
16:26
as it connects to AI. So I'm Kate Calvin.
16:31
I'm NASA's Chief Scientist and senior climate advisor.
16:34
As Chief Scientist, my office is responsible for representing and
16:38
enabling science across the agency, including the use of AI
16:42
to advance science. All right, I am AC Trinia, NASA
16:46
Chief Technologist in the Office of Technology Policy and
16:49
Strategy here at NASA Headquarters.
16:51
My job is to help shine a light on how we can be more
16:54
innovative, specifically with emerging technologies like AI.
17:00
Good afternoon. My name is David Salvinini and I
17:03
started here at NASA last summer as the Chief Data Officer
17:06
supporting the requirements in the Evidence Act and have since
17:09
assumed the role of the Chief artificial intelligence
17:11
Officers. So in both of those roles, I
17:14
will be looking after NASA's journey as it relates to meeting
17:18
some of the requirements from both executive orders and also
17:21
the OMB guidance, but more importantly, doing what's right
17:24
for NASA and addressing our journey and facilitating our way
17:30
forward as it relates to the use of AI in responsible and ethical
17:34
manner. Great and good afternoon.
17:37
I'm Jeff Seaton, the Chief Information Officer for NASA.
17:39
And I've got a responsibility for providing tools and
17:44
capabilities that enable the the work of the agency, leveraging
17:49
IT technology and also ensuring the security of our data and
17:53
systems across the agency and AI falls into to that sphere as
17:57
well. Thank you all.
17:59
And I want to start first with you, Dave.
18:01
First of all, congratulations on your new role as Chief
18:04
Artificial Intelligence Officer here at NASA.
18:06
Would love to hear from you about what your role actually
18:10
entails and what you're hoping to see as we move forward with
18:14
with your new role. No, thank you very much for the
18:16
question. So just a little bit of context
18:18
about this, if you're not tracking and I'm sure many of
18:20
you already are, the regulatory guidance coming from the
18:24
administration is not new. In December of 2020, they
18:27
released the Responsible AI Executive Order, which created
18:32
the need for a responsible AI official within federal
18:35
organizations and that really went after the need to be
18:38
ethical, responsible, transparent, safe, and and it it
18:43
accounted for accountability as it related to how federal
18:47
organizations use AI. Fast forward from there, we have
18:50
a new executive order last October, which now UPS the ante
18:53
to some degree and says, Nope, a responsible AI official is not
18:57
enough. What we'd like as a chief
18:58
artificial intelligence officer. And that person is going to be
19:01
responsible for ensuring the safety and rights of U.S.
19:05
citizens and also really driving agencies towards more innovative
19:10
practices and risk management. And then since that was issued
19:14
in in October, OMB released a memorandum in 28, the 20th of
19:20
March, which offered a lot of additional detail for how
19:24
federal agencies ought to address that.
19:26
So in part, my role is addressing those requirements
19:31
from the executive order and the OMB memo.
19:35
So what does that mean? Well, that means, you know,
19:37
establishing my role, establishing a leader who's
19:40
ultimately responsible for this journey at NASA, and then really
19:44
establishing some of those mechanisms that director, Deputy
19:47
Director Melroy and Administrator Nelson talked
19:51
about as it relates to how we manage risk associated with AI,
19:55
but also I would say manage risk while also managing opportunity.
20:00
So what do I do there? So I create situational
20:02
awareness. I would say the role is largely
20:05
an orchestration coordination role.
20:06
I think of myself if if NASA were a Symphony, I'm the
20:09
conductor. And I'm harmonizing the various
20:13
different instruments and sections of the orchestra in a
20:16
way where we're all rowing in the same direction.
20:19
We're situationally aware of what the other is doing, and
20:22
we're capitalizing on that knowledge of each other's work
20:25
so that we can build upon it as it relates to our specific
20:29
mission area. So there's that role, there's
20:32
certainly the compliance role, there is the role of standing up
20:36
governance and doing so in a mission informed manner.
20:39
So it's not burdensome, but it's a value add to the organizations
20:42
that look to leverage it. There's a work forcing component
20:46
to this. In other words, dealing with,
20:47
well, how do we equip the workforce to address the change
20:51
in AIAI has been around for years, but AI has also changed
20:55
dramatically in the last 12 to 18 months.
20:58
And there's some responsibilities that we all
21:00
have as it relates to how we use these tools responsibly,
21:04
ethically, transparently and so on.
21:07
So that imparts the responsibility to make sure
21:10
we're reskilling and upskilling the workforce to, to be aware of
21:15
the change and understand their responsibilities.
21:18
So I think that's a good synopsis of the responsibilities
21:22
associated with the role. So I'll stop there.
21:24
And that really ties into what Pam and Bill mentioned about how
21:27
NASA has been using AI responsibly and effectively for
21:30
years. And Kate, I'd like to pull you
21:32
in here. Can you share how AI has
21:33
contributed and been applied to scientific research?
21:37
Absolutely. So scientists have been using AI
21:39
for a long time. And one of the things that AI is
21:42
really good at is analyzing large data sets, like the data
21:45
that we get from our Earth observing satellites or space
21:48
telescopes or our other science missions.
21:51
And I want to hone in on one one of the ways that we use AI
21:55
around with what scientists like to call anomaly detection or
21:58
change detection. Essentially, when you look
22:00
through a data and you look for something that looks distinct, a
22:03
specific feature, and once you've identified that feature,
22:06
you can count it, you can track it, you can avoid it, you can
22:10
seek it depending on what your goal is.
22:12
So some of the ways we've used AI, one is to track wildfire
22:15
smoke. So we can teach the computer,
22:17
this is what smoke from a wildfire looks like, and then
22:20
look through all of our satellite imagery and find other
22:23
wildfire smoke and track that. Another way we've used it is to
22:27
count trees. So if you teach the computer,
22:29
here's a tree, find all of the trees.
22:32
And this is really important to use AI in these because for the
22:34
wildfire smoke with wildfire and other disasters, what we really
22:38
want to be able to do is do things quickly.
