[Senate Report 116-289]
[From the U.S. Government Publishing Office]


                                                       Calendar No. 580
116th Congress 	    }			         {     	         Report
				SENATE
 2d Session         }                            {              116-289
_______________________________________________________________________
                                     

       IDENTIFYING OUTPUTS OF GENERATIVE ADVERSARIAL NETWORKS ACT

                               __________

                              R E P O R T

                                 of the

           COMMITTEE ON COMMERCE, SCIENCE, AND TRANSPORTATION

                                   on

                                S. 2904
                                

		[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]


                November 9, 2020.--Ordered to be printed



                               __________



                     U.S. GOVERNMENT PUBLISHING OFFICE

19-010			     WASHINGTON : 2020







       SENATE COMMITTEE ON COMMERCE, SCIENCE, AND TRANSPORTATION
                     one hundred sixteenth congress
                             second session

                 ROGER F. WICKER, Mississippi, Chairman
JOHN THUNE, South Dakota             MARIA CANTWELL, Washington
ROY BLUNT, Missouri                  AMY KLOBUCHAR, Minnesota
TED CRUZ, Texas                      RICHARD BLUMENTHAL, Connecticut
DEB FISCHER, Nebraska                BRIAN SCHATZ, Hawaii
JERRY MORAN, Kansas                  EDWARD J. MARKEY, Massachusetts
DAN SULLIVAN, Alaska                 TOM UDALL, New Mexico
CORY GARDNER, Colorado               GARY C. PETERS, Michigan
MARSHA BLACKBURN, Tennessee          TAMMY BALDWIN, Wisconsin
SHELLEY MOORE CAPITO, West Virginia  TAMMY DUCKWORTH, Illinois
MIKE LEE, Utah                       JON TESTER, Montana
RON JOHNSON, Wisconsin               KYRSTEN SINEMA, Arizona
TODD C. YOUNG, Indiana               JACKY ROSEN, Nevada
RICK SCOTT, Florida
                       John Keast, Staff Director
               David Strickland, Minority Staff Director




                                                       Calendar No. 580
116th Congress                                                   Report
                                 SENATE
 2d Session                                                     116-289

======================================================================



 
       IDENTIFYING OUTPUTS OF GENERATIVE ADVERSARIAL NETWORKS ACT

                                _______
                                

                November 9, 2020.--Ordered to be printed

                                _______
                                

       Mr. Wicker, from the Committee on Commerce, Science, and 
                Transportation, submitted the following

                              R E P O R T

                         [To accompany S. 2904]

      [Including cost estimate of the Congressional Budget Office]

    The Committee on Commerce, Science, and Transportation, to 
which was referred the bill (S. 2904) to direct the Director of 
the National Science Foundation to support research on the 
outputs that may be generated by generative adversarial 
networks, otherwise known as deepfakes, and other comparable 
techniques that may be developed in the future, and for other 
purposes, having considered the same, reports favorably thereon 
with an amendment (in the nature of a substitute) and 
recommends that the bill (as amended) do pass.

                          PURPOSE OF THE BILL

    This bill would direct the National Science Foundation 
(NSF) and the National Institute of Standards and Technology 
(NIST) to support research on the outputs by generative 
adversarial networks, commonly referred to as ``deepfakes.'' 
NSF would be required to support research on manipulated or 
synthesized content and information authenticity. NIST would be 
required to support research for the development of 
measurements and standards necessary to accelerate the 
development of the technological tools to examine the functions 
and outputs of generative adversarial networks or other 
technologies that synthesize or manipulate content.

