[House Hearing, 118 Congress]
[From the U.S. Government Publishing Office]
[H.A.S.C. No. 118-66]
THE TECHNOLOGY AND AI FIGHT FOR
21ST-CENTURY OPERATIONS IN
THE DEPARTMENT OF DEFENSE
__________
HEARING
BEFORE THE
SUBCOMMITTEE ON CYBER, INFORMATION
TECHNOLOGIES, AND INNOVATION
OF THE
COMMITTEE ON ARMED SERVICES
HOUSE OF REPRESENTATIVES
ONE HUNDRED EIGHTEENTH CONGRESS
SECOND SESSION
__________
HEARING HELD
MARCH 22, 2024
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
------
U.S. GOVERNMENT PUBLISHING OFFICE
56-900 WASHINGTON : 2025
SUBCOMMITTEE ON CYBER, INFORMATION TECHNOLOGIES,
AND INNOVATION
MIKE GALLAGHER, Wisconsin, Chairman
MATT GAETZ, Florida RO KHANNA, California
LISA C. McCLAIN, Michigan SETH MOULTON, Massachusetts
PAT FALLON, Texas WILLIAM R. KEATING, Massachusetts
DALE W. STRONG, Alabama ANDY KIM, New Jersey
MORGAN LUTTRELL, Texas ELISSA SLOTKIN, Michigan
JENNIFER A. KIGGANS, Virginia JARED F. GOLDEN, Maine
NICK LaLOTA, New York PATRICK RYAN, New York
RICHARD McCORMICK, Georgia CHRISTOPHER R. DELUZIO,
Pennsylvania
Joshua Stiefel, Professional Staff Member
Michael Hermann, Professional Staff Member
Brooke Alred, Research Assistant
C O N T E N T S
----------
Page
STATEMENTS PRESENTED BY MEMBERS OF CONGRESS
Gallagher, Hon. Mike, a Representative from Wisconsin, Chairman,
Subcommittee on Cyber, Information Technologies, and Innovation 1
Khanna, Hon. Ro, a Representative from California, Ranking
Member, Subcommittee on Cyber, Information Technologies, and
Innovation..................................................... 2
WITNESSES
Martell, Dr. Craig, Chief Digital and Artifical Intelligence
Officer, Department of Defense................................. 4
Sherman, John, Chief Information Officer, Department of Defense.. 3
Skinner, Lt Gen Robert, Director, Defense Information Systems
Agency......................................................... 6
APPENDIX
Prepared Statements:
Martell, Dr. Craig........................................... 42
Sherman, John; and Skinner, Lt Gen Robert.................... 27
Documents Submitted for the Record:
[There were no Documents submitted.]
Witness Responses to Questions Asked During the Hearing:
Mr. Moulton.................................................. 55
Mr. Luttrell................................................. 55
Questions Submitted by Members Post Hearing:
Mr. Moulton.................................................. 59
Mr. Strong................................................... 59
Dr. McCormick................................................ 62
THE TECHNOLOGY AND AI FIGHT FOR 21ST-CENTURY
OPERATIONS IN THE DEPARTMENT OF DEFENSE
----------
House of Representatives,
Committee on Armed Services,
Subcommittee on Cyber, Information Technologies, and
Innovation,
Washington, DC, Friday, March 22, 2024.
The subcommittee met, pursuant to call, at 9:01 a.m., in
room 2118, Rayburn House Office Building, Hon. Mike Gallagher
(chairman of the subcommittee) presiding.
OPENING STATEMENT OF HON. MIKE GALLAGHER, A REP-
RESENTATIVE FROM WISCONSIN, CHAIRMAN, SUBCOMMIT-
TEE ON CYBER, INFORMATIONCHNOLOGIES, AND INNOVAT-
ION
Mr. Gallagher. Good morning, everyone. Welcome to today's
hearing, ``The Technology and AI [artificial intelligence]
Fight for 21st-Century Operations in the Department of
Defense.''
Since the start of this Congress, the Armed Services
Committee has held nine hearings dedicated to driving change in
DOD [U.S. Department of Defense] through adoption of new
technology and pivoting to an innovation-first posture.
No one, obviously, expects the Pentagon, the largest
bureaucracy in the world, to turn on a dime. It is an aircraft
carrier, not, you know, a little Corvette. But I don't think
any member believes the Department is moving with the speed,
force, or unity of action necessary to get our warfighters what
they need.
We have established that national security in this century
is contingent on rapidly embracing new technology,
experimenting, iterating, and pivoting from a platform-centric
combat to network-centric warfare. And yet there doesn't appear
to be any adversary as formidable as the change-resistant
culture in DOD.
My hope for this hearing is not only to hear about your
plans for the coming fiscal year but more so to understand the
precise obstacles and barriers to reform that you have seen,
that you have encountered on a day-to-day basis.
And as I have said before in this room, you know, for those
of us who attend the Reagan National Security Forum every year,
that you can almost, like, copy-and-paste the speech that the
Secretary of Defense gives there. It is all about, you know, we
are going to kill the Valley of Death. And we talk about it,
and the same people have the same conferences, and then it
doesn't actually get any better.
So, put differently, why it seems to this committee that, a
year later, the implementation of policies we have passed and
the directives on AI, zero trust, and software integration from
your bosses seem to have made little progress. And I assume we
all agree that speed is essential, and I hope that this hearing
fosters an honest conversation about what is holding you back,
what you are doing to go faster.
We are very pleased to have before us John Sherman, the
Department of Defense's Chief Information Officer; Dr. Craig
Martell, the inaugural Chief Digital and Artificial
Intelligence Officer; and Lieutenant General Bob Skinner, the
Director of the Defense Information Systems Agency and the
Commander of Joint Forces Headquarters-Department of Defense
Information Network.
General, you get the award for longest title after your
name.
And I will note that this is Dr. Martell's final appearance
before the committee, having announced last week that he is set
to depart his position on April 15th.
And before hearing from the witnesses, I yield to the
ranking member for any comments he may have.
STATEMENT OF HON. RO KHANNA, A REPRESENTATIVE FROM
CALIFORNIA, RANKING MEMBER, SUBCOMMITTEE ON
CYBER, INFORMATION TECHNOLOGIES, AND INNOVATION
Mr. Khanna. Thank you, Mr. Chairman. And thank you for your
bipartisan focus on how we make the Department more innovative,
agile, and capable of adopting the latest technology and
software and working with the private sector to remain the most
innovative military in the world.
I hope at today's hearing we can explore not just what the
Department is doing but what more flexibility is needed and how
we can have greater agility. The reality is, we have a
challenge of harmonizing this ecosystem, to bring it up to the
speed of modern innovation, while operating in our era of
great-power competition.
In particular, efforts such as securing the defense
industrial base, implementing zero-trust architectures,
accelerating the cloud transition and roll-out of a robust
identity, credentialing, and access management capability are
fundamental.
Two weeks ago, this subcommittee had a very interesting
hearing on software development and software acquisition. And,
a month ago, the committee held a hearing on accelerating the
pace of innovation across the Department.
As we approach the end of this term, I guess what would be
helpful to me is a few, two or three, very concrete
recommendations, particularly of what the committee can do or
needs to do legislatively to help with making the Department
more innovative.
Is it, at this point, that Congress has done everything it
needs to do and it is more a matter of implementation? Or do
you see a role for the committee to continue to do more
legislatively? And, if so, what would those recommendations be?
Thank you, Mr. Chairman, and I yield back.
Mr. Gallagher. Mr. Sherman, you are recognized for 5
minutes.
STATEMENT OF JOHN SHERMAN, CHIEF INFORMATION
OFFICER, DEPARTMENT OF DEFENSE
Mr. Sherman. Good morning, Chairman Gallagher, Ranking
Member Khanna, and distinguished members of the subcommittee.
Thank you for the opportunity to testify before you today.
I am glad to be here with Dr. Martell, as well as General
Skinner, as you noted, Director of DISA [Defense Information
Systems Agency] and Commander of Joint Force Headquarters-DODIN
[Department of Defense information networks].