22:40
People need that information urgently and the computer can do
22:43
it faster than a human can. In terms of the counting trees,
22:47
you know, it's not tractable for a person to count every tree
22:50
across the country or a continent.
22:52
So instead, we can have the computer help us and then we can
22:55
focus on what do we learn once we know how many trees.
22:58
So that's in looking at, you know, observational data sets.
23:00
But we can also use AI to help improve models that that do
23:04
weather and climate forecast in the future.
23:07
And I focused a little bit on Earth science applications just
23:09
because I think they're easier sometimes for people to relate
23:12
to. But we use this throughout
23:13
science. Pam mentioned heliophysics, but
23:16
also in the science mission director, we do things like
23:18
counting exoplanets. It's the same type of thing that
23:21
we do with counting trees. But now we're looking out in the
23:23
universe and seeing what else is out there.
23:26
And AC want to pull you in here and talk about the Artemis
23:29
missions. Can you talk about how AI is
23:32
supporting NASA's efforts to put men back on the moon and then
23:36
eventually put humans, bring humans to Mars?
23:39
Yeah, thanks. Kind of.
23:40
As Bill and Pam mentioned, I think artificial intelligence,
23:43
large language models can be used in combination with humans
23:47
and not to replace humans. We've used AI for many, many
23:50
years with humans to examine that data and to make us smarter
23:54
about the universe. Relative examples to your
23:57
question about how do we use artificial intelligence, large
24:00
language models and the like in terms of Artemis include helping
24:04
us observe the Moon and Mars. In terms of surface imagery,
24:10
surface features. We can look at how do we
24:12
leverage these kinds of technologies and human machine
24:15
integration interface to help humans more intuitively and
24:19
better work with machines. We can look at how we use this
24:23
technology to help us communicate with spacecraft at a
24:26
large distance to alleviate, accelerate mission operations,
24:30
not to replace decision making, but to accelerate our missions.
24:34
We can also look at machine vision technologies and
24:37
capabilities that AI enables to help our Rovers on the surface
24:41
of the Moon or Mars traverse farther and faster.
24:45
And finally, I'm also very excited with the potential for
24:48
AI agents, large language model agents to help us optimize,
24:53
understand, manage, observe these complicated interconnected
24:58
networks that we're establishing.
25:00
Thus, for instance, in the absence of crews, can these
25:02
technologies helps help us observe and maintain Rovers,
25:07
space stations, habitats, the like throughout the solar
25:10
system? And Jeff, you know, we hear a
25:13
lot about the concerns and the risks across the industry when
25:16
it comes to AI. Can you talk about the use of
25:19
the tool right here at NASA and what we're doing to kind of
25:21
eliminate that risk? Yeah.
25:23
But first, let me take a step back because you heard the
25:27
administrator, he said that it's a new generation of capability.
25:31
And it definitely is AI is, is not new.
25:35
We've been using it, as you've heard for years in our existing
25:38
missions. But there is something that's
25:40
new, right? And it's becoming more
25:42
accessible to, to all of us. And if you think about
25:46
generative AI, one of the questions is what is it?
25:49
Well, think about when you were a kid.
25:51
If you're like me, you like logic puzzles.
25:53
And I can remember these logic puzzles that say, here are 5
25:57
numbers, what's the 6th number in the series, right?
26:01
Generative AI is a little bit like that, except instead of
26:04
this simplistic series of five numbers, it's this trove of
26:08
information that we can now access because of the networks
26:11
that are connecting systems worldwide.
26:14
So instead of a simple series, you have this volume of
26:17
information that we can learn from that the computer
26:21
algorithms can learn from. And because of the progress
26:24
we've made over the last 20 years in natural language
26:27
processing and understanding, we're able to take human
26:31
language and prompts and ask questions and have mathematical
26:36
models. It's it's math, it's not magic
26:38
mathematical models that can then generate responses to those
26:42
questions. And so there's a huge promise in
26:44
that. But you talk about the some of
26:47
the challenges to the risks. And one thing I want to say is
26:50
that these are computers and programs and algorithms.
26:54
We have existing processes and approaches to appropriately
26:59
secure data computers, programs and algorithms.
27:02
So we're not going to create new processes and policies to
27:07
protect our systems and software.
27:09
We're going to use the ones that we already have.
27:12
Goes back to understanding the sensitivity of the data, goes
27:15
back to understanding where is that data going, how is it being
27:19
transmitted, What systems is it connecting to?
27:21
And we're going to be able to leverage those same processes to
27:24
make sure that you're able to access these new emerging
27:27
capabilities in a safe and secure and a responsible way.
27:31
So from 1 standpoint, there is a lot that's new.
27:35
From a security of the system standpoint, there's not a lot
27:40
that's new. We're just going to apply our
27:41
existing processes to to be able to do that and to make these
27:44
tools available to you ultimately.
27:47
And coming off of that with so many tools that that are going
27:49
to be available, Kate, the importance of having humans
27:52
actually review the work and the research that AI is is doing.
27:56
Can you talk a little bit about that?
27:58
Yeah, so human review of AI products is really, really
28:01
critical. AI can make mistakes, and we
28:04
have to look through it and check them and make sure that
28:06
it's robust. And we've been using AI and
28:09
science and engineering for a long time, but scientists and
28:12
engineers have processes to check the products.
28:15
So we do things like code review, where if you use
28:18
something to generate a computer code, you have processes for
28:21
reviewing that code, for checking that it's accurate, for
28:24
testing it. We use peer review in the
28:26
science so that we actually peer review the document and get
28:28
people looking at it and making sure that it's correct.
