[House Hearing, 115 Congress]
[From the U.S. Government Publishing Office]
.
[H.A.S.C. No. 115-122]
DEPARTMENT OF DEFENSE'S
ARTIFICIAL INTELLIGENCE STRUCTURE, INVESTMENTS, AND APPLICATIONS
__________
HEARING
BEFORE THE
SUBCOMMITTEE ON EMERGING THREATS AND CAPABILITIES
OF THE
COMMITTEE ON ARMED SERVICES
HOUSE OF REPRESENTATIVES
ONE HUNDRED FIFTEENTH CONGRESS
SECOND SESSION
__________
HEARING HELD
DECEMBER 11, 2018
[GRAPHIC NOT AVAILABLE IN TIFF FORMAT]
__________
U.S. GOVERNMENT PUBLISHING OFFICE
34-978 WASHINGTON : 2019
--------------------------------------------------------------------------------------
SUBCOMMITTEE ON EMERGING THREATS AND CAPABILITIES
ELISE M. STEFANIK, New York, Chairwoman
BILL SHUSTER, Pennsylvania JAMES R. LANGEVIN, Rhode Island
RALPH LEE ABRAHAM, Louisiana RICK LARSEN, Washington
LIZ CHENEY, Wyoming, Vice Chair JIM COOPER, Tennessee
JOE WILSON, South Carolina JACKIE SPEIER, California
FRANK A. LoBIONDO, New Jersey MARC A. VEASEY, Texas
DOUG LAMBORN, Colorado TULSI GABBARD, Hawaii
AUSTIN SCOTT, Georgia BETO O'ROURKE, Texas
JODY B. HICE, Georgia STEPHANIE N. MURPHY, Florida
(Vacancy)
Eric Snelgrove, Professional Staff Member
Lindsay Kavanaugh, Professional Staff Member
Jamie Jackson, Deputy General Counsel
Neve Schadler, Clerk
C O N T E N T S
----------
Page
STATEMENTS PRESENTED BY MEMBERS OF CONGRESS
Langevin, Hon. James R., a Representative from Rhode Island,
Ranking Member, Subcommittee on Emerging Threats and
Capabilities................................................... 2
Stefanik, Hon. Elise M., a Representative from New York,
Chairwoman, Subcommittee on Emerging Threats and Capabilities.. 1
WITNESSES
Deasy, Dana, Chief Information Officer, Department of Defense.... 5
Porter, Dr. Lisa, Deputy Under Secretary of Defense for Research
and Engineering, Department of Defense......................... 4
APPENDIX
Prepared Statements:
Deasy, Dana.................................................. 32
Porter, Dr. Lisa............................................. 27
Stefanik, Hon. Elise M....................................... 25
Documents Submitted for the Record:
[There were no Documents submitted.]
Witness Responses to Questions Asked During the Hearing:
[There were no Questions submitted during the hearing.]
Questions Submitted by Members Post Hearing:
Mr. Larsen................................................... 43
DEPARTMENT OF DEFENSE'S
ARTIFICIAL INTELLIGENCE STRUCTURE,
INVESTMENTS, AND APPLICATIONS
----------
House of Representatives,
Committee on Armed Services,
Subcommittee on Emerging Threats and Capabilities,
Washington, DC, Tuesday, December 11, 2018.
The subcommittee met, pursuant to call, at 3:30 p.m., in
room 2118, Rayburn House Office Building, Hon. Elise M.
Stefanik (chairwoman of the subcommittee) presiding.
OPENING STATEMENT OF HON. ELISE M. STEFANIK, A REPRESENTATIVE
FROM NEW YORK, CHAIRWOMAN, SUBCOMMITTEE ON EMERGING THREATS AND
CAPABILITIES
Ms. Stefanik. Thank you for your patience. The subcommittee
will now come to order.
Welcome, everyone, to this open hearing of the House Armed
Services Subcommittee on Emerging Threats and Capabilities.
Today we will examine the DOD's [Department of Defense's]
efforts to transform the delivery of artificial intelligence-
enabled [AI] capabilities to the warfighter.
AI and machine learning are topics of priority and deep
interest among the members of this subcommittee as we build a
blueprint for the battlefield of the future. Over the last
year, we have explored these technology issues closely and
heard from numerous outside subject matter experts on the
emerging opportunities, challenges, and implications of
adopting commercial artificial intelligence solutions into the
defense enterprise.
We have also closely examined our adversaries' investments
in AI and related technologies, including China's whole-of-
society approach, which threatens our competitive advantage. In
response, this committee has taken deliberate bipartisan
actions to better organize the Department of Defense to
oversee, accelerate, and integrate artificial intelligence and
machine learning technologies.
The John S. McCain National Defense Authorization Act for
Fiscal Year [FY] 2019 directed the Secretary of Defense to
conduct a comprehensive national review of advances in AI
relevant to the needs of the military services.
Section 238 further directed the Secretary to craft a
strategic plan to develop, mature, adopt, and transition
artificial intelligence technologies into operational use.
Additionally, section 1051 established the National
Security Commission on AI, an independent entity inside the
executive branch, to take a holistic view of the
competitiveness of U.S. efforts and elevate the national
conversation surrounding the national security implications of
AI.
Today, we will continue this conversation and hear about
the DOD's efforts to reorganize and more effectively oversee
the execution of AI programs across the military services. We
will also examine the Department's investments in basic
research to generate groundbreaking AI capabilities for future
conflict.
The transformation and prioritization of AI inside the
Department today will shape the efficiency of DOD's business
functions, and most importantly, the effectiveness of our
forces in future battle.
Let me welcome our witnesses here today: Dr. Lisa Porter,
Deputy Under Secretary of Defense for Research and Engineering
at the DOD, and Mr. Dana Deasy, Chief Information Officer at
the DOD.
We look forward to your testimony.
And finally, I want to take this time to recognize and
express this subcommittee's gratitude to two staff members who
will be departing the committee this month, Neve Schadler and
Mark Pepple. Thank you so much for all of your work this past
year and years prior. Your contributions to this committee are
appreciated from both sides of the aisle, and we wish you best
of luck in your next endeavors.
Let me now turn to Ranking Member Jim Langevin for his
opening comments.
[The prepared statement of Ms. Stefanik can be found in the
Appendix on page 25.]
STATEMENT OF HON. JAMES R. LANGEVIN, A REPRESENTATIVE FROM
RHODE ISLAND, RANKING MEMBER, SUBCOMMITTEE ON EMERGING THREATS
AND CAPABILITIES
Mr. Langevin. Thank you, Madam Chair, and I want to thank
and welcome our witnesses here today.