Together, Lieutenant General Skinner and I lead a team that
provides direction, oversight, and technical expertise to
secure and modernize the Department's information technology;
enhance warfighting command, control, and communications, or
C3; manage the DOD's use of electromagnetic spectrum; and
cultivate a digital workforce.
We look forward to sharing the progress on the Department's
digital transformation efforts and discussing our key
priorities for fiscal year 2025.
As we have seen in Asia and Europe, the cyber threats we
face today are evolving, and we must keep pace to both support
the Nation's warfighters and protect key national-security
capabilities.
Protection of our networks and the networks of the defense
industrial base is critical. We are laser-focused on zero-trust
implementation. And, earlier this year, we expanded eligibility
for the Defense Industrial Base Cybersecurity Program that will
significantly enhance the cybersecurity posture for these
companies.
Cloud computing and software modernization remain central
to our IT [information technology] modernization efforts. When
I testified last year, the Department was just beginning the
enterprise cloud journey, and I am happy to report significant
and successful progress.
Through the Joint Warfighting Cloud Capability, or JWCC,
DISA has successfully awarded more than 47 task orders over the
last year, and over 50 more are in the pipeline right now. We
also publish DOD guidance to streamline cloud contracting and
reduce contract sprawl across the Department.
In today's environment, and as you all focused in on your
hearing last Wednesday, it is critical, more than ever, that we
provide DOD personnel with secure and resilient software when
and where they need it. We recognize the urgency of this issue
and are working hard to ensure we are successful.
Our software modernization strategy and ecosystem of more
than 55 software factories are transforming the way DOD
develops and delivers this force multiplier. This requires
changes to our processes, policies, workforce, and technology.
Accelerating the authority to operate, or ATO, and
strengthening reciprocity are absolutely key to this effort.
Just as important as the software is ability to operate in
any environment. Our adversaries have spent decades investing
in capabilities to make the best use of electromagnetic
spectrum, and that is critical to defend the Nation. However,
we also understand the increasing commercial demand for
spectrum and are working with the White House, Department of
Commerce, and other interagency partners to explore ways to
address increasing Federal and commercial demand for spectrum
access without compromising national security.
5G is also critical to both DOD and civilian industry. Last
fall, my office became responsible for this mission, and we are
accelerating the deployment of 5G on military installations,
advancing enterprise capabilities, and addressing resource
requirements.
Each of these missions and others are critical to our
warfighters and would be impossible without the right people.
This past year, we laid the foundations to enable DOD to grow a
dynamic and innovative workforce needed to succeed in the 21st
century. Our implementation of the DOD Cyber Workforce Strategy
is ongoing to ensure we work with industry to recruit and
retain the right people with the right skills for the right
jobs.
As people are our greatest resource, I understand their
interaction with the Department's IT infrastructure, weapons
systems, and business systems directly affect the mission and
morale of each warfighter, civilian, and contractor. To enhance
user experience, we established a Customer Experience Office to
harness resources and data to tackle this multifaceted
challenge.
Thank you for your consistent and dedicated support and for
the opportunity to testify this morning, and I look forward to
answering your questions.
[The prepared statement of Mr. Sherman and General Skinner
can be found in the Appendix on page 27.]
Mr. Gallagher. Thank you.
Dr. Martell.
STATEMENT OF CRAIG MARTELL, CHIEF DIGITAL AND ARTIFI-
CIAL INTELLIGENCE OFFICER, DEPARTMENT OF DEFENSE
Dr. Martell. Thank you. Chairman Gallagher, Ranking Member
Khanna, and distinguished members of the subcommittee, I thank
you for allowing me to testify here today and to share the work
the CDAO [Chief Digital and Artificial Intelligence Office] has
been doing.
I will read through this, but I want to start by answering
the direct question, which is, what have we been doing and how
can we make it faster?
We have spent the last year building what I could call a
virtuous cycle. That virtuous cycle is delivering value,
creating new demand, and then iterating.
And so I will talk through the tools that we have been
using to do that. But I think, to answer the direct question at
the beginning, we have to continue that virtuous cycle, and I
think that is what is going to increase the acceleration and
delivering value to the warfighter.
The CDAO is accelerating Defense-wide adoption of data
analytics and AI so that the DOD can make better decisions
faster, from the boardroom to the battlefield.
We create sustainable change at scale through two
functions: leading and overseeing--and this is our Principal
Staff Assistant function, mostly through policy guidance and
oversight--but also actively delivering capabilities across the
full range of the hierarchy of needs. So the hierarchy of needs
is how we analyze what needs to get done when.
This enables change at both speed and scale--speed, by
delivering apps and products directly to the end user quickly--
think both warfighters and running the business of the DOD--and
scale, by delivering platforms and services to allow builders
at the edge, the owners of the problem, to best solve their
problem.
So think about it this way: We are a centralized
organization. If our job is to build all of the AI of the
Department, that does not scale. We will build onesie-twosie
solutions. And we are tackling very large individual
solutions--JADC2 [Joint All Domain Command and Control], for
example. We can talk about Harbinger and other things that we
have been working on. And that is really important for the
centralized org with the expertise to drive those issues.
But it is also really important to create the tools,
policies, processes and best practices so that the folks at the
edge can actually deliver value when they need that value, so
they can deliver AI when they need that AI. They are the ones
who understand the problem; they are the ones who are going to
understand the solution. It shouldn't be only centralized. It
has to be distributed as well.
So we have been tackling both of these fronts
simultaneously. The first allows for winning the fight tonight.
The second allows for sustainability of those wins.
And we do this, as I said, through the hierarchy of needs.
The hierarchy of needs for us is that you have to get the data
right. So the lowest layer, the foundational layer is: Do we
have the right data? Is the data accessible? Is the data
understandable? And can people use that data to build apps that
provide solutions? So think about a separation between data and
apps.
The next level above that is simply analytics and metrics.
We spent a great deal of time this year getting people to move
from effort-based metrics to outcome-based metrics. And that
has allowed for the running of the Department in a way that we
haven't seen before. So every monthly meeting with the Deputy,
there is a dashboard with the metrics for that Principal Staff
Assistant and how well they are doing. And gathering the data
for that has been significant.
And on the top of that is AI, because without metrics to
know how well you are doing--because, remember, AI is
statistics at scale. You measure the past to predict the
future. You have to know how well you are predicting the
future. To do that, you need the metrics to measure yourself
against it, and you need the quality data.
So this hierarchy of needs combined with our agile approach
is how we are going to drive the sustainable change through our
virtuous cycle.
And that is a bunch of buzzwords, but what do I mean by an
agile approach? We have had major success on CJADC2 [Combined
Joint All Domain Command and Control] this year, in particular
by having data owners and industrial software engineers sitting
right next to warfighters, literally in the room, and as those
warfighters need something new, the data owners go find that
data, and the software engineers change the software, and it is
delivered within days. This sort of iteration, this sort of
agile iteration, is what is going to continue to drive that
virtuous cycle.
So we deliver by learning by doing, shipping fast and
iterating quickly, with warfighters, as I said, developers
sitting side-by-side. This is done on real networks with real
data to learn and deliver for real warfighters fast.
And we couldn't have done it without working closely with
our industry partners. And we have used a lot of the tools that
you have given us to be able to build contracts fast and to
procure their services very quickly.
So CDAO looks forward to working closely with the
subcommittee on these issues and others as we scale DOD's
current and future use of data analytics and AI for national
security.
Thank you.
[The prepared statement of Dr. Martell can be found in the
Appendix on page 42.]
Mr. Gallagher. Thank you.
General Skinner.
STATEMENT OF LIEUTENANT GENERAL ROBERT SKINNER,
DIRECTOR, DEFENSE INFORMATION SYSTEMS AGENCY
General Skinner. Good morning, Chairman Gallagher, Ranking
Member Khanna, and distinguished members of the subcommittee.