28:31
And we need to think about those same things when we're looking
28:33
at AI. There might be some cases where
28:35
you look at an AI product and it's clear that there's a
28:38
mistake in it. So Pam was talking about auto
28:41
correct earlier. So if you see it something
28:43
that's auto corrected and it's not the word you meant, you'll
28:46
know it, right? You know what you wanted to say,
28:48
you'll see that it's incorrect. But when you're generating a
28:50
whole paragraph that's maybe outside your expertise, you
28:53
still need that review because you have to check that it's
28:56
correct. AI only knows what it's been
28:58
trained on, and it could be trained on things that are out
29:01
of date. It's also combining multiple
29:03
data sources, and sometimes it does that in ways that are
29:06
really powerful and helpful. Sometimes it makes mistakes and
29:10
we have to check it. So human review is really
29:12
important. And you know, Jeff, we we
29:15
received a lot of interest in generative AII know you spoke a
29:17
little bit about what that is. Can you go in a little deeper
29:20
talk about NASA's current policy and what we can expect to see in
29:24
the future as these tools are evaluated for safety?
29:28
Sure. And following on to what Kate
29:30
said, I think it's good to maybe kind of create two different
29:33
buckets. One is the the ad that's built
29:35
into our missions where you have program and project processes
29:38
that are doing the review and evaluation of how we're
29:41
leveraging the the capabilities. And then we have what is really
29:45
emerging now. And these are the, the general
29:48
tools that will be accessible to to so many of us to maybe do
29:53
some things in ways we never even imagined possible.
29:56
And so I think I'll focus on that latter category right now.
30:00
And that's you. You probably, if you listen to
30:03
the radio, listen to podcast, you can't go 30 minutes without
30:06
hearing an advertisement of some company that's saying they're
30:09
now the AI company right there. Everybody is an AI company.
30:12
And we're going to see AI capabilities built into many of
30:15
the tools that we're already using today here at NASA.
30:18
We do a lot of our work from a kind of business product
30:21
activity standpoint with Microsoft tools.
30:24
We use others, but Microsoft is is definitely one of the primary
30:29
providers of some of our capabilities.
30:31
And you've probably heard of things like Copilot.
30:33
Well, Copilot is a term that's an umbrella for a number of
30:38
different products and offerings and integrations that Microsoft
30:42
is building in to its product suite.
30:45
Similarly, many of you might use products from Adobe to create
30:48
images or videos or audio, and they're building in AI
30:52
capabilities into their products.
30:54
Many companies are doing that and, and we use many, many
30:57
vendors. So one of the things that we are
31:00
looking to do is evaluate how those vendors are building these
31:05
capabilities into their products.
31:07
And it can be a little bit frustrating sometimes though,
31:09
because at home you might have access to some of these
31:12
capabilities today because they're in the generally
31:15
available products. And within the government, we
31:18
have a few more requirements that we have to to make sure are
31:21
met before we use them. One example would be in the the
31:25
general utilization, a provider might be using data centers that
31:29
are worldwide, but for the government, we have to use data
31:33
centers that are within the continental US typically.
31:36
And so we have to wait for providers to create a version of
31:40
products that do meet the federal government requirements.
31:43
So we're working with many different providers to
31:46
understand where they are, understand what capabilities
31:49
they are building in and how they are protecting those
31:53
capabilities to ensure that we're being responsible with the
31:56
data that we have, the data that we are working with.
32:01
And while all that is happening, there's safety, right, Dave?
32:04
That's a priority as well. Can you talk a little bit about
32:07
how the workforce can actually play a role to make sure that
32:10
safety is made a priority in an effective way?
32:14
Yes, certainly so and and I would say it's, it's beyond
32:18
safety, but just responsible use, right.
32:20
So as Jeff talked about and Kate talked about an AC talked about
32:24
the use of AI and the many opportunities you know, there
32:27
are. I, I almost wanna, I wish
32:29
artificial intelligence was not referred to as artificial
32:32
intelligence. I wish it was referred to as
32:34
assistive intelligence assistive.
32:36
In other words, it is my digital assistant.
32:38
It is that resource that I now have access to that can help me
32:43
in my decision process. And Kate talked about this quite
32:46
a bit. So I think it's incumbent upon
32:48
all of us. You know, the AI is not
32:51
accountable for the outcome. The person is the human is
32:56
right, So and the vendor who may offer an embedded AI capability
33:01
as part of a product suite is not accountable as well.
33:04
You know, the user is the person who leverages that tool is
33:08
responsible. So we have to 1st own that and
33:11
we have to understand what responsibilities come with us.
33:14
You know, I was thinking about this a little bit earlier and
33:17
you know, I was thinking about hurricanes and I was thinking
33:19
about weather forecasters that released as it relates to
33:21
hurricanes and think about the validation that goes into sort
33:24
of hurricane forecasting. You know, and you've all seen
33:27
the forecast where they'll show that the projected track and
33:31
they'll be multiple lines and each one of those lines is
33:33
represented by a model. Why do they represent multiple
33:37
lines? Because they're cross
33:38
correlating the outcomes and trying to ascertain what do we
33:43
reasonably believe is, you know, an accurate projection and can
33:48
we assert with confidence what that projection is?
33:51
Well, it's based on the aggregation of those outputs.
33:53
So again, human judgement, not AI doing our job for us.
33:58
So I think that's the important part.
34:00
So then, you know, how do we be safe about this?
34:03
We understand our responsibility as the ultimate accountable
34:06
person as it relates to the use, as it relates to our work
34:09
products. And then if we happen to use AI
34:12
as part of the generation of a work product, that's fine, but
34:16
just understand its capabilities and limitations.
34:18
Another thing I'd like to talk about is, is, you know, there's
34:21
this notion of hallucinations where maybe the AI kind of gives
34:25
you a false answer and maybe it generated that false answer, but
34:30
there's that can be somewhat obvious and more easily
34:34
detected. I would be more worried about
34:37
errors of omission. So what about the AI that gives
34:40
you an answer, but there was a whole bunch of data that it
34:44
actually didn't reference. And maybe if that data were
34:48
referenced in the response from the AI, the answer would have
34:51
been different. So again, it's incumbent upon us
34:55
as we think about our use of these tools.
34:58
This digital assistive technology is going to be in
35:02
every part of our work day. It's already in many parts of
35:06
our life. I drove to work this morning.
35:08
I used a navigation application that, you know, referenced
35:13
traffic, OK. And by the way, I actually
35:17
didn't take the recommended course.