This year the Emerging Threats and Capabilities
Subcommittee has placed a significant emphasis on how
artificial intelligence, machine learning, and associated
technologies can be used to advance U.S. warfighting and
deterrence capabilities and bring efficiencies to business
processes and systems in the Department.
In June, the subcommittee held an industry roundtable where
we focused largely on the implementation of AI in the defense
innovation system and how the Department of Defense can best
leverage in-house and commercial capabilities to support
military functions.
During the roundtable discussion I expressed serious
concerns about what I perceived as a disjointed, ad hoc
approach by DOD in developing Department-wide AI policies,
strategies, and programs. Since then, I am pleased to see that
the Department has made some strides toward refining and
refocusing its AI programs and initiatives.
Most notably, the Department launched the Joint Artificial
Intelligence Center [JAIC]. I look forward to better
understanding how this center, located under the Chief
Information Officer, will bring synergy to Department-wide
efforts.
More specifically, I hope to hear today about the center's
structure, mission, roles and responsibilities, coordination
with the military services, and plans for delivering and
scaling critical AI capabilities.
Finally, I would like to better understand how the center
fits into the Department's cloud initiative.
Now, many claim that data is the new oil. Access to data,
data integrity, and data labelling are key issues facing the
Department.
In addition to hearing about the Joint AI Center, I look
forward to hearing from Mr. Deasy about how he is setting
standards and issuing other guidance to the services, agencies,
and other entities pertaining to these issues.
In August, through Chairwoman Stefanik's leadership, we
successfully authorized the National Security Commission on
Artificial Intelligence. I commend the Chair for her work. I
was proud to join her in that effort.
The commission has been tasked with comprehensively
examining U.S. advances in AI with regard to investments in
basic and advanced research, efforts to recruit top-notch
talent, ethical and safety considerations for military
applications, and strengthening our global competitive
advantage in the field.
I appreciate the DOD's partnership in standing up the
commission and look forward to hearing more about its plans to
prioritize funding and resources for the commission during
today's hearing.
I also look forward to hearing more about the division of
roles and responsibilities for the AI portfolio between the
Department's Under Secretary of Defense for Research and
Engineering and the Chief Information Officer, as well as
efforts to synthesize AI strategies and plans with the
services.
There is enormous momentum around AI, and it is exciting,
and it is critical that the U.S. capitalize on this momentum in
order to maintain its technological edge. As a matter of
national security, I strongly encourage the Department to
continue to strengthen its partnerships with academia and the
private sector, better leverage Federal labs, invest in
cutting-edge research, and continue to explore applications of
AI with the interagency to ensure that we remain at the
forefront of AI innovation.
Before I yield back, I, too, want to join Chairwoman
Stefanik in recognizing Dr. Mark Pepple, and Neve Schadler,
clerk, for their service to the committee as well as they
depart at the end of the year. I want to thank them for their
work. They have made great contributions to our work here on
the committee, and we are grateful for their service.
So thank you. And I yield back.
Ms. Stefanik. Thank you, Ranking Member Langevin.
I also want to welcome the chairman of the full committee,
Chairman Thornberry, who is here with us today. And this issue
is of deep interest to him, as reflective of the interest of
committee members beyond this subcommittee.
Without objection, the witnesses' prepared statements will
be made part of the record. I ask that you please keep your
opening remarks to no more than 5 minutes.
And Dr. Porter, we will begin with you.
STATEMENT OF DR. LISA PORTER, DEPUTY UNDER SECRETARY OF DEFENSE
FOR RESEARCH AND ENGINEERING, DEPARTMENT OF DEFENSE
Dr. Porter. Good afternoon, Chairwoman Stefanik, Ranking
Member Langevin, and distinguished members of the subcommittee.
Thank you for inviting me to appear before you today to discuss
artificial intelligence, particularly as it relates to national
security applications.
As this subcommittee knows, artificial intelligence, or AI,
is not a new thing. As long as there have been computers, there
have been engineers who have dreamed of enabling machines to
think the way humans do. In fact, DARPA [Defense Advanced
Research Projects Agency] funded much of the early work in AI
decades ago.
Today we are experiencing an explosion of interest in a
subfield of AI called machine learning, where algorithms have
become remarkably good at classification and prediction tasks
when they can be trained on very large amounts of data.
There are numerous examples of successful applications of
machine learning techniques. Some of the obvious ones include
facial recognition in photographs and voice recognition on
smartphones.
However, there has also been a significant amount of hype
and confusion regarding the current state of the art. It is the
USD (R&E) [Under Secretary of Defense for Research and
Engineering] position that we must not abandon the tenets of
scientific rigor and discipline as we pursue the opportunities
that AI presents.
Today's AI capabilities offer potential solutions to many
defense-specific problems. Examples include object
identification in drone video or satellite imagery and
detection of cyber threats on networks. However, there are
several issues that must be addressed in order to effectively
apply AI to national security mission problems.
First, objective evaluation of performance requires the use
of quantitative metrics that are relevant to the specific use
case. In other words, AI systems that have been optimized for
commercial applications may not yield effective outcomes in
military applications.
Second, current AI systems require enormous amounts of
training data, and the preparation of that data in a format
that the algorithms can use, in turn, requires an enormous
amount of human labor.
Furthermore, AI systems that have been trained on one type
of data typically do not perform well on data that are
different from the training data. For example, algorithms that
are trained on internet images will generally underperform when
used on drone or satellite imagery.
Another well-known limitation of current systems is that
they cannot explain what they do, making them hard to trust.
Furthermore, current systems require robust processing
power.
And finally, current systems are susceptible to various
forms of spoofing, known as adversarial AI.
We are working to address these challenges and
vulnerabilities through multiple efforts, most of which will
lever the complementary roles of the Joint Artificial
Intelligence Center, the JAIC, and the USD(R&E) enterprise.
The JAIC will offer a means to rapidly determine the
appropriate metrics for operational impact for a variety of
applications, as well as the operational performance
limitations of current tools. And these insights will help
inform algorithm and system development across multiple
USD(R&E) efforts.
Furthermore, the JAIC's focus on scaling and integration
will drive innovation and data-curation techniques, while DARPA
will pursue algorithms that can be robustly trained with much
less data.
In order to address AI's trust issue, DARPA's Explainable
AI program aims to create machine learning techniques that
produce more explainable models while maintaining a high level
of performance. The High Performance Computing Modernization
Program is designing new systems that will provide ample
processing power for AI applications on the battlefield.
Finally, countering adversarial AI is one of the key focus
areas of DARPA's AI Next campaign.