I am honored to be here today and represent the
approximately 19,000 personnel who support the missions of the
Defense Information Systems Agency and Joint Force
Headquarters-Department of Defense Information Networks.
I am also honored to sit alongside Dr. Craig Martell and
one of my two bosses, the Honorable John Sherman, a key ally
and partner in the campaign to innovate, modernize, secure, and
defend the Department's networks, systems, and data to achieve
and maintain information superiority.
Each day, we are energized and focused on ensuring the
Joint Force is in a position of advantage against any Nation or
group that desires to harm us or our allies as we set the globe
in the cyber domain.
Along with our key partners within the Federal Government--
allies, industry, research, and academia--we continue to
leverage lessons learned from the ongoing conflicts in Ukraine
and Gaza; the nefarious activities of our pacing threat, the
People's Republic of China; and global cyber events to
strengthen our digital technologies and forces.
Through the support of this committee, we have made
significant progress over the last year and look forward to
highlighting our future plans.
Resiliency, agility, survivability, and velocity are key
tenets in describing our initiatives, starting with our
Department's flagship Joint Warfighting Cloud Capability
contract, which was awarded in December of 2022 and provides
access to multiple global cloud fabrics that ensure our
warfighters can conduct operations anywhere in the world.
Along with the 47 task orders Honorable Sherman mentioned,
we have also successfully deployed an initial overseas cloud
capability in support of INDOPACOM [U.S. Indo-Pacific Command]
missions. Recently, we also kicked off a pilot, the Joint
Operational Edge, that will deploy hybrid cloud capabilities to
remote locations across the globe.
As discussed in the hearing you held earlier this month, it
is critical that we ensure that DOD personnel have secure and
resilient software regardless of where they are located. We
continue to develop agile and modern software development
methodologies with templates, tools, and automation in our
software factory integration platform which is available to the
Department.
Equally important to operational software availability is
our perimeter defenses. We have two key pilots ongoing with
commercial companies that are increasing our ability to see and
counter adversary activity, which are increasing each and every
day.
Along with these activities, we are lockstep with the
Department's zero-trust strategy through our Thunderdome
initiative. In 2023, we deployed Thunderdome to 15 sites on
classified and unclassified cyber terrain and will accelerate
deployment to more than 60 sites this year.
Thunderdome, which is part of our next-generation network
and data environment, will provide defense agencies and
combatant commands with improved user experiences while also
increasing cybersecurity by knowing who is accessing the
network and data, limiting individuals to only that data they
are authorized to access, enhanced data analytics through
artificial intelligence, and increased segmentation, like fire
breaks in houses that prevent fires from spreading.
While we continue to make significant advances, our work is
not done. Leaning forward, I believe our new strategic plan
outlining the Agency's fiscal year 2024 to 2029 goals sets the
North Star to enable best value capabilities to our Department
and our warfighters.
A final area to highlight is our commitment to no-fail
missions, such as the National Leadership Command Capability.
Within our strategy, DISA will deploy an integrated, multiple-
level, secure voice and video communications and conferencing
capability to provide direct support to senior leaders,
including the President, the Secretary of Defense, the
Chairman, and the nuclear command and control community.
None of these initiatives are possible without our bold,
innovative, and critically thinking workforce. In line with
DOD's Cyber Workforce Strategy, DISA has released our Workforce
2025 strategy that outlines key objectives to recruit, retain,
and professionally develop our team to ensure the right
personnel with the right skill sets are in the right positions.
Our overall readiness, strength and resilience, and
warfighter success relies on the strong support that this
subcommittee has provided for many years. I am grateful for
your support and the opportunity to testify this morning. I
look forward to your questions.
Thank you.
[The prepared statement of Mr. Sherman and General Skinner
can be found in the Appendix on page 27.]
Mr. Gallagher. Thank you.
We will now move to questions.
Dr. Martell, INDOPACOM has pieced together the Joint Fires
Network in the absence of a JADC2 solution or operational
capability.
Elaborate on how your organization--what you have done to
contribute to the JFN [Joint Fires Network]----
Dr. Martell. Absolutely.
Mr. Gallagher.--and, by extension, INDOPACOM command and
control. I guess, put differently, how do you see the joint
operating system and CDAO's JADC2 efforts fitting into the JFN?
Dr. Martell. Yeah. Thank you for the question, Chairman
Gallagher.
We work hand-in-hand with INDOPACOM--with R&E [research and
engineering] and INDOPACOM--with research and experimentation
and INDOPACOM on JFN. The underlying data layer, which drives
the right data to JFN to be able to make the decisions, to be
able to close chains, comes from the CDAO. Some of it was
theirs already, but they needed a whole bunch more.
We have a team that sits with them, and when they need more
data, we help find that data, we help deliver that data, and we
help make sure that data is there at speed and scale so that
they can use it on a regular basis.
So they were a key part of our last guide experimentation,
and it was in INDOPACOM. And the success and some failures in
that experimentation was due to that partnership. I am happy to
go into deeper about exactly what worked and what didn't work
in the closed session, but we are tightly aligned.
Mr. Gallagher. Can you elaborate, taking a step back, just
elaborate for us laypeople on the difference between the joint
operating system and the data integration layer?
Dr. Martell. I would say, the joint operating system is
part of the data integration layer.
So think about the data integration layer as two major
components: a catalog to help you find where data is; and APIs
[Application Programming Interface] that allow that data to be
served. That is really what--and that data has to, obviously,
sit on a cloud somewhere, so the work with CIO [Chief
Information Office] is really important there.
So the data is in the right cloud place. The data has an
API that allows an application builder to access that data when
needed. And it is discoverable. And not only is it
discoverable, when it is discovered, it is understood. So there
is metadata that describes what that data means, how that data
is used, and how that might be ingested into an application
that is being built.
JOS [Joint Operation System] does specific things that we
had originally contracted them for, and they are now part of
that data integration layer. They have data that they are
providing, Anduril has data that they are providing, that is
part of--that has an API layer and is discoverable through that
centralized catalog.
So you want to think about the data integration layer as a
multi-vendor, heterogeneous entity that really is the data that
is needed, is discoverable and accessible. And so that is what
we are driving forward.
Mr. Gallagher. And, then, in PB25 [President's Budget
2025], it says that the JOS is moving to production.
I like the idea of CDAO using all sorts of acquisitions
authority to move rapidly, taking commercial technology like
the JOS through prototyping and into production. But could you
elaborate more on how you see the JOS in production?
Dr. Martell. I may have to take that for the record or for
the closed session, because I don't actually know specifically
what we can talk about about what it delivers. But they
deliver--and my team can tell me if I am allowed to now.
But they deliver specific information that was necessary
for the Joint Fires Network. It is their hardware that is
distributed--it is Anduril's hardware that is distributed
across the INDOPACOM theater that allows that data to flow
easily. And it is discoverable through our catalog and through
our sets of APIs.
I think I didn't answer the question more specifically than
I did before. Sorry about that.
Mr. Gallagher. And you are leaving soon. As you look back,
what are the metrics that you would judge your tenure on?
Dr. Martell. Massive increase in demand for getting it
right. We just did a presentation across the whole Department
about the current state of JADC2 and the successes that we had.
A year ago, there were a lot of ``I don't know what this
means,'' ``I don't know what you mean by 'getting the data
right,''' ``I don't know what you mean by 'having the data
accessible to the warfighter at the right time and the right
place.''' That is now well understood.
CJADC2--I was very pleasantly surprised, there is no longer
an ambiguity, at least within our org, about what CJADC2 is. It
isn't a thing. It isn't a system. It isn't even a destination.
It is a set of behavioral patterns and the underlying dataflows
to support those behavioral patterns.
I mean, CJADC2 is command and control for the 21st century.
You do command and control in the 21st century by the right
flow of data. The demand for that data flowing correctly has
not quite exponentially but geometrically increased, I would
say, over the last year and a half. And if there is a win that
I am strongly willing to claim, it is that.