                          BACKGROUND AND NEEDS

    Generative adversarial networks (GANs) are a type of 
algorithm that utilize two neural networks to produce synthetic 
data that appears real.\1\ One output of a GAN, commonly known 
as a deepfake, is a convincing digital video, imagery, or audio 
of events that never occurred. Most commonly, deepfakes appear 
in computer-assisted productions of highly believable audio and 
video in which real people appear to be saying things or doing 
actions that they never said or did.\2\ While the sophisticated 
manipulation of image and video applications can be solely 
based in personal amusement or artistic value, other such 
manipulations are for adversarial purposes such as propaganda 
and misinformation campaigns.\3\ Deepfakes have the potential 
to be used in information warfare or to manipulate 
elections.\4\ The development of standards for authentication 
or simply identification is a growing necessity as deepfakes 
move closer to illegal activities such as copyright 
infringements and data breaches.\5\
---------------------------------------------------------------------------
    \1\National Institute of Standards and Technology, ``Georgia Tech: 
The Unlinkable Data Challenge,'' Public Safety Communications Research 
Division, press release, updated Jan. 22, 2019 (https://www.nist.gov/
ctl/pscr/georgia-tech) (accessed Sep. 3, 2020).
    \2\David Chu et al., White Paper: Deep Fakery--An Action Plan, 2019 
(available at https://www.nsf.gov/mps/dms/documents/
Deep_Fakery_Workshop_Report.pdf) (accessed Sep. 3, 2020).
    \3\Dr. Matt Turek, ``Media Forensics (MediFor),'' Defense Advanced 
Research Projects Agency (https://www.darpa.mil/program/media-
forensics) (accessed Sep. 3, 2020).
    \4\Donie O'Sullivan, ``Lawmakers Warn of `Deepfake' Videos Ahead of 
2020 Election,'' CNN, Jan. 28, 2019 (https://www.cnn.com/2019/01/28/
tech/deepfake-lawmakers/index.html) (accessed Sep. 3, 2020).
    \5\Ian Sample, ``AI-generated Fake Videos Are Becoming More Common 
(and Convincing). Here's Why We Should Be Worried,'' Guardian, Jan. 13, 
2020 (available at https://www.theguardian.com/technology/2020/jan/13/
what-are-deepfakes-and-how-can-you-spot-them) (accessed Sep. 3, 2020).
---------------------------------------------------------------------------
    Due to the nature of the internet and the rapid advancement 
of technology, the production of deepfakes does not require 
complex processing systems. Academic and industrial researchers 
and even amateurs are able to acquire the computer resources 
necessary to create deepfakes. The increase in the prominence 
of deepfakes has also lead to an increase in quality awareness. 
Poor quality videos have become easier to detect, exposing 
inconsistencies such as lighting deficiencies and audio 
glitches.
    Agencies, universities, and private industry have launched 
research and development initiatives to enhance deepfake 
detection.\6\ Companies such as Amazon, Facebook, Microsoft, 
and others have joined the Deepfake Detection Challenge (DFDC), 
which invites people from around the world to build innovative 
new technologies to help in the detection of manipulated 
media.\7\ The Department of Defense, through the Defense 
Advanced Research Projects Agency (DARPA) has commissioned 
researchers across the United States to develop deepfake 
detection methods. DARPA, in collaboration with the University 
of Colorado Denver, is working to create convincing videos in 
order to develop technology to detect the real from the 
fake.\8\ Researchers at the Georgia Tech Research Institute 
have been working on a grand prize initiative to generate 
differentially private synthetic data using GANs. This data 
will then be able to be utilized for a variety of analysis 
efforts, which may include classification, regression, 
clustering, and answering unknown research questions.\9\ These 
developments have improved understanding of this issue and have 
shown the opportunity for agency and private partnerships, as 
well as the need for standards to detect bad actors.
---------------------------------------------------------------------------
    \6\Id.
    \7\Deepfake Detection Challenge, ``Building Tools to Detect 
Deepfakes Together,'' Facebook, 2019 (https://
deepfakedetectionchallenge.ai/) (accessed Sep. 3, 2020).
    \8\Donie O'Sullivan, ``When Seeing Is No Longer Believing: Inside 
the Pentagon's Race Against Deepfake Videos,'' CNN Business, 2019 
(https://www.cnn.com/interactive/2019/01/business/
pentagons-race-against-deepfakes/) (accessed Sep. 3, 2020).
    \9\National Institute of Standards and Technology, ``Georgia Tech: 
The Unlinkable Data Challenge,'' Public Safety Communications Research 
Division, press release, updated Jan. 22, 2019 (https://www.nist.gov/
ctl/pscr/georgia-tech) (accessed Sep. 3, 2020).
---------------------------------------------------------------------------

                         SUMMARY OF PROVISIONS

    S. 2904, as amended, would direct NSF and NIST to do the 
following:
   Support research on generative adversarial networks 
        or deepfakes.
   Support research on manipulated or synthesized 
        content and information authenticity.
   Support research for the development of measurements 
        and standards necessary to accelerate the development 
        of the technological tools to examine the functions and 
        outputs of generative adversarial networks or their 
        technologies that synthesize or manipulate content.
   Report to Congress on the feasibility of utilizing 
        public-private research partnerships to detect 
        manipulated or synthesized content.