Ultimately, as we look to the future, we anticipate a focus
on developing AI systems that have the ability to reason as
humans do, at least to some extent. Such a capability would
greatly amplify the utility of AI, enabling AI systems to
become true partners with their human counterparts in problem
solving.
It is important that we continue to pursue cutting-edge
research in AI, especially given the significant investments
our adversaries are making. We are therefore grateful for the
leadership and support that the members of the subcommittee
have shown regarding AI.
We also appreciate the establishment of the National
Security Commission on AI, whose charter is appropriately
focused on key areas that must be assessed objectively to
assure that the U.S. maintains a leadership position in AI-
enabled technologies and systems.
Thank you for your interest in this important topic, and I
look forward to answering your questions.
[The prepared statement of Dr. Porter can be found in the
Appendix on page 27.]
Ms. Stefanik. Thank you.
Mr. Deasy.
STATEMENT OF DANA DEASY, CHIEF INFORMATION OFFICER, DEPARTMENT
OF DEFENSE
Mr. Deasy. Good afternoon, Ms. Chairwoman, Ranking Member,
and distinguished members of the subcommittee. I thank you for
this opportunity to testify on the Department's progress in AI
adoption and the establishment of the Joint Artificial
Intelligence Center.
I am Dana Deasy, the Department of Defense Chief
Information Officer. I am the principal adviser to the
Secretary of Defense for a set of responsibilities that
integrate together to ensure that DOD has the information and
communications technology capabilities needed to enable the
broad set of missions we perform as a joint force.
The application of AI is rapidly changing a wide range of
businesses and industries. The 2018 National Defense Strategy
[NDS] foresees that ongoing advances in AI will change society
and, ultimately, the character of war.
In June, Deputy Secretary Shanahan directed my office to
establish the Joint Artificial Intelligence Center as a focal
point for that endeavor. In parallel, DOD submitted its first
AI Strategy to Congress, an annex to the NDS. JAIC's formation
also dovetailed section 238 of the latest NDAA.
Going forward, JAIC will benefit from and help bring into
reality recommendations of the National Security Commission on
AI.
In talking about the Joint Artificial Intelligence Center,
I would like to highlight three themes today.
The first is delivering AI-enabled capabilities at speed.
JAIC is collaborating now with teams across DOD to
systematically identify, prioritize, and select mission needs,
and then rapidly execute a sequence of cross-functional use
cases that demonstrate value and spur momentum.
Projects fall into two main categories: National Mission
Initiatives [NMI] and Component Mission Initiatives [CMI]. NMIs
are driven and executed by JAIC, whereas CMIs are component-led
and are able to make use of JAIC's common tools, libraries,
best practices, and more.
I will note that our new emphasis on rapid, iterative
delivery of AI complements the Department's ongoing work at the
other end of the AI spectrum and fundamental research, as Dr.
Porter shared with you today.
Two examples of early projects. First, predictive
maintenance. The NMI helps address Secretary Mattis' direction
to the services to improve their maintenance readiness rates
and offers well-defined return on investment criteria.
A second example, humanitarian assistance and disaster
relief. This NMI is an open mission to apply AI to saving lives
and livelihood. We are applying lessons learned and reusable
tools from the DOD's AI pathfinder, Project Maven, to field AI
capabilities in support of such events as hurricanes and
wildfires.
The second theme is all about scale. JAIC's early projects
serve a dual purpose: to deliver new capabilities to end-users,
as well as to incrementally develop the common foundation that
is essential for scaling AI's impact across the DOD. This means
shared data, reusable tools, libraries, standards, and AI cloud
and edge services that help jump-start new projects.
We will put this in place, this foundation, in a manner
that aligns with the DOD enterprise cloud adoption. Let me
underscore that point. Our enterprise approach for AI and
enterprise cloud adoption via the DOD-wide cloud strategy are
mutually reinforcing, mutually dependent undertakings.
The third theme is we build the initial JAIC team. It is
all about talent. And this will be represented across all the
services and all components.
Today we have assembled a force of nearly 30 individuals.
Going forward, it is essential that JAIC attract and cultivate
a select group of mission-driven, world-class AI talent,
including pulling these experts into service from industry.
In closing, 2 weeks ago, in front of 380 companies and
academic institutions at DOD's AI Industry Day, I announced we
had achieved a significant milestone: JAIC is now up and
running and open for business.
I look forward to continuing to work with Congress in this
critical area in an ongoing dialogue on our progress in AI
adoption and the ways in which JAIC is being used to accelerate
that progress.
Thank you for this opportunity to testify this afternoon,
and I look forward to your questions.
[The prepared statement of Mr. Deasy can be found in the
Appendix on page 32.]
Ms. Stefanik. Thank you for those opening statements.
I want to ask a broad question to begin. I am deeply
concerned, as I read headline after headline announcing the
U.S.'s looming defeat when it comes to the global race for AI
dominance. It seems like every week there is a new headline.
I want to quote a recent article, of the fall of this year,
in Foreign Policy:
``There will not be one exclusively military AI arms race.
There will instead be many AI arms races as countries (and,
sometimes, violent nonstate actors) develop new algorithms or
apply private sector algorithms to help them accomplish
particular tasks.
``In North America, the private sector invested some $15
billion to $23 billion in AI in 2016. That is more than 10
times what the U.S. Government spent on unclassified AI
programs that same year.
``China says it already holds more than 20 percent of
patents in the field and plans to build its AI sector to be
worth $150 billion by 2030.''
My broad question is, are we falling behind already? If so,
how far behind? And how do we jump-start it to make sure that
we do not lose our technological edge when it comes to AI?
Dr. Porter, I will start with you.
Dr. Porter. So I would say we are not behind. Right now we
are actually ahead. However, we are in danger of losing that
leadership position. So your concern is certainly valid.
Ms. Stefanik. And let me, I am going to jump in there. How
are we ahead? How do we measure that?
Dr. Porter. Absolutely. So there are a lot of ways to
assess that, but if you look in terms of our talent,
particularly in our academic base, the United States, along
with our partners in the U.K. [United Kingdom] and Canada in
particular, are seen, even by the Chinese, as having quite a
lead.
And I will tell you the reason for that--and this is an
important point to make, and I alluded to it in my opening--
DARPA, in particular, and also the NSF [National Science
Foundation], have been funding this field for decades. So we
have built an extremely robust and deep bench in the
disciplines that are required to advance this field.
Even when you hear about AI winters in the past and so
forth, the United States continued to invest robustly in this
domain for decades. And then we have this vibrant private
sector that is able to turn around and take that research and
rapidly convert it to commercial products and create new
markets.