Mr. Gallagher. And in the 7 seconds I have left, how would
you respond just to the articles about the poor command climate
surveys?
Dr. Martell. Yeah, I think that is great and--that is a
great question and a really important one.
We did a very hard task. We took four organizations; we
merged them to get four--with very distinct cultures. And we
had to break a lot of expectations in doing that. There were
lots of people who wanted to do particular things, and we said,
``Yeah, no, we are going to look for economies of scale, and we
are going to do these things instead.'' That is a monumental
task for any merger and acquisition, and it is going to upset
some people.
We have worked really hard over the last year to work with
some industrial partners on getting our culture right, on
making sure that our teams are being heard, that communication
flows better. We did sort of biased execution, maybe, to the
expense of communication with our people in the beginning, and
I own that. I am looking forward to next month when the new
pulse surveys are released, because I feel confident that they
will say good things.
Mr. Gallagher. Thank you.
Mr. Khanna is recognized for 5 minutes.
Mr. Khanna. Thank you, Mr. Chair.
Dr. Martell, I see that you worked at Dropbox and Lyft
before your service to our country.
Can you describe some of the collaboration with the Silicon
Valley technology companies that you have and how that can
improve?
Dr. Martell. We couldn't have done the job without tight
partnership with key companies. Palantir is one. Anduril is
another. Databricks is another.
Most of the things--almost everything we do--and then all
of the cloud providers, so Google, Amazon----
Mr. Sherman. Oracle and Microsoft.
Dr. Martell. Oracle and Microsoft. Thank you, John.
Mr. Khanna. Please keep listing companies in my district.
No, I'm just joking.
Dr. Martell. I think every Silicon Valley company is in
your district, right?
So we couldn't have done it without them. We spent a great
deal of time talking to the leaders of those companies to get
buy-in for the story that I am telling you now.
Like, what I tried to do was take best practices from
Silicon Valley, which is, really, the data is more important
than the--the data has to precede the app that is on top of it.
It is not more important, but it has to precede--quality data
has to precede the app, sit on top of it. Quality data has to
precede AI. It is just logically inconsistent otherwise.
And so I vetted this view with all of my colleagues in
Silicon Valley. I had many conversations, not just with the big
companies but also with lots of startups, to make sure that
they were onboard and that they would be willing to engage as
we built out this multi-vendor, heterogeneous data-integration
layer that would allow for separation of data from apps.
Mr. Khanna. Thank you. Are there things we can do to
improve the collaboration, or do you think it is in a pretty
good place?
Dr. Martell. Well, I think we still have some work to
tackle. I think DIU, the Defense Innovation Unit, is doing a
really good job at tackling those. Doug Beck is a really strong
hire, and so I am really glad he is onboard.
You know, I have seen an increase in the way we have been
interacting with small businesses, for example. Because of the
marketplace we have built through Tradewinds, we are now up to
65 nontraditional company contracts, we are up to 75 small
business contracts. And many of these are done in 30 to 60
days. And that is because of the authorities that you all have
given us.
So we are seeing some growth, we are seeing some inertia
increase, but we need that to continue. I mean, I think the
biggest takeaway--and, actually, if I am candid, the thing that
I feel best about is, at the end of my tenure, people are
demanding more of what we are trying to do, and faster.
I 100-percent agree, faster has to happen. 100-percent
agree. But I think it is now clear what has to happen faster,
which is: getting the data right, having the data flow
correctly, being able to integrate with companies as quickly as
possible. And it shouldn't be something that we build; it has
to be multi-vendor, heterogeneous.
Mr. Khanna. Dr. Martell, I also see that you have a
computer science Ph.D. degree from the University of
Pennsylvania.
Now, many people rightfully have been concerned about anti-
Semitism, and there is no place for that on college campuses,
but some of the rhetoric in this building has gotten pretty
excessive. One Member of Congress said, I want to start
defunding these universities, and the ``rot'' in higher
education.
Could you talk about what it would mean for our national
security and our ability to have a lead in AI if we just
started defunding MIT [Massachusetts Institute of Technology],
University of Pennsylvania, Harvard, and many of these
universities?
Dr. Martell. Thank you for that land-mine question, Ranking
Member Khanna.
I think it is a tough call. The environment of a
university, in order to be effective, has to allow for everyone
to be able to think freely.
But I do agree that if we don't continue to fund STEM
[science, technology, engineering, and mathematics] the way we
need to fund STEM and continue to fund the technology in the
way that we need to fund the technology at not just the top
universities but at all universities, then that is going to put
us behind.
So continuing to fund has to be the case.
Mr. Khanna. I yield back, Mr. Chairman.
Mr. Gallagher. Mr. Gaetz?
Mr. Gaetz. General Skinner, are you a gamer?
General Skinner. No, sir, I am not.
Mr. Gaetz. No ``Madden''? No, like, EA----
General Skinner. My son i ----
Mr. Gaetz. --Sports, ``Call of Duty''?
General Skinner. My son is, but I am not.
Mr. Gaetz. I am not either. But it seems as though a lot of
our servicemembers are. I mean, I have seen, a lot of these
USOs [United Service Organization], they are building all these
great gaming complexes.
And I took interest in a report that we were feeding some
of the ``StarCraft II'' game models into an integration with
ChatGPT, and this piece, ``AI Surpasses Humans--U.S. Military
ChatGPT Outperforms in War Scenario Planning.''
And I guess what they did in this case was, they told
ChatGPT to function as an assistant to a military commander in
this engagement. So I imagine, like, an AI Dwight Schrute,
assistant to the regional manager. But it turns out, in
changing conflict dynamics, the military assistant turned out
to be quite capable.
So, Dr. Martell, you seem to have some familiarity with
this circumstance. What can we learn from the integration of
gaming models, AI, and military strategy?
Dr. Martell. So I will talk less about gaming, because I am
not a gamer, but about how we might effectively use tools,
generative AI like ChatGPT.
We have been working really hard to figure out where and
when generative AI is going to be useful and where and when it
is going to be dangerous.
The danger is, it is extremely difficult--it takes a very
high cognitive load to validate the output of this model. And
so there is a very large demand signal for AI to replace
experts and allow novices to replace experts. That is where I
think it is dangerous.
Where I think it is going to be most effective is helping
experts be better experts or helping someone who knows their
job well be better at the job that they know well.
Mr. Gaetz. I don't know, Dr. Martell. We are all on the
front end of this wave, but I find a lot of novices showing
capability as experts when they are able to access these large
language models----
Dr. Martell. So, if I can, sir, I think the reason is, it
is extremely difficult to validate the output, right?
So, as long as there is a way--I am totally onboard as long
as there is a way to easily check the output of the model.
Because hallucination hasn't gone away yet. There is lots of
hope that hallucination will go away. There is some research
that says it won't ever go away. That is an empirical open
question that I think we need to really continue to pay
attention to.
But, most importantly, if it is a difficult-to-validate
output, then it is going to be--then I am very uncomfortable
with it.
Mr. Gaetz. Yeah. I mean, I have even used, just in my
modest way, like, Claude to audit the hallucinations of ChatGPT
and vice versa.
And I can see the young people behind you nodding in the
affirmative. So sometimes that can be a check.
And I think you are right on the outputs, I think you are
dead there. But I think it is--dead-right--I think it is
interesting, though, to think about this in terms of--
Dr. Martell. Absolutely.
Mr. Gaetz. --an assistant. And then you think about that in
the air domain, in the space domain.
But I want to get to the inputs as well. Because you made a
really good point in your testimony about the quality of the
data dictating, kind of, the ceiling on this enterprise.
Dr. Martell. Right.
Mr. Gaetz. And I envision a circumstance where we are in
this room marking up the NDAA [National Defense Authorization
Act] and there is a big fight among all the lobbyists for the
defense contractors about who owns the data. Right? Just as we
see in the large language models, The New York Times and these
entities saying, ``Well, you trained on our data and our stuff,
and so we have some ownership interest in the work product that
comes out.''