35:19
I applied judgement and I said, well, I know the flow of
35:23
traffic. I know preferred courses and
35:25
routing and so on. I'm gonna take the second
35:28
course. I applied judgement.
35:30
So I think that's the key here. We're ultimately accountable.
35:33
This is assistive technology and you know, we're not outsourcing
35:37
our thinking to the AI. We're applying judgement and
35:40
using it as data points to enable us to make good
35:43
decisions. And I'm, I'm glad you touched on
35:46
AI and hallucination and fabricating sources because that
35:49
was a concern that we heard from a lot of, a lot of people here
35:52
at NASA and AC turning to you. I mean, what does growth in this
35:55
field look like when it comes to missions and innovation and
35:58
opportunities here at NASA? I I think there's a few aspects
36:02
to this. One is inspiration inspiring us.
36:04
Technologies of all types inspire us to think differently,
36:07
innovatively. I also think we should leverage
36:10
these tools to help all of our directorates, including mission
36:13
support in terms of helping us think about how do we accelerate
36:18
our day-to-day practices separate from science and
36:21
engineering. And then finally, in some sense
36:24
we are an engineering and science agency, also a data
36:26
agency. And how can we uncover
36:30
knowledge, new knowledge from these various data sets we have
36:34
historically we're collecting today.
36:36
Our missions of the future are going to provide even more data
36:40
downstream that we can leverage and analyze from both human and
36:43
robotic missions across the board.
36:45
We should be working with experts of both internally and
36:48
externally to help us manage and ride this wave of innovation.
36:52
And finally, I think about this town hall I was mentioning to
36:56
someone earlier this morning. These are the kinds of town
36:59
halls I came to NASA for. These are the kinds of town
37:01
halls NASA should be having in the 21st century in terms of
37:05
leveraging technologies, leading the way of leveraging our data,
37:08
showcasing how we use it to other government agencies as
37:12
best practices or the practices of of use leveraging this
37:16
technology. So it's a pretty exciting time
37:18
at the agency of how we use these safely, responsibly, but
37:22
innovatively to achieve our mission faster and more bolder,
37:26
I think in more bold fashions if I could.
37:28
Add a little bit to what AC said too, because he talked about
37:30
here in the mission support realm, applying some of these
37:32
technologies and I mentioned we use Microsoft and today if
37:36
you're in a Teams meeting, you can already get an automated
37:38
transcript generated, right? It's listening to the audio and
37:41
doing a pretty good job of transcribing that meeting.
37:44
Well, one of the things that I would love to see happen and I
37:46
hope happens in the near future is we can take that transcript
37:49
and we can say give me a one page summary of that one hour
37:52
meeting and automatically the system gives that to you, right.
37:57
So that OK and we'll have to evaluate and see is that really
38:00
valid. But hopefully as the technology
38:03
matures, it will be a pretty valid synopsis of an hour long
38:07
meeting that is generated almost in the blink of an eye.
38:11
We've already been working to automate many processes over the
38:14
last years. We've saved 10s of thousands of
38:16
hours using existing technologies to automate manual
38:20
processes and that work is ongoing.
38:22
The mission support director at the NSSC working with various
38:26
organizations. And now if we are able to apply
38:29
these, you know, evolving AI capabilities, I think we'll be
38:33
able to automate much more. And I don't know about you, but
38:36
I don't hear a lot of people saying I've got so much free
38:39
time Jeff, give me more to do what I hear is I'm overworked
38:43
help, right? And so to apply some of these
38:46
technologies to address some of the the mundane tasks, to free
38:51
us up to do other tasks that we would love to get to, but we
38:54
don't have time to because we're wrapped up in the bureaucracy or
38:58
the process, right? I think there's a lot of
39:00
opportunity there. So it's not about replacing the
39:02
people, it's about enabling us to do more than we can today.
39:06
And I'm really excited about that.
39:08
And I, I want to pull the, the audience into the conversation
39:11
here, but before we do, I just want to see if there's anything
39:13
else that you all wanted to touch on or mention.
39:19
And here I am, the Chief data Officer and now Chief Data and
39:22
AI officer here at NASA. And I've talked less about data
39:25
than anyone else on the panel. So I want to say that, you know,
39:29
in Jeff's case of, let's say, the transcript of a meeting,
39:33
it's a very, very narrow data set.
39:35
You know, that's a pretty easy use case when you think about
39:38
it. What's much more complicated is
39:40
a large corpus of holdings, where you're now asking
39:43
generative AI to comb through a large corpus of holdings and
39:46
come up with some kind of logical conclusion to be able to
39:49
do that effectively. There's a a practice of data
39:51
management that's actually critical to the success of the
39:55
algorithm. You know, Eloquent algorithms
39:58
can really go bad with poor data.
40:01
So we also have a responsibility when we're thinking about our
40:05
use of AI is understanding the data that enables the AI to give
40:09
us the answer that it's providing.
40:12
And is that data complete or not complete?
40:15
Is it reliably sourced or not necessarily reliably sourced?
40:20
So think about, you know, AI also, you know, do we understand
40:24
the data? Do we understand the origins of
40:26
the data? Do we understand or have high
40:29
confidence in the accuracy of the data and its completeness?
40:32
Because if if the answer to those questions isn't yes, then
40:35
you know our confidence in the AI outcome should should be
40:41
declined or diminished. Yeah, I just add one thing.
40:45
So the four of us have had a lot of opportunities to talk about
40:48
AI over the last six months. And I think all four of us sort
40:50
of we're really excited about the opportunities that the
40:53
things like Jeff said that AI can do to help make our jobs
40:56
easier so that we can focus on the things that are new, that
40:59
are innovative. And I think we really are.
41:01
We're here to help. Yeah.
41:04
And I guess to add on to that a little bit about maybe what you
41:06
can do. And you asked a policy question
41:09
early and I don't think I really answered it right, because last
41:11
year I put out an agency policy on generative AI and said, hey,
41:15
at this point we don't have approved tools that are in the
41:19
environment. So no, we shouldn't be
41:21
installing things and using them.