                          LEGISLATIVE HISTORY

    S. 2904, the Identifying Outputs of Generative Adversarial 
Networks Act, was introduced on November 20, 2019, by Senator 
Cortez Masto (for herself and Senator Moran) and was referred 
to the Committee on Commerce, Science, and Transportation of 
the Senate. On May 20, 2020, the Committee met in open 
Executive Session and, by voice vote, ordered S. 2904 reported 
favorably with an amendment (in the nature of a substitute), 
with a first degree amendment.
    H.R. 4355, the Identifying Outputs of Generative 
Adversarial Networks Act, was introduced on September 17, 2019, 
by Representative Gonzalez [R-OH-16] (for himself and 
Representatives Stevens [D-MI-11], Baird [R-IN-4], and Hill [D-
CA-25]) and was referred to the Committee on Science, Space, 
and Technology of the House of Representatives. On December 9, 
2019, H.R. 4355, as amended, was passed by voice vote in the 
House of Representatives and was referred to the Committee on 
Commerce, Science, and Transportation of the Senate.

Hearings

    On June 13, 2019, the Permanent Select Committee on 
Intelligence of the House of Representatives held a hearing 
entitled ``The National Security Challenge of Artificial 
Intelligence, Manipulated Media, and `Deepfakes'''.\10\ This 
hearing specifically examined deepfakes and other types of AI 
generated synthetic data.
---------------------------------------------------------------------------
    \10\U.S. Congress, House Permanent Select Committee on 
Intelligence, The National Security Challenge of Artificial 
Intelligence, Manipulated Media, and ``Deepfakes,'' 116th Cong., 1st 
sess., Jun. 7, 2019, press release and webcast (https://
intelligence.house.gov/news/documentsingle.aspx?DocumentID=657) 
(accessed Sep. 3, 2020).
---------------------------------------------------------------------------
    On January 15, 2020, the Committee on Commerce, Science, 
and Transportation of the Senate held a hearing entitled 
``Industries of the Future.'' This hearing included an 
examination of the opportunities and issues associated with the 
development of increasingly sophisticated artificial 
intelligence capabilities, including deepfakes.\11\
---------------------------------------------------------------------------
    \11\U.S. Congress, Senate Committee on Commerce, Science, and 
Transportation, Industries of the Future, 116th Cong., 1st sess., Jan. 
15, 2020, press release and webcast (https://www.commerce.senate.gov/
2020/1/industries-of-the-future) (accessed Sep. 3, 2020).
---------------------------------------------------------------------------

                            ESTIMATED COSTS

    In accordance with paragraph 11(a) of rule XXVI of the 
Standing Rules of the Senate and section 403 of the 
Congressional Budget Act of 1974, the Committee provides the 
following cost estimate, prepared by the Congressional Budget 
Office:

	[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]