As you depart government service, on to something else,
what advice can you give the subcommittee about how to have as
much of that data open-source and acceptable so we don't have a
circumstance where, like, Lockheed Martin is saying, ``Well, we
have to protect our source code----
Dr. Martell. That is right.
Mr. Gaetz. --on the F-35'' and back to that sort of stuff?
Dr. Martell. So I think the right tactic there is to
separate stovepipe solutions--which really are data all the way
up to the end user--into two layers, a data layer and an app
layer, an application layer, and then create two separate
marketplaces.
So the app marketplace makes perfect sense. You guys get
it. Who is going to open my Word doc? Is it going to be Google
or is it going to be Microsoft or some third party? That is the
app layer. But then the actual word-processing doc is the data
layer.
But you can also build a marketplace in the data layer. If
Lockheed Martin had invested a great deal of IP [internet
protocol] and work to building out that data layer, well, maybe
we pay them for access to that data and figure out interesting
contracting ways where they can actually make money selling
that data not just to themselves but to other app builders, but
require that data be accessible.
So I really think----
Mr. Gaetz. That is a really important point. You know, I
hope we are able to get back to it at some point. Because what
I worry about is, they will create the cost of that data as so
cost-prohibitive so as to vertically integrate all the features
of the contract.
And Mr. Sherman, who testified about contract sprawl
earlier, I think, is acknowledging that we are going to have to
confront that. And I think----
General Skinner. Yes, sir.
Mr. Gaetz. --it is a sticky wicket.
General Skinner. I think that is right.
Mr. Gaetz. Thank you, Mr. Chairman. I yield back.
Mr. Gallagher. Mr. Moulton.
Mr. Moulton. Thank you, Mr. Chairman.
Mr. Sherman, I co-led the legislation to establish 988, the
new National Mental Health Hotline. And in the first year and a
half, calls are up over 50 percent, texts are up over a
thousand percent, literally saving thousands of lives.
The Commandant called me last year to share what he has
heard from Marines about what a difference it has made in their
lives and their units.
I know the leadership of the Department is working on
reducing our shocking number of suicides, and one
straightforward way to help is to simply make sure that 988 is
dialable from all DOD phones. It is a legal requirement that
you can dial 911, but not yet 988.
So what can we do to make sure we get there?
Mr. Sherman. Congressman, first, thank you for sponsoring
that legislation. As Secretary Austin has said, mental health
is health, and this is imperative.
So, within the Department of Defense, we have validated its
use in the Pentagon, Fort Myer. We even called it from my
office to make sure, because, I am going to be honest, the
first few times we were doing this, to make sure you don't have
to put a 9 in front of it or something like that, to make sure
it could work.
So, working with General Skinner, with his DISA hat on, and
also Personnel and Readiness, making sure that that guidance is
out there.
We are going to continue to validate. And I want to make
sure I am clear on this. This is going to be an ongoing process
for whether it is at Fort Cavazos or Camp Pendleton or Edwards
Air Force Base, wherever a servicemember or civilian is that
needs to be able to get to that number, that we are going to
have to continually validate this.
So this is something we have been talking about just
recently here, that this is going to be an ongoing process to
make sure that 988 is reachable without having to put 1-1 in
front of it or 9 or things that I would have had to do at Fort
Stewart in calling from the day room down at 15-ADA (ph).
So, sir, this is something that has got our attention for
large and small installations alike to make sure we can
validate that.
Mr. Moulton. Once the troops know this number by heart--and
we have done a lot to make sure it is publicized on bases--it
is going to be important to have it accessible from places like
Okinawa and Landstuhl as well. So I hope you will keep your
horizons far.
Mr. Sherman. Yes, sir.
Mr. Moulton. Dr. Martell, thank you very much for your time
with the Department.
We all clearly agree that we need to invest in AI
capabilities for the future. And we also understand, as you
have said, that AI is dependent on data.
It is kind of insane to me that American taxpayers have
spent more money on the J-35 than any defense system in
history, and yet we download its data from many missions and
literally throw it out because we don't have a place to store
it.
When is this going to stop?
Dr. Martell. I am getting all the great questions, aren't
I?
That is a great question, and I appreciate it. We call that
data ``exhaust,'' where valuable data is being blown out the
tailpipe, and there is an example of it.
I don't feel that I have the expertise to answer that
question, when it is going to stop. But I can say that we have
to continue the demand for quality data to deliver solutions.
And I believe that virtuous cycle is going to then start
creating the demand to understand how that data can be most
effective.
Mr. Moulton. Well, I would just point out that, if my
laptop or my phone told me that I was out of space, I could
solve that problem in a couple hours. So I don't expect DOD to
move that fast, but the fact that this has taken literally
years and you still can't give us an answer I think is a real
problem.
Dr. Martell. I will have to take the rest of that as a
question for the record and get back to you. I am not
sufficiently familiar with----
Mr. Moulton. Okay. Well, it would be very helpful to have a
timeline.
[The information referred to can be found in the Appendix
on page 55.]
Dr. Martell. Okay.
Mr. Moulton. And then, of course, once we simply retain
that data--you have addressed this in some of the other
questions members have asked--how are we going to make it
safely available not just to, you know, the primes, right ----
Dr. Martell. That is right.
Mr. Moulton. --but to some of the leading private-sector
companies in AI innovation?
Dr. Martell. So I think there are two parts to that: the
safely and the available.
From the available perspective, the technology is there. It
is the right APIs that have the right limitations of
accessibility on them. But it is more about the contracting, to
make sure that the folks who have produced the data--I mean,
the data is fundamentally the government's data, but then a lot
of work goes into massaging that data to make it accessible.
And that is IP that should be paid for, in my opinion. But we
have to make the contracts clear about what that payment looks
like and how it must maintain its accessibility. That is a
really big set of work that we have to do.
On the safely accessible part, I defer to my colleagues on
how we can use zero trust to make sure that only the right
people are accessing that.
Mr. Moulton. Well, look, I would just point out that,
because of civil-military fusion in China, China is doing this
today, right? I mean, they have their top AI companies working
on their military problems. So this is another thing where we
have to have the urgency to get this done.
A quick question on workforce recruitment, Mr. Sherman: How
are we going to find someone to replace Dr. Martell who brings
the level of academic expertise and private-sector experience
to the job?
Mr. Sherman. Well, we are certainly going to miss Dr.
Martell. He has been a huge combat multiplier.
I think Dr. Radha Plumb, his designated replacement, is
pretty darn amazing too. And I think she is going to pick up
the baton where he is leaving it there, sir.
Mr. Moulton. All right.
Thank you, Mr. Chairman.
Mr. Gallagher. Mr. Luttrell.
Mr. Luttrell. I am glad that you are nodding your head
``yes,'' Dr. Martell. That puts me kind of in my comfort zone,
that somebody is coming in behind you with some expertise to
keep widening the road that you have paved.
I picked up on the cross--I call it cross-functional teams,
where you have your engineers sitting with your operators, and
that dynamic back and forth just lends itself to forward-
leaning success. So I am sure the expansiveness of that is
growing, yeah?
You lead the data integration layer for CJADC2, correct?
Dr. Martell. Correct.
Mr. Luttrell. This is likely, in my opinion, the most
critical position in that Department. Given it is centrality in
moving and exchanging data between and within each services,
this likely requires not just coordination but direction to the
various stakeholders.
Can you explain what directive authority you have been
afforded in building the data integration layer?
Dr. Martell. That is a great question, in terms of the
authorities. I can tell you what we have been doing, and then I
can take as a question for the record what authorities we have
been doing them under.
Mr. Luttrell. Okay.
Dr. Martell. But, right now, we have been building out the
prototype of what it would mean for the hardware to support the
flow of data across combatant commands, for example, so
combatant commands can have a unified picture of what is going
on in the world.