41:22
And we're going to get to that. And we're still, you know,
41:24
working in that direction. But one of the things that was
41:27
noted is, hey, on your personal computers with, you know,
41:30
publicly available data, you can, you know, access and, and
41:34
so that's one thing I'd like to say right now, as we're
41:36
continuing to move forward to bringing tools into the NASA
41:39
environment, I would encourage you to play right now.
41:43
You can on your home computer, actually on your NASA computer,
41:46
right? We're not blocking certain
41:48
sites. You can go to
41:50
copilot.microsoft.com. You can go to ChatGPT today and
41:54
you can use your NASA computer. It's not a block site.
41:57
And you could just type in queries and questions, you know,
42:01
I'm taking a vacation this summer.
42:03
What's a five day itinerary for Mount Rainier National Park and
42:07
see what it says, right? Start to experiment with some
42:10
personal questions and things. Just be aware that you send
42:14
something out. It's out.
42:15
It's not in your possession anymore.
42:17
That's why we're saying, hey, no NASA sensitive data because it,
42:20
it's outside of our control. But on my phone, alright, my
42:24
personal phone, I've got a couple of these apps installed
42:27
and I asked ChatGPT to explain generative AI.
42:30
So let's have a little fun. So it did good, good, good
42:34
explanation. Then it said, OK, tell it to me
42:38
as a Limerick. So here we go.
42:41
There once was an AI so keen it learned from all things it had
42:45
seen it wrote and it drew made new from the old 2A marvel of
42:51
tech quite serene. And I said, OK, how about let's
42:56
tighten it up a little bit. How about as a haiku?
43:00
AI learns from all, creates new from what it knows.
43:04
Art and words unfold. So I would just encourage you to
43:10
experiment on your own, you know, little, maybe a little bit
43:15
of work. Time to go to some of the
43:17
available sites and start playing with some of the
43:20
prompts. Cause one of the things that we
43:21
see is the questions that we ask actually have a pretty
43:25
significant influence on the results that we get out of these
43:28
generative AI tools. And so prompt engineering,
43:32
right? What are the questions that
43:33
we're asking is going to be something that's going to be
43:35
important as we move forward. So I would just encourage
43:38
everybody to get a little smarter by experimenting a
43:42
little bit using non sensitive data, right?
43:46
Because pretty soon we're gonna have tools at your disposal that
43:49
will be able to use in our everyday work with sensitive
43:53
data internal to the agency or in approved cloud capabilities.
43:57
And the more you know now, the more prepared you'll be for that
44:00
future that's on the horizon. And if I can add one more thing,
44:04
Jeff and I are part of the NASA 24 Technology Work stream.
44:08
You may have heard of NASA 2040 of how we look at how the agency
44:12
should operate in the future. And on the technology work
44:14
stream, we're thinking about how do we work digitally leveraging
44:18
data sets. AI and Jeff and I are committed
44:22
to leveraging that work stream that the agencies provided us
44:25
under the guise of 2040 to help David and others using these
44:29
technologies. So I think there's a very great
44:32
synergy and we recognize it of this AI initiative, the NASA
44:36
2040 initiative, and how we can leverage 2040 to accelerate
44:40
everything we've talked about today.
44:43
Thank you all. And we're we're not done yet.
44:45
Got a lot more. If you're in the room and you do
44:47
have a question, we have a microphone up front and we would
44:50
invite you to come up and ask that question.
44:53
So while we give folks a chance to do that, we also received a
44:56
lot of questions at conference dot IO.
44:58
So I'm going to go ahead and give you guys a few of those so
45:00
we can get started. One of the questions when it
45:03
comes to ensuring that AI is being is being used is done in
45:06
an equitable manner. How will NASA ensure the AI used
45:10
by the agency doesn't develop unforeseen biases?
45:17
I can I can start with that question.
45:20
So I love the bias question only because we often think about the
45:24
bias as a negative and and I would offer this as an analogy,
45:28
right? Think about think about an
45:31
autonomous vehicle and think about the algorithm having a
45:34
bias. Do you want the autonomous
45:36
algorithm that drives your car to be biased towards speed and
45:42
performance or towards safety? I think everyone would admit
45:47
safety, right? We want some margin, we want
45:49
some threshold. We want to know that we're going
45:51
to be safe in the vehicle. So first of all, we have to look
45:54
at bias as there's positive bias, right?
45:58
And it's a very well worded question because it says
46:00
unforeseen bias. So what about the bias that we
46:03
don't understand? And by the way, with AI, you can
46:06
adjust that bias. So I can crank up or down the
46:10
margin of safety, OK? I can be very deliberate about
46:14
it. And this applies to so many
46:16
different use cases. So bias is a good thing when
46:20
applied appropriately. Now, unforeseen bias could be
46:23
something like, you know, address an outcome related to
46:27
some of the concerns the administration has referenced as
46:31
it relates to privacy. But let's say equitable
46:33
underserved communities being underrepresented in the outcomes
46:37
of AI. How do we address this?
46:39
You know, another thing could be age discrimination.
46:41
Think about a data set where the response to people who are,
46:44
let's say, older is different than the response to questions
46:49
about people who are younger. You have to understand the data
46:52
and you have to be on the lookout for those biases and you
46:55
have to test. So again, I go back to sort of
46:59
the accountability and one of the things we have to be careful
47:03
about as we onboard various different AI technologies that
47:06
are coming from our vendor partners is do we really
47:09
understand how they work? And have we thoroughly tested
47:12
them for bias and have we again understand the data?
47:17
Is the data leading the algorithm toward a biased
47:20
outcome or not? So that really is what it comes
47:24
down to. So we have often thought about
47:26
onborn technology from a security perspective in cyber.
47:29
Is it safe from a cyber perspective?
47:30
Now we have to add a set of attributes around, OK, is the AI
47:36
protected against bias? There are things like model
47:39
drift where AI can actually drift based on its use over
47:43
time, and we have to be on the lookout for that as well.