So we have a lot going for us, and I believe China has,
unfortunately--or fortunately if you are from China, I guess--
they have figured out that that is one of the key ingredients
to our success, that we have a multitiered approach in this
country to ensuring that we continue to stay on the cutting
edge. We invest heavily in academia, we invest heavily in our
labs, and then we figure out how to convert those investments
quickly and rapidly into products and creating new markets.
They recognize that, and that is why you are seeing a
tremendous increase in their investments, particularly in
academic as well as startup community, which I think you were
alluding to.
Ms. Stefanik. What about the race for data? You talked
about some of the challenges that we face. Obviously, data is
the fuel for AI. So when we talk about an AI arms race, part of
that is a race for data and being able to analyze data in a
comprehensive way. Can you comment on that, Dr. Porter? Beyond
the challenges, what are our solutions to ensure that we have
the fuel to help propel our AI research?
Dr. Porter. Yes. So data is a very key element. I have
commented on this and so has The New York Times. Probably you
saw the article a couple weeks ago highlighting how China has
created these places where people are sitting there and
essentially labelling data, right, for their needs.
So one of the things we have to do is we have to be smarter
about how we understand how these algorithms are working, and
that is why DARPA always looks at a problem and says: Okay, how
do we do this better? Our industrial sector also recognizes
these challenges, so they are going to look at how do we do
this better.
So it is not going to be just about how do we get a lot of
data; it is going to be about how do we develop algorithms that
don't need as much data; how do we develop algorithms that we
trust as they are using the data and they are evolving the
data.
And this is where the JAIC comes in. I think this is why
the powerful connection between R&E and JAIC is so important,
because we have an opportunity now, as we want to test out new
ideas, we get to a point where we use something like DIU
[Defense Innovation Unit], that says, hey, what is going on in
the private sector; how are they trying out new things; let's
try to prototype; and then get them out into the operational
place more quickly and say, how is that actually working?
So this can be a very powerful way for us to accelerate the
experimentation that is going to be continual to stay ahead of
the game, because it can't just be about labelling data. It has
got to be about being smarter using the data that you have got.
Ms. Stefanik. Thank you, Dr. Porter.
Mr. Langevin.
Mr. Langevin. Thank you, Madam Chair.
And thank you again to both of our witnesses for your
testimony today.
Dr. Porter, I would like to start by asking you to expand
on what you talked about in terms of the DARPA project and
using less data to get better outcomes, if you want to talk a
little bit more about that.
Dr. Porter. Sure. That is just one of the many areas that
DARPA has been focusing on. I think some of you are aware of
their AI Next campaign, which they have announced publicly, and
they are trying to address all of these weaknesses that--or
several of the weaknesses, I should say, that I outlined, and
one of them has to do with this reality of the big data
problem.
Folks on the cutting edge are now talking about how we
can't just be using traditional machine learning. What is the
next step? How do we combine other elements to get after this?
And if I can brag about something that DARPA has done
recently, because I think it will give you some hope when I say
I think the United States has ways to stay ahead. They recently
started something called AI Exploration, and this is a way that
they very rapidly get money out, particularly into academia,
and the labs, and the small businesses, to say, all right, by
the time I announce my concept I want you guys to go after,
within 90 days I am going to get you the money. Not from the
time I tell you, you are selected, but from the time I post it
till you get the money.
They have already done this once and within 90 days they
had 16 awards out, each about a million dollars or so. And the
problem they are tackling is, can we bring some physics into
machine learning so that we don't need as much data and we
don't have to worry so much about these fragile and brittle
things that I was talking about.
So I am telling you this story because I think you have got
a lot of innovation going on within the DOD enterprise to say:
How do we get to speed, as well as scale? And this is what Dana
and I are going to continually try to work together on, is how
do we move faster, because AI is all about speed. It really is.
This is one of those domains where things are just going very,
very quickly.
Mr. Langevin. Thank you, Dr. Porter.
Mr. Deasy, as I mentioned in my opening statement, data is
the fuel that powers AI and machine learning. So what efforts
are you undertaking to promote policies and practices that
ensure DOD owns data collected under its authorities?
Mr. Deasy. So it is interesting, when I joined the
Department and we kicked off the JAIC, sir, one of the earliest
questions I got asked was: What are going to be some of the
earliest stumbling blocks you are going to face in the
successful standup of JAIC? And I said: I can almost predict
now that as we roll out the first two, three, four
applications, the thing that will be hitting us over and over
again will be data.
And what do I mean by that? It will be: Where is the single
source of the truth coming from? How do you ingest it? What are
its formats? Do we have duplicate data? And how do we bring it
together.
Part of the reason why you heard me comment in my opening
remarks about the integration of cloud: cloud provides us the
physical capacity to take this enormous amount of data and
bring it together.
True, it will still continue to sit in different formats.
But what we will do in development of JAIC is we will start to
define with different problem sets and different algorithms
what is the expectation in terms of the data standards that
need to be deployed.
So if we are looking at audio versus we are looking at
image data, or if we are looking at textual or good old-
fashioned tables, one of the things that the JAIC will need to
do is--two things--technically describe what it is we need to
do to ingest the data and what are the tools; and then two is
what are the policies and standards that need to be put in
place on the correct formats of data as people develop new
systems going forward.
Mr. Langevin. Yeah, but you didn't answer my question about
what are we doing to ensure that DOD owns the data collected
under its authorities. I need that. But I also need to ask you,
how are you incorporating publicly available data sets into
your efforts, and have you had challenges accessing data sets
owned by other entities? So those two.
Mr. Deasy. Yeah. Too early from the standpoint of JAIC, as
JAIC is just stood up. So we haven't had a program right now
where we are actually accessing public data. DARPA may be in a
position to describe what they have done on that standpoint.
But, indeed, there will be programs eventually where we will
need to incorporate that, and we will have to be very clear on
the ownership of that data.
If the data is truly being created, whether it be from an
intel [intelligence] community, a mission partner, very clear
rules of the roads will have to be established early on as to
the ownership of that data.
Part of our job in standing up JAIC--and I need to stress
this throughout today--is that this is going to be an iterative
learning cycle. We are going to take something in, we are going
to learn what are the issues.
One of the issues, the one you probably bring up here, who
owns the data? Where is the legal authorities for that data?
And we are going to have to actually take these on a case-by-
case basis, then develop ongoing policy that can be applied for
more missions as we go forward.
Mr. Langevin. Okay. I am glad we are thinking about these
things now for sure.
I know my time is expired. I have other questions. If we
get to a second round, I will ask those then. If not, I will
submit them for the record. But thank you, and I yield back.
Ms. Stefanik. Thanks.
Dr. Abraham.