We are doing that as the key learning exercise--and we do
it every 90 days through these guide exercises--as the key
learning exercise to understand what combatant--and through
war-gaming--what combatant commanders would need to see and
what all of the components under the combatant commanders would
need to see, would need to exchange, and how data would need to
flow in order for it to go from swivel-chair and PowerPoint and
email to digital dataflows as that information goes across the
combatant commands and within the combatant commands.
So the next step, which we should be doing within the next
3 to 6 months, is building out a set of requirements so that
industry can help--I mean, we have been doing this with key
industrial partners, mostly Palantir and Anduril, but it needs
to be--it needs--and the technology that we have built we are
leaving behind. It is there; it is available. That is why we
call it a minimum viable capability. It is viable. We then need
to build out clear requirements that allow other industrial
partners to join in and to expand that data integration layer
and then also to expand those capabilities.
Mr. Luttrell. That sounds like something this--we might
need to tweak this a little bit as you are starting to
transition something that you wish you would have had, that the
incoming can have. That is something you need to address with
this committee, I would say.
Dr. Martell. Absolutely. And I would prefer to take this as
a question for the record----
Mr. Luttrell. Yes, sir. Absolutely. I understand.
Dr. Martell. --and get you an absolutely clean----
Mr. Luttrell. I understand.
Dr. Martell. --clean definition of what we need.
[The information referred to can be found in the Appendix
on page 55.]
Mr. Luttrell. Mr. Chairman, are we doing two rounds or just
one?
Mr. Gallagher. I was planning on one, but----
Mr. Luttrell. Okay.
Mr. Gallagher. --do you need two?
Mr. Luttrell. I could use about four or five. All right.
Dr. Martell, can you explain what would be the consequences
of affording your position with some sort of directive
authority over the services and their development of CJADC2
solutions?
Dr. Martell. I am not a fan of that. And I am not a fan of
a hard authority there. And I may be saying the wrong words
because I haven't quite got up, in the 2 years I have been
here, in all the bureaucratic lingo I should.
I think the services understand--in general, let me just
say, in general, the center should provide oversight and policy
and best practices. The edge knows their problems best.
So solving the problems from the center and imposing it
upon the edge I think is dangerous. It is going to creates one-
size-fits-all solutions that don't----
Mr. Luttrell. You should write that down and put it on a
plaque so we can hang it in every room in this building.
Dr. Martell. I will take that as an activity for the
record, sir.
Mr. Luttrell. Okay.
Dr. Martell. That was a joke. We are not really going to do
that, guys.
And the services understand what they need better than OSD
[Office of the Secretary of Defense] is going to understand
what they need. But there does have to be authority about the
interface. So data is going to have to go down into the
services, and the right kind of data is going to have to flow
up out of the services.
So where we have been spending our energy with the
services--say, project convergence or with overmatch--is
figuring out how would that data flow. We still have the policy
questions or the workflow questions about what data should
flow.
So I actually see--what we have been doing, to be clear, is
putting the tech before the policy. The tech to allow the data
to flow is going to force the right questions: What data should
flow? Right?
Like, we can't answer Title 10 and Title 50 combinations a
priori. We can't just start with that. We have to watch the
data flow, and then there is going to be an increased demand
for data to flow, and the right data to flow. And then we can
say, ``Well, now, that is a policy issue that we can tackle,''
or, ``That is a small version of the policy issue for just that
one piece of data.''
That is the way that the change is going to happen. The
change is not going--my very strong opinion--the change is not
going to happen by some large, a priori view, philosophical
view, of the way the world should be and then trying to
implement that.
Mr. Luttrell. Okay. Thank you, sir.
Mr. Gallagher. Dr. McCormick.
Dr. McCormick. Thank you, Mr. Chair.
I find it fascinating, we just had our first AI task force
meeting just two floors away from us, literally right in line
with us. And there is so much focus on this technology right
now, it is just unbelievable, from every aspect of, whether it
be cybersecurity, defense, mechanical applications, labor
force, education, law and what do we consider legal and how do
you verify facts, and then of course our security.
This is such a comprehensive issue right now that everybody
is focused on right now. I am just going to kind of dive in on
a more broad scope so you can give me some more insight into
how you are approaching it.
Dr. Martell, much like the computer power and data storage
capacity, artificial intelligence advances at an increased rate
right now--obviously, exponential rate.
What challenges does this rate of acceleration--in other
words, as we are trying to keep up legislatively and
regulatory-wise, what does this acceleration create for
integrating data, artificial intelligence, and digital
solutions for the DOD and how it applies, like, just trying to
keep up with this pace?
Dr. Martell. So it is remarkable. And I will say, in my 30-
year career, this is the most exciting time to be an AI expert.
The science has been really remarkable even just the last 5
years, but if you go back 15 years, it has been amazing as
well.
I don't know if the science supports the marketing, though.
So I think one is to be--``skeptical'' is the wrong word. I am
very bullish on this technology. But I think it is important to
be skeptical about--or to verify--``trust, but verify,'' if I
can quote someone, the claims of the marketing. So that is one.
Number two, I think it is extremely important to not see
this as a monolithic technology, which is how it is sold: Get
this thing; if we have it, we win, and if they have it, we
lose. And I think that is fundamentally flawed. It is neither a
panacea, nor is it a Pandora's box.
So if we have it, it doesn't mean every problem we want
solved will be solved. And if they have it, it doesn't mean
every fear that we have will come to pass. We always have to
think about it on a use-case-by-use-case basis, as we always
have with any technology.
So the reassurance is, it is just a technology. It will
work in some use cases; it won't work in other use cases. We
have to have quality data and evaluation metrics to be able to
say it is working now or it is not working now.
So I get extremely frustrated with my colleagues in
industry who pitch it as magic. I actually think that is the
case often. But then when you ask them about the use cases,
they have a clear set of small use cases that they know how to
measure and are effective: information retrieval via RAG, or
first-draft generation, or a number of things which we all feel
fairly confident are working well.
But the, sort of, pitch that it is going to reason for us,
it is going to solve our problems, we are going to be able to
do away with brains and use this instead, that is just way-
overhyped.
Now, last year, I think that hype was greater. I think we
have all learned in the last year where that hype is starting
to come down, and that is making me very happy.
Dr. McCormick. So you mentioned a couple things, claims of
marketing and also maybe--I would dispute one thing that you
said, because we talk about them having it, us not having it.
When a country literally talks about taking over an island that
produces basically 100 percent of our AI chips and we don't
have a backup, that is kind of a denial of one from the other.
And it is different than what you were discussing, I
understand----
Dr. Martell. No, and I am 100 percent--look, the underlying
hardware--for example, the CHIPS [Creating Helpful Incentives
to Produce Semiconductors] Act is the right direction to go,
because we need to be--that, I think, is a national-security
issue, the ability to generate the underlying infrastructure
and the underlying data that is going to allow this to be
successful. I just meant the models themselves, we have a gap
between the science and the marketing.
And one of the things our organization has been doing in
Task Force Lima is trying to rationalize that gap. We are
building what we are calling a maturity model, very similar to
the autonomous driving maturity model, so, you know, Level 1
through Level 5 autonomous driving. That is a really useful
model, because people have claimed Level 5 but, objectively
speaking, were really at Level 3, with a couple folks doing
some Level 4 stuff.
Dr. McCormick. As an ER [emergency room] doc, I am all for
kids not driving in cars, figuring out how to do it safely. But
I will say, from a military aspect and my military experience,
I am very concerned about the kill chain and what it could mean
to us when we have disruptive Chinese technologies that are
literally competing with us at this cybersecurity level. That
really worries me the most.
I don't know if----
Dr. Martell. Sir, I would like to--if I could respond?
Dr. McCormick. Sure.
Dr. Martell. I am 100 percent in agreement. So I am not
saying we don't have to worry about specific use cases. We do
have to worry about specific use cases. We just shouldn't worry
that it is monolithic.
So, in particular use cases, we should do what we always
do: evaluate what our adversaries can do, evaluate what we can
do, invest in ways that we can do it better, and invest in ways
in which we can maybe contribute to our adversaries not doing
it so well.