    S. 2904 would require the National Science Foundation (NSF) 
to support research on manipulated digital content and 
information authenticity. The bill also would direct the 
National Institute of Standards and Technology (NIST) to create 
measurements and standards for the development of technological 
tools that examine generative adversarial networks (GANs), 
which are used to produce manipulated content.
    For this estimate, CBO assumes that the legislation will be 
enacted in late 2020. Under that assumption, the affected 
agencies could incur some costs in 2020, but CBO expects that 
most of the costs would be incurred in 2021 and later.
    Using information from the NSF, CBO estimates that 
implementing the bill would have no significant cost for the 
NSF because the agency is already carrying out the required 
activities through its existing grant programs. Using 
information from NIST, CBO estimates that the agency would 
require 10 additional employees at an average annual cost of 
$175,000 each through 2023 to establish a research program on 
GANs and similar technologies. S. 2904 also would direct NIST 
and the NSF to report to the Congress on related policy 
recommendations. Based on the costs of similar tasks, CBO 
estimates that developing the report would cost less than 
$500,000. In total, CBO estimates that implementing S. 2904 
would cost $6 million over the 2020-2025 period; such spending 
would be subject to the availability of appropriated funds.
    On October 29, 2019, CBO transmitted a cost estimate for 
H.R. 4355, the Identifying Outputs of Generative Adversarial 
Networks Act, as ordered reported by the House Committee on 
Science, Space, and Technology on September 25, 2019. The two 
pieces of legislation are similar; the differences in CBO's 
estimated costs in 2020 reflect different assumed dates of 
enactment.
    The CBO staff contacts for this estimate are Janani 
Shankaran and David Hughes. The estimate was reviewed by H. 
Samuel Papenfuss, Deputy Director of Budget Analysis.

                      REGULATORY IMPACT STATEMENT

    In accordance with paragraph 11(b) of rule XXVI of the 
Standing Rules of the Senate, the Committee provides the 
following evaluation of the regulatory impact of the 
legislation, as reported:

Number of Persons Covered

    S. 2904, as amended, would cover the NSF, NIST, DARPA, 
Intelligence Advanced Research Projects Activity (IARPA), other 
relevant Federal agencies, Congress, and public and private 
academic and scientific stakeholders on forensic science and 
generative adversarial networks.

Economic Impact

    S. 2904, as amended, would have no negative expected 
impacts on the scientific community. Rather, S. 2904 would 
assist in accelerating the development of technological tools 
to examine the outputs of generative adversarial networks and 
encourage collaboration across public and private sectors.

Privacy

    S. 2904, as amended, would have no further impact on 
privacy.

Paperwork

    S. 2904, as amended, in section V would require the 
Directors of NSF and NIST to submit a joint report to Congress.

                   CONGRESSIONALLY DIRECTED SPENDING

    In compliance with paragraph 4(b) of rule XLIV of the 
Standing Rules of the Senate, the Committee provides that no 
provisions contained in the bill, as reported, meet the 
definition of congressionally directed spending items under the 
rule.

                      SECTION-BY-SECTION ANALYSIS

Section 1. Short title.

    This section would provide that the bill may be cited as 
the ``Identifying Outputs of Generative Adversarial Networks 
Act'' or the ``IOGAN Act''.

Section 2. Findings.

    This section would establish the current state of affairs 
regarding artificial intelligence and generative adversarial 
networks, the current work conducted by NSF, and the potential 
for the development of new credible techniques.

Section 3. NSF support of research on manipulated or synthesized 
        content and information security.

    This section would require the Director of NSF, along with 
other Federal agencies, to support merit-reviewed research on 
manipulated or synthesized content and authenticity, which may 
include fundamental research on authenticity and detection 
technologies, identification technical tools, social and 
behavioral research, public perception and awareness research, 
and research awards coordinated with other Federal agencies.

Section 4. NIST support for research and standards on generative 
        adversarial networks.

    This section would require the Director of NIST to support 
research for the development of measurements and standards to 
examine the function and outputs of GAN or other manipulative 
technologies. The Director of NIST would be required to receive 
input from public, private, and academic stakeholders and 
consider the feasibility of ongoing engagement in the 
development of standards and measurements.

Section 5. Report on feasibility of public-private partnership to 
        detect manipulated or synthesized content.

    This section would require the Directors of NSF and NIST to 
submit a joint report to Congress, not later than 1 year after 
the date of enactment, detailing the feasibility for research 
opportunities with the private sector and any policy 
recommendations to facilitate and improve communication and 
coordination among the private sector, NSF, and other Federal 
agencies with respect to generative adversarial networks or 
other synthesizing and manipulative technologies.

Section 6. Generative adversarial network defined.

    This section would define the term ``generative adversarial 
network'' for the purposes of this bill.

                        CHANGES IN EXISTING LAW

    In compliance with paragraph 12 of rule XXVI of the 
Standing Rules of the Senate, the Committee states that the 
bill as reported would make no change to existing law.

                                  [all]