Dr. Abraham. [Inaudible--off mic] but certainly on other
sides of this globe. I refer to even gene editing here. Of late
we have seen that go awry on the eastern part of the globe. So
I worry about the scientific discipline that will be involved
with our data.
To follow up on Jim's question a little bit, Mr. Deasy, the
algorithms that are constructed by, I am assuming, commercial
industry, they own that data. Am I correct there, the way the
law stands as of now?
Mr. Deasy. Yes. So in the case of some solutions that we
built, for example, in Maven, where we have used partners, part
of that case will be commercial available solutions and
algorithms that they will own.
Dr. Abraham. And you said JAIC wants to incorporate people
from industry to be part of the total family.
Mr. Deasy. Absolutely.
Dr. Abraham. Is that a correct statement?
Mr. Deasy. It will be a combination of those solutions that
will be developed by our own organization, JAIC, and those that
will be developed through partners. So there will be no single
solution where we will probably come from either all
commercial, internally, but we will be using a combination of
both.
Dr. Abraham. But I just go back to a few years ago where
the VA [Department of Veterans Affairs] had a physician develop
a drug that was used, and who owned that particular patent was
a big mess.
So I just implore--and I am sure you are ahead of the curve
here--but if we have the rules of the roads in place before
those algorithms are developed and then we have to get into
this debate, I think it is prudent to do that.
Madam Chair, I yield back. Thank you.
Ms. Stefanik. Thank you, Dr. Abraham.
Mr. Larsen.
Mr. Larsen. Thank you very much. Thanks for coming today.
So one of the criticisms or, I guess, concerns when we
compare ourselves to our competitors, especially China, is they
can take a top-down approach, sort of drive everything through
state-owned enterprises, through what they consider public
financing, where we have to have a more of a bottom-up approach
because we have such a very active private sector innovative
economy.
And so how are you trying to balance that? How are you
trying to drive the innovation so that it creates options for
the DOD to pick from? Because we are probably not going to
drive it to any one solution, but a set of solutions, and then
you can choose partners as you move forward. Who might be best
to answer that?
Mr. Deasy. Well, I will talk about it from an operational
production.
Mr. Larsen. Yeah, sure.
Mr. Deasy. We will let Dr. Porter discuss it more from a
research and science.
Mr. Larsen. Yeah.
Mr. Deasy. So the way that JAIC is being established is
going to be very much a hub-and-spoke model. There will be a
physical entity that we are creating in the Washington, DC,
area. But we recognize that we are going to need talents that
are going to exist, for example, outside in the academia
environment.
So part of our spoke model is we will be establishing
locations next to academic environments, we are actually in the
process of selecting those right now, where they will have
certain skill sets. And so what we are actually doing is going
through an inventory process of identifying what are the
problems we believe are most in need to solve for and what
institutions.
Between that and the fact in our AI Day that we ran
recently, the reason we ran that day was we are now getting in
white papers that are coming in from the commercial sector, as
well as the academic sector, starting to describe what are
their solutions against the problem sets we are trying to
solve. We are right now in the case of actually building out an
inventory of these solution sets.
Mr. Larsen. Interesting. And Dr. Porter.
Dr. Porter. So one of the things that when Dana and I talk
about this--and this may be a helpful thing if you can
visualize it--we think about near, mid, and long term. And in
the near term, of course, that is where JAIC resides.
And in the mid term gets to kind of your question. This is
where DIU, for example, says: All right, what is going on in
the private sector? Because those problems that I articulated
that we have to address, the private sector has to address as
well.
So if an algorithm isn't very robust, my recommender system
doesn't tell you that the movie that I am recommending to you
makes any sense to you, you are not going to use my system
either, and that is going to cause revenue problems. So I have
got to solve that problem.
So the DIU, in places like that, they look and say: Well,
what are they coming up with in the near term or the mid term
that we can fold back in and test.
And so DIU has a very effective way of basically doing
proof of principle and projects at a lower level and say: Okay,
JAIC, I think we have got this; we want you to scale it up and
really test it, and wring it out, and tell us where we are
missing things and continue to iterate. So that is kind of a
unique capability.
Now, if you go a little further out, to your point, some of
these problems the commercial sector are not going to solve,
because they are hard or they are not relevant to their
markets. That is why we need a DARPA. That is why we need our
national labs. That is why we need our underpinning across that
entire spectrum of our academic experts who can guide us in the
near, mid, and far term to think about what can we solve now
and where do we need to have long-term strategic investment.
And, again, the willingness as a nation to continue to
invest in the hard problems even as some of those are going to
lead to missteps and we are going to have to try again, right?
That is that high-risk, high-payoff realm. So we have to cover
that entire spectrum. If we do that, we can optimize on
benefitting from the private sector, as well as pushing to
solve the problems we care about most that are hard.
I am sorry, I went over the time.
Mr. Larsen. No, that is fine. I have a little less than a
minute.
So the Center for Strategic and International Studies just
published a report late last month on AI and national security,
and the argument they made was the need for robust supporting
capabilities or an ecosystem around AI, especially within DOD.
And I don't know if your folks have evaluated that. But it
might be--it is not an easy read for people like me, but it is
a good read for folks like you to use it maybe as a marker
standard to compare yourself against. There are other folks
writing about this as well. But I would commend that to you.
And they outline a variety of areas: trust and security of
AI, the people part of it, the digital capability, and the
policy. Which you have already outlined some of those concerns
and the things you are trying to focus on. I would just lay
that out there if you are thinking about how to compare
yourself to where maybe you ought to be versus where you are
today.
And I will wait for a second round. Thanks.
Ms. Stefanik. Mr. Hice.
Mr. Hice. Thank you, Madam Chair.
China has identified AI as a strategic technology for them,
and they plan to develop an AI industry worth over $21 billion
by 2020.
As we all know here in this room, China also has a strong
history of both government and industrial espionage, and this
just creates a great deal of concern personally.
So what are we doing to protect ourselves, specifically
from China, but really from anyone, from hackers? What are we
doing to make sure our AI program remains ours?
Mr. Deasy. Okay, I will start.
So I would say a couple things on that. Interestingly
enough, we are actually going to apply AI to help us address
this problem.
So I mentioned earlier we have two types of initiatives,
National Mission Initiatives and Component Mission Initiatives.
We are actually doing some work right now to start to evaluate
with U.S. Cyber Command, how is it we can apply AI in pattern
recognition in signatures, where are you looking for anomalies
that are going on in your network, and how can you use AI to
quickly assess that there has been a change to what is a normal
pattern.
If you think about how hackers actually try to penetrate,
they will go to the point of least resistance, and once they
are in, they will go laterally. And then what you are looking
for is exfiltration.