47:46
So part of what we'll be doing, and you'll see announcements
47:49
soon is the Summer of AI, which is a training initiative where
47:52
everyone in NASA is going to have an opportunity to learn
47:54
more about AI. It's literally a campaign.
47:57
It's going to be kind of a surge, if you will, of training
48:00
opportunity. So stay tuned for those
48:02
announcements, but I would encourage you, you know,
48:04
participate in those courses and learn about bias and learn about
48:08
how you can prevent some of the bias that would be unforeseen.
48:12
Learn about the data that enables an AI to do what it
48:15
does. And how do you, if you want to
48:17
take age bias out of an, an algorithmic outcome from an AI,
48:21
maybe you take the age parameter out of the data set, but you
48:26
have to be thinking about that. Or if there's other ways where
48:29
you can weight the outcomes differently as it relates to age
48:33
in that particular data set, that would be an alternative as
48:35
well. But again, just exercising
48:38
judgment and understanding the technology and not just treating
48:42
it as a black box and just assuming, well, it's seems right
48:46
most of the time, it must always be right.
48:49
Well, not necessarily. We've also heard from you all
48:52
today about, you know, there are so many organizations and tools
48:55
out there. This next question, how will
48:57
NASA's AI tools differ from AI tools used by the public?
49:01
And is NASA looking to partner with any of the leading private
49:04
AI organizations? Maybe I'll start with that.
49:06
In terms of the organizations, yes, I mean, I mentioned we work
49:10
with many, many different vendors and providers of
49:11
products. And so there are conversations
49:14
ongoing with some of those providers already in terms of
49:18
what their plans are and how those can potentially roll out
49:21
into the the government NASA environment.
49:24
As I mentioned, it can be a little bit frustrating because
49:26
it takes longer for tools, the kind of publicly available
49:29
generally known tools to to get into our hands in the government
49:34
environment. We are working on that front
49:37
with many, many different providers.
49:39
So there's that. And then there's another piece
49:42
going more towards the the mission technical side of
49:45
things. And we do have established in at
49:50
a FISMA low level, right. In terms of the the data
49:53
sensitivity, we have generative AI large language model
49:58
capability that's being you know, tested out by some some of
50:02
our folks within the agency. And we anticipate that by mid to
50:07
late summer we'll have that environment rated at a FISMA
50:10
moderate level. So we can start leveraging some
50:14
sensitive internal data and experimenting.
50:17
And so that's another, I think value in terms of engagement,
50:21
getting engaged with the AI community that is being formed
50:25
and that's that's growing right to be aware of what are some of
50:29
the capabilities that are emerging also to have your voice
50:32
in, Hey, what are some of the things we should be doing?
50:34
Right, because this is not the four of us up on this stage are
50:38
not the the know it alls as far as what's happening with Jenner
50:42
today. We've got a lot of smart people
50:43
across this agency and I think together we can guide some of of
50:46
the investments we've got limited time, limited resources,
50:49
we all do. And so how do we apply those
50:51
most effectively to advance this, you know, kind of journey
50:55
that we're on that's going to take involvement from all of us?
50:58
And we have a question here in the room, Sir.
51:02
Hi Moon Kim from OCFO. Thanks for this and super
51:04
excited to use Jenny AI or AI going forward.
51:08
Coming from a budget perspective since I'm from OCFO.
51:12
AI is not cheap, right? It cost GPU clouds, cloud
51:15
environment is expensive. In the middle of a budget
51:19
constrained environment. Do you foresee any of the AI
51:23
tools being openly available for everybody at NASA despite how
51:27
much budget you have in your your division in your office or,
51:30
or and do you see any tools that might be just free, like not
51:34
free, but Excel, you know, all attached to every single
51:36
computer like Excel or office. Any plans to navigate around
51:40
this? Constrained budget please.
51:43
Good question. I guess I'll start see if you
51:45
all have anything you want to add to that.
51:46
Margaret, the CF OS in the room somewhere I think right.
51:49
I saw her walk in there she is up front.
51:50
That was a question that you planted right.
51:53
So, and but that's a a great question because we we all know
51:56
that budgets are tight and companies need to make money and
52:01
so they'll be looking to leverage these new capabilities
52:04
to make profit. Sure that's that's what they
52:07
should do. So that's a valid question.
52:11
What we need to ask ourselves is the value of these capabilities.
52:15
So to back to what AC mentioned about 20-40 in the tech work
52:21
stream. So one of the things we're
52:23
taking a look at in that work stream is, hey, where should we
52:25
be investing more in technologies that can actually
52:28
enable the NASA mission to be successful maybe more rapidly,
52:33
maybe in different ways. And so the, the investment
52:36
question is a very real one. I think as an agency, we do need
52:41
to invest more in our technology, foundational
52:45
capabilities and some of these advancing capabilities.
52:47
And so honestly, that's a leadership conversation that
52:49
we'll be having, right. So as we take a look at some of
52:53
the products that we'll be rolling out and we see the cost
52:56
models and my guess is the cost models are going to continue to
52:59
evolve. That's what's happened in, you
53:01
know, the cloud world for the last five to 10 years.
53:05
The business models have evolved and changed and we've had to try
53:09
to understand and adapt to them. That's going to happen in the
53:11
generative AI space as well. So yes, I think we need to
53:15
invest. Yes, I think there needs to be a
53:17
set of capabilities that are just broadly available to the
53:20
NASA workforce. Go back to my example about the
53:25
summation of a hour long teams meeting, you know, if that is a
53:29
capability that rolls out, I don't think that the three
53:32
people that can pay for that should have access to it and
53:35
everybody else not so much, right.
53:37
So I think there should there should be and there will be a
53:41
general level of capability that we see value in providing to to
53:45
the broad workforce. And then I think they'll be
53:47
communities that will say there's value in the investing
53:50
in this tool that the whole agency doesn't need, but this
53:54
community does. And so we'll have to take a look
53:57
and understand what's available to us and how we optimize the
54:01
investments we make. Yeah, just following on a Jeff
54:04
statements on 2040 to give you a little bit of insight.
