H.R. 4355 — what changed
Identifying Outputs of Generative Adversarial Networks Act
From Introduced in House to Reported in House. 5 sections amended between Introduced in House and Reported in House.
Sec. 2 Findings
Congress finds the following:
changed
Research gaps currently exist on the underlying technology needed to develop tools to identify authentic videos, voice reproduction, or photos from manipulated or synthesized content, including those generated by generative adversarial networks.
The National Science Foundation’s focus to support research in artificial intelligence through computer and information science and engineering, cognitive science and psychology, economics and game theory, control theory, linguistics, mathematics, and philosophy, is building a better understanding of how new technologies are shaping the society and economy of the United States.
The National Science Foundation has identified the “10 Big Ideas for NSF Future Investment” including “Harnessing the Data Revolution” and the “Future of Work at the Human-Technology Frontier”, in with artificial intelligence is a critical component.
The outputs generated by generative adversarial networks should be included under the umbrella of research described in paragraph (3) given the grave national security and societal impact potential of such networks.
Generative adversarial networks are not likely to be utilized as the sole technique of artificial intelligence or machine learning capable of creating credible deepfakes and other comparable techniques may be developed in the future to produce similar outputs.
Sec. 3 NSF support of research on manipulated or synthesized content and information security
changed
The Director of the National Science Foundation, in consultation with other relevant Federal agencies, shall support merit-reviewed and competitively awarded research on the science manipulated or synthesized content and ethics of material produced by generative adversarial networks, information authenticity, which may include—
changed
supplementing fundamental research on digital media forensic tools or comparable other technologies for detection of verifying the outputs authenticity of information and detection of manipulated or synthesized content, including content generated by generative adversarial networks completed by the Defense Advanced Research Projects Agency and the Intelligence Advanced Research Projects Activity;networks;
changed
fundamental research on developing constraint aware generative adversarial networks; andtechnical tools for identifying manipulated or synthesized content, such as watermarking systems for generated media;
changed
social and behavioral research on related to manipulated or synthesized content, including the ethics of the technology, technology and human engagement with the networks.content;
added
research on public understanding and awareness of manipulated and synthesized content, including research on best practices for educating the public to discern authenticity of digital content; and
added
research awards coordinated with other federal agencies and programs including the Networking and Information Technology Research and Development Program, the Defense Advanced Research Projects Agency and the Intelligence Advanced Research Projects Agency.
Sec. 4 NIST support for research and standards on generative adversarial networks
changed
In general— The Director of the National Institute of Standards and Technology shall support research for the development of measurements and standards necessary to accelerate the development of the technological tools to examine the function and outputs of generative adversarial networks.networks or other technologies that synthesize or manipulate content.
Outreach— The Director of the National Institute of Standards and Technology shall conduct outreach—
changed
to receive input from private, public, and academic stakeholders on fundamental measurements and standards research necessary to examine the function and outputs of generative of generative adversarial networks or to develop constraint aware generative adversarial networks; and
changed
to consider the feasibility of an ongoing public and private sector engagement to develop voluntary standards for the function and outputs of generative adversarial networks or comparable technologies.other technologies that synthesize or manipulate content.
Sec. 5 Report on feasibility of public-private partnership to detect manipulated or synthesized content
changed
Not later than one year after the date of the enactment of this Act, the Director of the National Science Foundation and the Director of the National Institute of Standards and Technology shall jointly submit to the Committee on Science, Space, Science, and Technology of the House of Representatives and the Committee on Commerce, Science, and Transportation a report containing—
changed
the Directors’ findings with respect to the feasibility for research opportunities with the private sector, including digital media companies to detect the function and outputs of generative adversarial networks or comparable technologies; other technologies that synthesize or manipulate content; and
changed
any policy recommendations of the Directors that could facilitate and improve communication and coordination between the private sector, the National Science Foundation, and relevant Federal agencies through the implementation of innovative approaches to detect media products digital content produced by generative adversarial networks or comparable technologies.other technologies that synthesize or manipulate content.
Sec. 6 Generative adversarial network defined
changed
In this Act:Act, the term “generative adversarial network” means, with respect to artificial intelligence, the machine learning process of attempting to cause a generator artificial neural network (referred to in this paragraph as the “generator” and a discriminator artificial neural network (referred to in this paragraph as a “discriminator”) to compete against each other to become more accurate in their function and outputs, through which the generator and discriminator create a feedback loop, causing the generator to produce increasingly higher-quality artificial outputs and the discriminator to increasingly improve in detecting such artificial outputs.
removed
Generative adversarial network— The term generative adversarial network means, with respect to artificial intelligence, the machine learning process of attempting to cause a generator artificial neural network (referred to in this paragraph as the “generator” and a discriminator artificial neural network (referred to in this paragraph as a “discriminator”)) to compete against each other to become more accurate in their function and outputs, through which the generator and discriminator create a feedback loop, causing the generator to produce increasingly higher-quality artificial outputs and the discriminator to increasingly improve in detecting such artificial outputs.
removed
Comparable technology— The term comparable technology means technology that utilizes similar techniques to achieve the same outputs as a generative adversarial network.
removed
Constraint aware— The term constraint aware means, with respect to artificial intelligence, the generation of realistic relational data by a machine with constraint on the modules generated by an adversarial network.