And so we believe actually AI will be a very good machine
use case for looking at how we look at signatures and patterns
of data across our network and actually use that to help ensure
that we don't have exfiltration occurring from folks like the
Chinese.
Mr. Hice. Dr. Porter, would you like to add anything to
that?
Dr. Porter. Sure. I think you are highlighting an extremely
important point. I think there are specific technical
approaches that we are going to be working. And that example is
a good one, because we are not going to get it all right, and
it is going to be iterative, and DARPA, in fact, is also
looking at this from their perspective.
But I would want to emphasize the broader point you are
making. I think we have to be vigilant and aware of this
problem. I actually spent time at In-Q-Tel and I have spent
time in the intel community. And I know this committee was
briefed, I think back in June, about the Thousand Talents
Program that China has, and I know you guys were told at that
time exactly your point: They have a goal of facilitating both
legal and illicit transfer of U.S. technology, intellectual
property, and know-how, and we have to be cognizant of that in
the community.
So across our research domain and spectrum we are thinking
about that. It isn't just about protecting against hacking,
although that is certainly a big part of it, it is all of those
ways that they have to try to capture that intellectual
property, which I think is what you were alluding to.
Mr. Hice. Absolutely it is. And I know some of this
probably would be best served in another environment than this.
Dr. Porter. Right.
Mr. Hice. But I would like to dive deeper into this issue
if we can in another setting.
But going back to what Mr. Larsen said, I want to just get
a little more clarity. How do you plan to recruit talented data
engineers and scientists? Specifically in the near term, I
guess.
Mr. Deasy. Right. So right now, we have approximately 30
people inside. It is a combination of civilians, which are DOD
employees, as well as military.
The philosophy is over time we are going to need to
actually build out an internal capability that will include
people inside the military.
So what we have done recently is we brought in 10 very
highly talented, skilled individuals from the various services
into JAIC. We are going to team them with data scientists,
``been there, done it'' people that we are recruiting. And the
idea is to use this pairing system so people can leave JAIC, go
back into the services, and then use that to increase the
flywheel.
How we are recruiting people is a combination of commercial
contacts, academia contacts, think tank contacts. We have quite
a list of people that we are currently identifying.
We expect at some point we may have to put something in
place like the Cyber Excepted Service, which is going to allow
us to recruit in a way that has a lot of additional speed. It
is going to have to handle compensation differently. And it is
going to handle how we onboard them in a much better fashion
than you would normally onboard into government.
Mr. Hice. Okay. Thank you very much. I yield back.
Ms. Stefanik. Mr. Veasey.
Mr. Veasey. Thank you, Madam Chair.
I wanted to ask Dr. Porter or Mr. Deasy about the $2
billion that DARPA has announced as a multiyear investment for
AI Next.
Can you explain to me exactly what the $2 billion is going
to be used for? Is it just to sort of develop a kind of a basic
groundwork on how we should move forward? Or is it going to
advance specific technologies?
Dr. Porter. So it is kind of both, because that is what
DARPA does. Now, to be clear, the $2 billion is over 5 years,
so it is roughly $400 million a year. And they have several
thrust areas that target these problems that I was talking
about.
So one I already told you about, this Exploration program,
and this is that really rapid getting stuff out there and
getting really great ideas funded. So we do exactly what you
just said and provide that foundation for larger efforts.
There is also a lot of focus on what I mentioned,
adversarial AI. This is where it has been proven, and if you
read the popular press there are these examples that people are
publishing almost daily now, where they can spoof AI systems
pretty easily.
One of the ones that is notable, because in the self-
driving car community they really took note of this, is there
is a team at Berkeley, at the school out in California, they
put tape on stop signs. And when you put the tape on the stop
signs, the AI system thought the stop sign was a speed limit
sign for 45 miles an hour. So you can imagine that is a little
bit of a problem, right?
And there are countless examples of this now. It is almost
a game now where people are showing all the ways they can spoof
these systems.
So, obviously, if we are going to trust this and we are
going to apply it to things where there are high stakes, i.e.,
the DOD mission, we have got to do much better at understanding
how we ensure that people can't spoof our systems. There is a
lot of research to be done there, and that is one of the key
thrust areas in the AI Next program.
Mr. Veasey. As we try to gain a better understanding, is
the $400 million, is that like a good starting number? Or where
does the number ideally need to be in order for us to sort of
stay on track?
And right now, I think you had mentioned earlier that when
it comes to China and other competitors, that we actually are
ahead. But financially, like where do we need to be to make
sure that we stay ahead and that we can continue to work on
things like making this AI more smarter, to where tape can't
throw it off?
Dr. Porter. You have got it.
I think it is a reasonable investment level. And one of the
things DARPA likes to emphasize, which I fully agree with, is
it is not just the amount of money you invest in, it is how you
do it.
So DARPA has a model, right, where they try things that a
lot of people won't try because it is risky and it may not pay
off. And if it doesn't work, no harm, no foul, we will try
something else, because we are trying to pursue the really hard
things.
That whole model that DARPA has is pretty unique, and, in
fact, when you couple that with a robust funding effort, as the
$2 billion over 5 years is, you can actually get significant
jump-aheads. And that is really what I am trying to emphasize
here. That is one of our unique secret sauce ingredients in the
United States.
Mr. Veasey. Thank you very much.
Madam Chair, I yield back.
Ms. Stefanik. Mr. Bacon.
Mr. Bacon. Thank you both for being here. I am grateful for
your expertise and sharing it.
It seems to me that until a few years back, or maybe even a
decade ago, DOD would drive a lot of the technology. The
private sectors would then leverage that. And then we saw a
period of time where there was probably a lot of even synergy.
But in my visits recently to the private sector and some of
the larger companies, it seems to me they are producing
technology faster than DOD can install it, or with the
requirements process, testing, by the time we do field it, it
is already 2 to 3 years out of date, if not more.
Are we positioning ourselves right in the AI to stay
abreast and not fall behind?
Dr. Porter. So I will start, and then I will, because I
think this is a joint answer.
I think you highlighted the problem we are very much
interested, the two of us, in trying to address. If we do
nothing else, we are going to still have this problem, because
even if DARPA gets us ahead of the game and the private sector
takes off with those ideas and goes their own way and creates
these great products, we have got to have a way to more rapidly
transition that innovation back in, learn from it, and continue
the cycle.
And both Dana and I have talked multiple times about the
speed challenge, and this is why we are really trying very hard
to figure out, how do we coordinate that cycle, so that spin
cycle, if you will, so that we get multiple spins very quickly,
rather than three, four, five multiple-year cycles just to
insert something.