54:07
You know, we're having conversations with senior
54:09
leadership on these technology investments of how do we work
54:12
digitally. So we've divided those future
54:15
investment opportunities in various strings from
54:17
cybersecurity to data management to AI and having conversations
54:21
with Jeff and David and others to say in those dreams, what are
54:25
some investments over the next few years we need to start
54:27
making. And so I think I'm committed
54:30
personally along with Jeff and others to make sure that gets a,
54:33
a proper voice within the 2040 environment and they're offering
54:37
that to us. So I think that's a very
54:38
positive sign of recognizing Jeff, Scott, Jeff's budget over
54:42
the next few years. But how can we enhance it?
54:44
How can we enhance directorate funding to enable these tools to
54:48
come online and all these various streams that are pretty
54:51
important to how do we work digitally.
54:53
So that's going that, that is going on in terms of those
54:55
budget conversations, I think and those investment
54:58
opportunities. Thank you, super excited.
55:00
I, I might share one other thought and that is I talked
55:02
earlier about, you know, shared situational awareness.
55:05
The thing is let's not duplicate effort.
55:07
So if we know that one organization is exploring an AI
55:10
capability to enable some part of their business, you know,
55:14
having that situational awareness and not having to
55:16
recreate that elsewhere across NASA will go a long way to
55:20
helping us be efficient about how we pursue this technology.
55:24
So, and I think there's a governance activity related to
55:27
that as well. So when we think about shared
55:29
awareness, getting us all rowing in the same direction as it
55:32
relates to our pursuit of AI as a, you know, Team NASA, not as
55:37
individual organizations. Yeah.
55:41
And I think that that underscores a good point Dave
55:43
that I want to mention cause Dave in his new role as Chief AI
55:46
Officer, he he came and working in my organization as the Chief
55:49
data Officer, both data and AI. I think that's an apartment
55:53
description of of Dave's role. He's trying to facilitate and
55:57
coordinate for the benefit of the agency, right.
55:59
So that awareness, that understanding, pulling people
56:01
together to have the shared conversations is I think, a key
56:05
piece. And then it goes back to what I
56:06
mentioned about community. Getting folks involved, being
56:09
part of those conversations and helping us to frame up where
56:11
we're going as an agency is an opportunity I think that many
56:14
have. And we we have another question
56:17
here in the room. Hi, Jenny Modder, Art Director
56:20
for NASA Science. I work a lot with the creative
56:23
community, and we've noticed that uptick in generative AI
56:26
imagery falsely attributed to NASA appearing online and in
56:29
stock libraries. And I'm just curious if there's
56:31
any plans for authenticating NASA imagery to protect our
56:34
credibility or if that's a worthwhile investment in time
56:37
and energies. Thank you.
56:42
Soy there is a concern and everyone has probably heard the
56:46
term deepfakes right where you see the falsification of imagery
56:50
and even the the fact that AI can create an image that
56:53
actually didn't occur, a representation of something that
56:56
didn't occur. The good news, at least right
56:59
now at this moment in time, is, is that AI is actually pretty
57:02
good at detecting that, better than maybe a human might be
57:07
because there is a sort of a footprint associated with that
57:12
type of that type of misuse, if you will, of AI.
57:17
There are ways in which we can validate our content.
57:21
And I think this gets to also being careful about our security
57:25
and where we post content. And you know, if you want to go
57:29
and find authoritative representation of NASA imagery,
57:33
I would probably go to NASA as opposed to maybe a third party
57:37
where there's a potential that something could have happened
57:40
along the way. Other than to say that, I mean,
57:44
there are other more sophisticated ways of sort of
57:46
hashing and encrypting and doing other things with imagery,
57:49
digital rights protection type of technology.
57:52
We will have to look at that in our future as this is this is
57:55
sort of a fast evolving area. So your point is very well taken
58:01
and and we'll have to keep an eye on it.
58:02
Yeah. And I'd say this is, I mean it's
58:03
early days with the tools too and they are evolving.
58:07
So if you go back just a few months, some of the generative
58:11
AI tools that were out there would produce results, but they
58:14
wouldn't tell you how they generated those results.
58:16
Where did that come from? And now some of the tools will
58:20
cite that this is an AI generated image or they will
58:23
reference at the end of the output, they give you the three
58:28
or four primary sources that were used to derive that result.
58:32
So the tools are evolving too, I think.
58:34
So some of the concerns that that we have that are
58:36
legitimate, right? I'm hopeful that the
58:40
organizations developing these tools will help with some of the
58:43
solutions as well so that we won't have to create them
58:45
ourselves necessarily, but we'll be voicing the concerns.
58:49
And just one last thing to add to that, I think this is an
58:51
ongoing discussion, you know, throughout the country and the
58:55
world about what do we do about this and how do we identify.
58:57
And it's something that we're tracking.
58:59
I've been following a lot sort of the university professor
59:02
discussions around how do you identify AI?
59:04
And there are these AI tools. There's also sorts of things to
59:07
look for that identify it. And we're also establishing
59:10
these methodologies, like what Jeff said, where you actually
59:13
acknowledge your use of AI. And so I think we're going to,
59:15
we'll keep tracking those conversations, you know, across
59:18
the US government with the university community.
59:20
And this will evolve over time, but it is something we're very
59:22
aware that people are concerned about and we're watching.
59:26
Thank you. And I want to ask you all a
59:28
question that's coming from conference dot IO.
59:31
Is there any intention of obtaining a closed off
59:33
deployment of an LLM that could be used with internal data?
59:37
Yeah. And I, I touched on that little
59:39
bit, little bit earlier. We do have one that's rated at
59:42
the FISMA low level working to get that to be able to handle
59:45
some sensitive data. I will say that's not at a
59:48
production capability level though.
59:49
It's for sort of a, you know, AI early adopter kind of experts
59:55
experimentation phase right now. So we need to take a look at
59:59
within the agency, what kinds of capabilities do we need, do we
1:00:03
want, do we invest in going back to that resource question?
1:00:06
And then figure out what that looks like in terms of
1:00:08
deployment. But that's not to to prevent.