My only caution was, don't think about it as a scary
monolithic technology. Think about particular use cases, and
then do the right evaluation of where they are and where we are
and invest accordingly.
Mr. Gallagher. Mr. Fallon.
Mr. Fallon. Thank you, Mr. Chairman.
I knew it was going to be a good day when I walked in and I
saw so many Air Force folks here, being a former Air Force
member.
Dr. Martell, what concerns me about a conflict with a near
peer--and, obviously, talking about China, Taiwan--and a
concern really of everyone on the committee and everybody and
all the witnesses is the tremendous potential for loss of life.
Another thing that really bothers me is, when you are
looking at war games, where there are some scenarios where we
lose in just a very short timeframe a thousand fighter
aircraft. That is devastating. And to avoid that kind of a
catastrophic loss, I think we need to employ more unmanned
aircraft, most notably, of course, autonomous platforms----
Dr. Martell. Agreed.
Mr. Fallon. --for not only strike purposes, for sustainment
as well.
Dr. Martell. Uh-huh.
Mr. Fallon. To do this, the DOD needs to not only invest in
hardware but, of course, software--probably even more
importantly, the software. So that exists in the commercial
space, but companies like Amazon don't have the requirement to
operate in GPS-denied environments like the military.
So the question I want to ask you is, what do you think the
significance of autonomous unmanned air platforms will have on
helping us win a future near-peer fight? And are we doing
enough to get where we need to be?
Dr. Martell. I completely agree with you; I think they are
going to be extremely important. And I think we have to
continue to drive that forward.
One of the things the CDAO is doing is supporting the
Replicator initiative by creating a sandbox, if you will, where
companies can come, using our data, to evaluate the success of
their unmanned aircraft software with respect to the scenarios
that we think are going to be effective.
So we are strongly supporting this. Part of what we call
``AI scaffolding'' that we are building out, part of this way
in which we are trying to help scale the Department by giving
tools to folks closer to the edge is this Replicator, this
Replicator sandbox.
And so we will provide data to that, we will provide
scenarios. I am pretty sure--I will double-check--that we are
going to provide some software for that. And then the folks who
are part of the Replicator initiative will come and try out
their software, their prediction, their detection on our data.
So we think it is really important, and we are strongly
supporting it.
Mr. Fallon. Could you talk a little bit more about that--
you mentioned the AI scaffolding?
Dr. Martell. So, you know, as I mentioned in my opening
remarks, we really see our job as twofold. One is immediate
help to the warfighter; and building out tools, processes,
policies, best practices that will allow other folks, the ones
closer to the problem, to easily build the solution to the
problem. Because a centralized team, it cannot build
everything, right? As we talked about before, that creates a
one-size-fits-all solution.
Part of our support to be able to scale is what we are
calling ``AI scaffolding.'' So AI scaffolding includes things
like the ability to do data labeling as a service. Because
knowing what you want to detect means humans have to label that
data. That is very difficult. Most teams closer to the edge
won't know how to do that, so we want to provide contracting
and expertise to allow them to do that.
Data transformation is a service. The data that has to come
into your problem is going to be in a thousand different
formats. The amount of work to transform a PDF, a Word doc, et
cetera, et cetera, et cetera, into structured data that you
need is pretty massive, and that will take up, for the folks on
the edge, all of their time. We want to be able to provide
services like that.
The building of the model, we strongly believe that should
be--because our IP is the data. The government's IP is the
data. We strongly believe the building of the model should be
contracted to industry. But on the other side--and we will
provide those tools, those contacts, those ability to build,
you know, initial models yourself. That is part of our AI
scaffolding. The final model should probably be built by
absolute experts in this field.
And, then, on the right side of the model building, we
think very hard about model monitoring. So one of, I think, the
biggest issues that industry just started tackling I would say
5 years ago, 5 to 10 years ago, and government is behind on is,
after we ship a model, does it continue to bring value?
Well, remember, AI is statistics at scale. We gather data
from the past to predict the future. In a warfighting scenario,
the future changes; it doesn't look like the past. So a model
trained on a pre-warfighting scenario is going to start to
degrade as the world changes in a fight. How do we measure that
degradation? How do we retrain to get that model back up to
where we need it to be?
So that is the model monitoring piece. And I wish we had
gotten further on that in the 2 years I have been here, but
that is going to be my biggest charge for Dr. Plumb, is really
focus on how do we model and measure the value of models over
time.
Right now--and it was this way in industry 5 years ago--
they are shipped and we forget. They are shipped and we
continue to believe. We don't forget. We just continue to
believe.
So our contracting, for example, should include model
monitoring. The way we deliver the model should include the
ability to monitor it.
And so we have done a lot of thinking about this. We need
to move out faster on actually executing on it.
Mr. Fallon. Okay.
Thank you, Mr. Chairman. I yield back.
Mr. Gallagher. Great.
We will now go into the classified session, which will be
upstairs in 2337 Rayburn. I know there were a variety of
questions that were not answered because of classification, so
we look forward to getting into those. And we will see you up
there in a few minutes.
With that, the unclassified portion is adjourned.
[Whereupon, at 9:57 a.m., the subcommittee proceeded in
closed session.]
=======================================================================
A P P E N D I X
March 22, 2024
=======================================================================
=======================================================================
PREPARED STATEMENTS SUBMITTED FOR THE RECORD
March 22, 2024
=======================================================================
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
=======================================================================
WITNESS RESPONSES TO QUESTIONS ASKED DURING
THE HEARING
March 22, 2024
=======================================================================
RESPONSES TO QUESTIONS SUBMITTED BY MR. MOULTON
Dr. Martell. I defer to the CDO of the F-35 program, USAF. [See
page 14.]
______
RESPONSES TO QUESTIONS SUBMITTED BY MR. LUTTRELL
Dr. Martell. The authority to build the Data Integration Layer
derives from the Deputy Secretary of Defense memo titled ``Initial
Operating Capability of the Chief Digital and Artificial Intelligence
Officer'', dated February 1, 2022. The memo directs the CDAO to
``create enabling digital infrastructure and services that support
Components' development and deployment of data, analytics, AI, and
digital-enabled solutions'', among other functions. [See page 16.]
=======================================================================
QUESTIONS SUBMITTED BY MEMBERS POST HEARING
March 22, 2024
=======================================================================
QUESTIONS SUBMITTED BY MR. MOULTON
Mr. Moulton. A significant portion of data on F-35s
collected/generated during missions is discarded because there
is no place to store it. When and how will DOD stop this data
exhaust from F-35s?
Dr. Martell. I defer to the CDO of the F-35 program, USAF.
------
QUESTIONS SUBMITTED BY MR. STRONG
Mr. Strong. Understandably, the cost of designing,
training, and deploying AI is top-of-mind for any organization.
Has the Department considered how it can leverage its existing
technological capabilities -including on-premise, private
cloud, public cloud, and hybrid cloud computing capabilities to
design, train, and deploy AI in a manner that ensures
compatibility with existing investments?
Mr. Sherman. Thank you, Mr. Strong, my team, and I
appreciate the offer to speak with your staff to connect to
Huntsville-area technology developers. To answer on the
efforts, the Department has long recognized the current and
future AI demand for computing power and capacity. There are
plans to develop Capability Planning Guidance--in concert with
CDAO input--about the necessary funding for AI designing,
training, and deploying. The adoption of cloud and our various
architectural models for cloud have and continue to remain
fundamental in modernizing our IT infrastructure for an AI
future.
We are relying on our commercial cloud service providers to
deliver the needed capability, innovation, and scale to enable
an AI-powered warfighter. Improving our visibility of cloud use
and cost through our JWCC contract and contract rationalization
efforts will help us understand critical trends and areas of
mission impact for more informed investment decisions across
the technology stack (both commercial and government). Given
our fiscal constraints, we must be measured in our investment
decisions in conjunction with CDAO, understanding that AI
initiatives today are still primarily narrow but one day, will
shift to that next level of intensity. We are preparing to be
ready with flexible processes that allow us to securely,
economically, and timely meet that need.