It is not solved, but I think what you are seeing here is a
real serious attempt by the DOD to say, let's line this up so
that we can improve this.
Would you not agree?
Mr. Deasy. Yeah. As someone who spent the majority of my
career in the private sector, I am often asked when I arrive,
what is it I have noticed most, and I say clock speed. How fast
we can embrace, either decide to work with something, get rid
of it and move on.
And one of the reasons we created the relationship we
created was you need two things in AI to be successful. You
need a maniacal focus on the here and now of operationalize and
getting things up and running, and that is that flywheel I talk
about. But you also need an intense focus on where the future
is going, where the science is going.
And you need a place to take that science. In this case,
what DARPA develops. Bring it in, rapidly decide whether or not
it can work or not work. If it doesn't work, move on. Tell
DARPA that is the case. Or if it is working and it just needs
tweaking, then let's do that.
This is why we think this model we have put in place is
actually going to help to address the very problem you raise on
how do we get the flywheel of innovation moving at a lot faster
clock speed.
Mr. Bacon. Are we confident DARPA is abreast of all of what
the various private sector companies are doing? I mean, do they
have their fingers on the pulse of a lot of different
companies? Are we confident of that?
Dr. Porter. So we are confident, but that is why we also
have DIU in our quiver. Because, as you know, DIU sits out in
Silicon Valley, but also sits in Austin, Texas, sits in Boston,
and it is keeping its finger on the pulse.
And, again, where they are going to really see the
innovation is a little bit nearer term, but it is that nice
bridge between where DARPA may be looking a little further out,
DIU is going to see where opportunities are in the next 12 to
24 months, which is much shorter than where DARPA typically
looks.
So we try to cover that landscape appropriately----
Mr. Bacon. Right
Dr. Porter [continuing]. So that we are seeing everything
we should be seeing.
Mr. Bacon. Two follow-on questions. Do we need to make any
revisions to our acquisition rules processes to help you out,
one.
Two, when I was recently visiting a company this past week,
they would say they come up with new technology, but because
the DOD didn't have a requirement for it, they didn't want to
really look at it. However, later on, they would say, yeah,
basically the requirements were shortsighted because they
didn't realize what the technology--what could be executed or
applied.
So my question, two of them, do we need to make any
revisions to our acquisition system? And two, do your
requirements keep up with some of the far-ranging technologies
that you are seeing in AI?
Thanks.
Mr. Deasy. So I will start with the first half of that, the
acquisition.
What I tell people often is, one of the things that we
struggle with at the DOD and government is what I call a
startup mentality. How you start AI is a very iterative
process. And many times the acquisition cycles are asking you
to define 30, 60, 90, 2, 3, 1 year, 2 years out, what the end
state will look like.
I am just trying to get the end state identified for the
next 90 days, 120 days, and then allow us to create this, what
I will call this iterative approach for how we are going to
build out. We are going to try a solution, we may acquire a
product. We will say that product didn't quite meet the needs,
and then we are going to need to go back out in a very rapid
cycle.
So, yes, I do believe there will need changes, and I
believe it is going to be, how do we move to a more startup
mentality when looking at technologies like AI?
Ms. Stefanik. We will now move to the second round of
questions and get through as many as we can before they call
votes.
My second question has to do with a previous testimony
before this committee. I believe it was Deputy Secretary
Shanahan talked about the fact that there are hundreds of AI
projects and programs within the DOD.
Can you speak, Mr. Deasy, to how we plan on integrating
those programs into the JAIC and how that process is going? And
can you also highlight one of the best examples of an AI
program that was started within the DOD that we can learn from?
Mr. Deasy. Yeah. So clearly what the Deputy was referring
to is there are a lot of programs that are using data learning,
machine learning, cognitive. You have to be quite thoughtful
when describing what is the universe of AI. I would argue that
some of those programs, when you really kind of dig under the
covers, are more business analytics, as they are as to true,
what I will call, machine learning.
With that said, there is no doubt that one of the biggest
benefits that JAIC will bring is trying to reduce the
replication and the duplication of tools, processes, and,
frankly, methodologies that are being used.
A good example of this--and it actually brings DIU into
it--is you think about the predictive maintenance. So this is
an area where how do you look at helicopters, planes, ships,
anything where there is a need to reduce the waste and the
cycle time of readiness. This is an example where DIU went out
and did some work, found some solutions in the marketplace.
They are now bringing that to us.
One of our first initiatives is predictive maintenance, and
we are actually going to use the learnings from DIU and the
commercial offerings as a way we are going to jump-start the
predictive maintenance.
Ms. Stefanik. Thank you.
Mr. Langevin.
Mr. Langevin. Thank you.
So for Mr. Deasy, our military force projection
capabilities are developed and tested almost entirely within
the continental United States, but the nature of warfighting is
largely expeditionary. That is, the great majority of
warfighting is going to occur far away from the wide
infrastructure and domestic regulatory constraints of the U.S.,
requiring flexible access to maneuver within a different
electromagnetic environment.
What role do you see for AI in overcoming this challenge?
And how would it potentially apply to other domains like space
and cyber?
Mr. Deasy. So interesting enough, I just came back from a
Five Eyes meeting over in the U.K. in which we
discussed with our mission partners what is the role that AI
can play in a lot of spaces. You mentioned the one,
electromagnetic spectrum. I mean, the nature of electronic
warfare is such that trying to degrade, spoof, and change the
nature of spectrum is such that clearly AI can play a role in
being able to quickly assess where spectrum has been
compromised and how do you then change the nature of the use of
that spectrum.
---------------------------------------------------------------------------
Australia, Canada, New Zealand, United Kingdom, and United
States intelligence alliance.
---------------------------------------------------------------------------
Another example is, if you think about mission partner
networks and how we need to share data in a classified or
confidential manner, we see that AI will be able to use, much
to my earlier comment, patterns and changes of behavior as we
are sharing data across our mission partner networks. And so we
have had conversations recently with our partners on what is it
that we should be doing more joined up in the matter of these
AI initiatives.
Mr. Langevin. Okay. Thank you. That is encouraging.
Also, getting back to data, what efforts are you taking to
set standards and guidance for data integrity? And finally,
what efforts are being taken to provide for a common lexicon
for AI and machine learning?
Mr. Deasy. So on the data front, one of the things that we
are doing, for example, is we are working now with the CMO
[Chief Management Officer] office. They have actually hired a
Chief Data Officer. And on the reform side, we are doing some
early work, because I have been quite a proponent of saying
that we are going to have to solve for do we really understand
where the sources of our data come from, what I like to refer
to as the single source of the truth.