1:00:11
We have a lot of great mathematicians, computer
1:00:14
scientists with the agency, these large language models,
1:00:17
some of them are available, right.
1:00:19
So we might be doing some internal experimentation in
1:00:21
various organizations as well as we go forward.
1:00:24
But I do think it will be important to have a kind of
1:00:26
inside the wire capability that we're experimenting with.
1:00:30
And then in some cases for our missions, we'll be deploying
1:00:33
those. I think the other thing I would
1:00:36
add about the experimentation piece of it, why that is
1:00:38
important. Our contractor community,
1:00:41
academic community are using these tools and if we want to be
1:00:44
smart buyers, we need to understand how these tools
1:00:47
operate even in experimental mode if not in a production
1:00:50
mode. So once again, I think important
1:00:52
to have that inside the wire for multiple reasons.
1:00:55
And then I might just add one other thought and that is from a
1:00:58
foundational model perspective, I would fully expect that
1:01:01
they'll be foundation model development specific to NASA use
1:01:05
cases. I was at a conference earlier
1:01:07
this week where a colleague from DoD in the intelligence
1:01:10
community was talking about some of the concerns that they have
1:01:12
with some of the at a box models and not really understanding the
1:01:17
underlying process within the model itself.
1:01:20
And they were actually talking about perhaps developing models
1:01:23
for themselves as well for that very reason.
1:01:26
If we develop it, we understand how it works.
1:01:28
If we buy it, we may not understand it to the same
1:01:30
degree. So again, an emerging area.
1:01:33
And this is an area where we can learn across mission
1:01:35
directorates because the Science Mission Directorate has done
1:01:37
some work with foundation models and partnerships externally to
1:01:40
build them. Thank you all.
1:01:43
Anything else that you'd like to add or mention, I'll say.
1:01:47
That it's exciting. There's a lot going on tomorrow,
1:01:51
actually, we have our inaugural Artificial Intelligence
1:01:54
Strategic Working Group session kick off.
1:01:57
That's at 2:00 PM Eastern. Organizations have been asked to
1:02:00
identify a working group member. This is the beginnings of our AI
1:02:06
governance here at NASA. What we wanted to do is we
1:02:08
didn't want to start at the top and work down.
1:02:10
We wanted to start at start at a lower level and work up.
1:02:13
The other thing we're going to be doing at that AI strategic
1:02:16
working group is we're going to be doing a spotlight series
1:02:18
where we highlight various different levels of various
1:02:21
different activities that are occurring across NASA and make
1:02:24
people again, situationally aware that, oh, Ames is doing
1:02:27
this, Langley is doing that, God is doing something else.
1:02:30
Oh, science is working in this particular area.
1:02:33
That shared awareness will go a long way to helping us
1:02:35
understand not only the opportunity space, but also what
1:02:39
the technology can do to, well, understanding the technology
1:02:42
better. So there's that.
1:02:43
I already mentioned the summer of AI as an opportunity.
1:02:46
So I would say get involved there on on teams.
1:02:50
You'll find an AI community of practice.
1:02:53
Feel free to participate, join that.
1:02:56
That community practices as robust as the people who
1:02:59
participate and post and share and so on.
1:03:02
So lots of opportunity. Get to know your AI strategic
1:03:06
working group member for your organization and ask them how I
1:03:10
can help. And then please pursue the
1:03:12
training. I would love the statistics on
1:03:14
the AI summer of learning to be off the charts as far as how
1:03:18
many people consume that training and how many courses
1:03:21
people actually complete throughout the period.
1:03:23
So thank you. Yeah.
1:03:25
And I was at a conference recently and this is going to
1:03:27
build on what Dave just said. And there were discussion was on
1:03:30
AI and there it was a physician surgeon on the stage and she was
1:03:35
asked, is AI going to replace you?
1:03:39
Is AI going to replace doctors and surgeons?
1:03:42
And her response was great. She was like, I'm not worried
1:03:45
about AI replacing me, but I do think what's going to happen is
1:03:51
surgeons and doctors who don't learn to work with AI are going
1:03:56
to be replaced by surgeons and doctors who have learned to work
1:04:00
with AI. And so I think that's true for
1:04:02
us in terms of the future is going to involve us continuing
1:04:05
to adopt and learn new approaches and leverage
1:04:08
capabilities and technologies. And so I think it is incumbent
1:04:11
upon us to what they've just said, take advantage to learn to
1:04:14
see how we can grow personally, how we can maybe be more
1:04:19
successful, capable at our jobs because we're learning about
1:04:22
these new technologies. So I would just encourage all of
1:04:25
you to play experiments, have those conversations, take the
1:04:29
take some training classes over the next several months.
1:04:33
And I'm excited, you know, maybe six months from now what's
1:04:36
happening within NASA because we're taking advantage of some
1:04:38
of these capabilities. I'd like to thank our panelists.
1:04:42
Thank you all for joining us today.
1:04:43
And thank you for everyone who participated in today's town
1:04:46
hall. It has been an enlightening
1:04:47
conversation. And as we continue to explore
1:04:50
how artificial intelligence is moving NASA forward, NASA
1:04:54
employees can continue to engage.
1:04:57
We had a lot of questions that we did not get to, but we want
1:05:00
to invite you to to check out your point and continue to have
1:05:03
the conversation. And thank you again.
1:05:06
And we look forward to having more conversations around AI.
1:05:10
I'm Melissa Howell. Thanks for joining us, everyone.
1:05:27
Hey, thank you so much for listening today.
1:05:28
I really do appreciate your support.
1:05:31
If you could take a second and hit the subscribe or the follow
1:05:33
button on whatever podcast platform that you're listening
1:05:36
on right now, I greatly appreciate it.
1:05:38
It helps out the show tremendously and you'll never
1:05:41
miss an episode. And each episode is about 10
1:05:44
minutes or less to get you caught up quickly.
1:05:47
And please, if you want to support the show even more, go
1:05:50
to patreon.com/stage Zero. And please take care of
1:05:55
yourselves and each other, and I'll see you tomorrow.