Mr. Strong. Mr. Sherman, one of your deputies, David
McKeown (Deputy Chief Information Officer for Cybersecurity)
spoke at AFCEA's TechNet Emergence conference held earlier this
month and said that no one in industry has come to him with a
cyber protection program using AI . . McKeown told the
contractor-heavy audience at AFCEA, ``if you happen to know of
any products out there that are leveraging AI to do
cybersecurity things, I'd love to talk to you. This is a key
part of our zero-trust strategy.''
The Armed Services Committee has done considerable work in
recent NDAA's to push for cyber technologies using AI, and I
know of several technology companies in my hometown of
Huntsville that are in fact developing cyber defense
technologies using AI. For instance, to address the Space and
Missile Defense Command (SMDC)'s Space & Ground Systems
operational cyber threat detection and prevention requirements
and support the zero-trust strategy, technology is being
developed to auto-configure, self-optimize, and self-train
defensive models to detect and mitigate cyber threats against
critical ground systems and space assets. This technology
leverages advanced Artificial Intelligence (AI), Machine
Learning (ML), and Reinforcement Learning (RL) technologies to
provide a holistic approach to space and ground systems cyber
protection by covering the entire threat domain.
Any comments on the future of technologies like this from
you would be appreciated. Further, my staff would like to
discuss this with your office further to help connect the
technology developers with the DOD end user.
Mr. Sherman. Our Team members routinely meet with
technology companies to identify areas in which they may bring
new or amplify existing capabilities to the Department. We
welcome the opportunity to learn more about how these companies
are leveraging AI, ML and RL in their approaches. Note that the
team is leveraging AI capabilities to support department's
Cyber Supply Chain and Software Supply Chain Assurance
activities. As you can imagine, the big data analytics required
to conduct supply chain risk analysis of our weapons and
information systems is significant. The Department has engaged
leading supply chain illumination companies, such as Exiger,
Ion Channel, and Interos, to promote and deliver supporting AI
technologies to the Department, down to program level, as well
as to our industry partners. In a recent example, one of these
illumination companies, Exiger, is leveraging Karambit.ai to
help evaluate the source code of --and at the request of-- a
market-leading U.S. company that provides critical software to
the Department. This company has developed code in a country of
concern for more than a decade; therefore, it has high risk
profile.
Examples of employing automated supply chain illumination
tools and AI provide transparency into the information and
communications technology supply chain. These tools can also
help industry regain assurance of their technology, ensuring
provenance and integrity of their code, in accordance with
Executive Order 14028, Executive Order on Improving the
Nation's Cybersecurity. Another area of interest is the
utilization of AI and ML capabilities is the area of continuous
monitoring and orchestration. AI/ML capabilities are well
suited for the DoD's continuous monitoring and orchestration
needs due to their abilities to comb through petabytes of data
at tremendous speed to inform risk-based decision-making with
accurate, real-time, threat-based intelligence. We welcome
industry's partnership in leveraging AI for cybersecurity.
We're hoping industry will step forward to demonstrate
capabilities that the Department can learn from and adopt.
Mr. Strong. Understandably, the cost of designing,
training, and deploying AI is top-of-mind for any organization.
Has the Department considered how it can leverage its existing
technological capabilities -including on-premise, private
cloud, public cloud, and hybrid cloud computing capabilities to
design, train, and deploy AI in a manner that ensures
compatibility with existing investments?
Dr. Martell. I defer to Mr. Sherman and Lt Gen Skinner to
best respond to this question.
Mr. Strong. Understandably, the cost of designing,
training, and deploying AI is top-of-mind for any organization.
Has the Department considered how it can leverage its existing
technological capabilities -including on-premise, private
cloud, public cloud, and hybrid cloud computing capabilities to
design, train, and deploy AI in a manner that ensures
compatibility with existing investments?
General Skinner. Yes, this is something that DoD is always
considering. It is a complex challenge involving technology
adoption and a need for compute capabilities available across a
spectrum of use cases, as there is not one-size-fits-all.
Current DoD policy and guidance encourage the use of agile
software development via numerous mechanisms, including DoD
Instruction (DoDI) 5000.87, ``Operation of the Software
Acquisition Pathway;'' the 2022 DoD Software Modernization
Strategy; the 2021 DoD Enterprise DevSecOps Strategy Guide; as
well as other issuances from both the Office of the Secretary
of Defense and the individual Services.
DoD should consider that any program with software
capabilities be developed via a Modular Open Systems Approach
(MOSA), which a few programs have already adopted. This
approach allows defining a broad set of end goals that allow
flexibility for industry partners to develop the most creative
and advanced capabilities to meet those goals rather than
develop capabilities that must meet predetermined DoD-mandated
technical specifications that may be outdated by the time the
capability is fielded. At a minimum, requirements documents
should ensure that data from systems being developed will be
collected, stored, and shared in a manner consistent with
Department data strategies, policies, and guidance. This would
enable greater interoperability and more iterative and agile
development practices.
With respect to compute in general, the DoD is considering
how it can best optimize the compute currently available to the
maximum extent practicable across the mix of use cases that
allow for the mix of on-premise, private cloud, public cloud,
and hybrid cloud computing capabilities. The generative AI use
case in particular is driving the need for increased compute
availability. The compute available today is insufficient to
satiate demand, nor agile enough with scale to respond to
what's easily knowable and foreseeable on the horizon. The
rubric being addressed is to define what the compute needs look
like in 5 years, and how the investments should be structured
to achieve the ability to meet demand in the changing and
evolving landscape.
In summary, your question is the right one, and we are
working thru thinking about this exact problem as this is a
foundational aspect of AI adoption that must be frequently
revisited to ensure a good match between capability and
infrastructure.
------
QUESTIONS SUBMITTED BY DR. MCCORMICK
Dr. McCormick. Mr. Sherman, the Department of Defense (DOD)
uses websites to communicate information and effectively engage
the public. It operates more than a thousand public websites
for the Military Departments, Joint Staff, Combatant Commands,
Department of Defense Agencies and DOD Field Activities. In
June 2020, the DOD tasked the Defense Media Activity (DMA) with
consolidating all DOD public-facing websites, but little
progress has been made toward that goal. Congress has supported
DMA through several budget cycles, most recently including a $5
million increase for DMA in the FY23 Appropriations Act.
However, in the recently released FY25 budget request, the DOD
has, again, not requested funding for this program, despite it
being highlighted in the DMA's strategic priorities list and
Congress' continued support.
Is Defense Media Activity's (DMA) public web program still
a strategic priority for the Department of Defense (DOD)? If
yes, why has DOD not requested funds for the program in its
budget requests and should we expect future year's requests to
include the proper resourcing for this critical initiative?
Mr. Sherman. Yes, consistent with the Annual Report on the
Progress to Implement the 21st Century Integrated Digital
Experience Act (IDEA), consolidating all DoD public facing
websites is still a priority. This report underlines the DoD's
process in achieving these goals. However, specific questions
on funding requests would be better directed to the director of
DMA consistent with DoD Directive 5105.74. https://
dodcio.defense.gov/Portals/0/Documents/21stCenturyIDEAReport-
2023.pdf
Dr. McCormick. Mr. Sherman, what is Defense Media
Activity's (DMA) status of consolidating all Department of
Defense public-facing websites, meeting security requirements,
and complying with the 21st Century IDEA Act?
Mr. Sherman. Per DoD CIO's Annual Report on the Progress to
Implement the 21st Century Integrated Digital Experience Act,
DMA is consolidating DoD public facing websites using the Web
Enterprise Business hosting solution. This system meets all DoD
standards for information security and continues to make
progress in implementing the 21st Century IDEA. For more
information, please refer to the more comprehensive Annual
Report on the Progress to Implement the 21st Century Integrated
Digital Experience Act. https://dodcio.defense.gov/Portals/0/
Documents/21stCenturyIDEAReport-2023.pdf
[all]