So we are partnering with the CMO and the Chief Data
Management Officer to start to identify what are those going to
be, those problematic data sets, where we are going to have to
get clearer standards, especially in the back office area of
reform. That is the area we are focusing on right now in the
Chief Data Management Officer.
Mr. Langevin. Dr. Porter, do you have anything to add to on
that?
Dr. Porter. Regarding the data integrity issue I think----
Mr. Langevin. Press your mic [microphone].
Dr. Porter. Oops. I am sorry about that. Regarding data
integrity there is also a research component to that as well.
And, again, it gets back to, as people recognize how important
your data is to training your algorithms, they are going to try
to mess with your data, right?
And so there is both the how do you ensure you are thinking
about AI not in an isolated way, but as was raised earlier in
the context of cybersecurity and other elements in your system
that have to work together.
And so one of the things I like to emphasize, which I think
you were touching on when you asked questions about space and
so forth, AI doesn't really mean anything until you think about
it in the context of the larger system that you are using it
in.
So how does it apply to your mission usually means it has
to be part of a larger system. How does it get integrated in a
way that you don't open up vulnerabilities because you have
forgotten that, wow, if my data is really easy to get into
someone is going to mess with it, so that I am training on the
wrong thing, as an example.
So there are research elements of this, because we have to
take a system-level approach, as we do with all technology,
when we think about integrating it into operations.
Mr. Langevin. So the last question I had is, to what extent
are the Department's challenges based on development of AI
technology--e.g., data processing and neural network
algorithms--versus a lack of infrastructure, such as big data
repositories, compute power, and cloud capabilities?
Mr. Deasy. [Indaudible]
Mr. Langevin. Microphone.
Mr. Deasy. Thank you, sir. I will start with that.
So I mentioned earlier the cloud. If you kind of step back
for a second and say, what has happened that has allowed AI to
suddenly be on the forefront of all conversations? And I would
argue there is the data science behind this, and I would say we
have entered an era now where there is unlimited compute power.
AI needs a massive amount of computer power, a massive amount
of storage, and, of course, you need the algorithms behind it.
The reason why I have been so vocal and energized about
wanting to get to an enterprise cloud capability is I want to
provide the Department of Defense with a way to handle that
unlimited compute capacity, unlimited storage, on demand, as
needed, with high integrity.
And this is why I have been such a strong advocate about
pushing the need for an enterprise cloud solution, because the
enterprise cloud is going to become the foundation for which
all the data and all that compute power will reside on top of
and those algorithms will use.
And understand that when I talk about cloud, I am not
talking about a centralized, single repository. I am talking
about a world where we need to work in a decentralized world.
If you are out at the warfighter, tactical edge, and we need to
be able to work in what I will call a compromised, degraded
mode. So it is clouds that can handle the edge, all the way
that clouds can handle the central.
Mr. Langevin. Thank you. My time has expired, so I yield
back. Thanks.
Ms. Stefanik. Thank you very much to our witnesses.
Votes have been called. For other members who we didn't get
to your second round of questions, please submit your questions
for the record.
And thank you, Dr. Porter and Mr. Deasy, for the testimony
today. We look forward to discussing this in the next Congress.
And I know I look forward to working with Mr. Langevin on it.
Thanks.
Mr. Deasy. Thank you.
Ms. Stefanik. The hearing is adjourned.
[Whereupon, at 4:30 p.m., the subcommittee was adjourned.]
=======================================================================
A P P E N D I X
December 11, 2018
=======================================================================
PREPARED STATEMENTS SUBMITTED FOR THE RECORD
December 11, 2018
=======================================================================
[GRAPHICS NOT AVAILABLE IN TIFF FORMAT]
=======================================================================
QUESTIONS SUBMITTED BY MEMBERS POST HEARING
December 11, 2018
=======================================================================
QUESTIONS SUBMITTED BY MR. LARSEN
Mr. Larsen. How do you assess the ability of military recruits to
work with AI? What recommendations would you make to improve K-12 and
community college curricula in order to make sure military recruits
have the necessary skills and are appropriately prepared to work with
AI-related applications?
Mr. Deasy. While some people joining the military today may have
skills suited for working with AI, overall we assess that the current
state of the existing workforce and military recruitment pipeline is a
critical shortfall for DOD. Although means of quantifying this
shortfall are still emerging, directional industry benchmarks indicate
DOD should build capacity of several thousand people with AI-specific
skills, such as data scientists and data coders. National investments
in skills training and high-quality K-12 and community college
education would be a significant force multiplier for DOD. Classes in
computational thinking as early as middle school and again in high
school will help establish a foundation for AI skills that will pay
dividends in the DOD workforce. Other recommendations to ensure
military recruits have the necessary skills and are appropriately
prepared to work with AI-related applications include the following:
1. Accelerate the use of digital content and ``flipped classroom''
pedagogy. There has been a renaissance in digital content such as
massive open online courses (MOOC), ebooks, and YouTube videos. This
content represents a new category of learning experience that presents
several advantages for K-12 and community college--generally high
quality, low cost, scalable, and adaptable to the needs of an
individual or community. In-person teachers can complement the online
content, resulting in faster and more enjoyable learning experiences
(``flipped classroom pedagogy'').
2. Evaluate guidelines, measurements, and incentives for AI
education in K-12. To establish consistent, measurable standards for AI
education and training, guidelines, measurements, and incentives should
be established across the country for curricula or key skills. As an
example of an external effort underway, the Association for the
Advancement of Artificial Intelligence (AAAI) and the Computer Science
Teachers Association (CSTA) are in the process of formulating
guidelines that will define what students in each grade should know in
AI.
3. Launch public-private partnerships, including open missions to
use AI to solve problems of societal significance. The use of public-
private partnerships can bring AI education to more K-12 classrooms
throughout the country. One type of partnership involves bringing to K-
12 and community colleges national security challenges and forming an
open mission to produce innovative AI technology to address real-world
problems. Such initiatives would enhance AI education, generate
excitement about working with the government, and inform potential
recruits of AI-related opportunities within the military. Similar
cyberspace initiatives have been very successful.
4. Establish clear pathways between K-12 and AI-enabled roles in
military service. Establishing a career track for computer scientists
in the military services provides potential recruits a clear path to
obtain sophisticated AI-related training and education. Designating AI-
related career fields allows for recruiting incentives such as
scholarships and bonuses.
5. Prioritize continued learning within military. The unique pace
of technological change in AI means that relevant knowledge decays more
rapidly than ever before. After entry, incentivizing continual learning
within military is imperative to maintain an ``AI ready'' workforce.
This should include expanding opportunities for internships,
fellowships, and exchanges between DOD and leading commercial AI
companies.
[all]