[House Hearing, 118 Congress]
[From the U.S. Government Publishing Office]
ARTIFICIAL INTELLIGENCE AND INTELLECTUAL
PROPERTY: PART III_IP PROTECTION FOR AI
ASSISTED INVENTIONS AND CREATIVE WORKS
=======================================================================
HEARING
BEFORE THE
SUBCOMMITTEE ON COURTS, INTELLECTUAL
PROPERTY, AND THE INTERNET
OF THE
COMMITTEE ON THE JUDICIARY
U.S. HOUSE OF REPRESENTATIVES
ONE HUNDRED EIGHTEENTH CONGRESS
SECOND SESSION
__________
WEDNESDAY, APRIL 10, 2024
__________
Serial No. 118-69
__________
Printed for the use of the Committee on the Judiciary
[GRAPHIC NOT AVAILABLE IN TIFF FORMAT]
Available via: http://judiciary.house.gov
__________
U.S. GOVERNMENT PUBLISHING OFFICE
55-378 WASHINGTON : 2024
-----------------------------------------------------------------------------------
COMMITTEE ON THE JUDICIARY
JIM JORDAN, Ohio, Chair
DARRELL ISSA, California JERROLD NADLER, New York, Ranking
MATT GAETZ, Florida Member
ANDY BIGGS, Arizona ZOE LOFGREN, California
TOM McCLINTOCK, California SHEILA JACKSON LEE, Texas
TOM TIFFANY, Wisconsin STEVE COHEN, Tennessee
THOMAS MASSIE, Kentucky HENRY C. ``HANK'' JOHNSON, Jr.,
CHIP ROY, Texas Georgia
DAN BISHOP, North Carolina ADAM SCHIFF, California
VICTORIA SPARTZ, Indiana J. LUIS CORREA, California
SCOTT FITZGERALD, Wisconsin ERIC SWALWELL, California
CLIFF BENTZ, Oregon TED LIEU, California
BEN CLINE, Virginia PRAMILA JAYAPAL, Washington
KELLY ARMSTRONG, North Dakota MARY GAY SCANLON, Pennsylvania
LANCE GOODEN, Texas JOE NEGUSE, Colorado
JEFF VAN DREW, New Jersey LUCY McBATH, Georgia
TROY NEHLS, Texas MADELEINE DEAN, Pennsylvania
BARRY MOORE, Alabama VERONICA ESCOBAR, Texas
KEVIN KILEY, California DEBORAH ROSS, North Carolina
HARRIET HAGEMAN, Wyoming CORI BUSH, Missouri
NATHANIEL MORAN, Texas GLENN IVEY, Maryland
LAUREL LEE, Florida BECCA BALINT, Vermont
WESLEY HUNT, Texas
RUSSELL FRY, South Carolina
Vacancy
------
SUBCOMMITTEE ON COURTS, INTELLECTUAL PROPERTY, AND
THE INTERNET
DARRELL ISSA, California, Chair
THOMAS MASSIE, Kentucky HENRY C. ``HANK'' JOHNSON, Jr.,
SCOTT FITZGERALD, Wisconsin Georgia, Ranking Member
CLIFF BENTZ, Oregon TED LIEU, California
BEN CLINE, Virginia JOE NEGUSE, Colorado
LANCE GOODEN, Texas DEBORAH ROSS, North Carolina
KEVIN KILEY, California ADAM SCHIFF, California
NATHANIEL MORAN, Texas ZOE LOFGREN, California
LAUREL LEE, Florida MADELEINE DEAN, Pennsylvania
RUSSELL FRY, South Carolina GLENN IVEY, Maryland
CHRISTOPHER HIXON, Majority Staff Director
AARON HILLER, Minority Staff Director & Chief of Staff
C O N T E N T S
----------
Wednesday, April 10, 2024
OPENING STATEMENTS
Page
The Honorable Darrell Issa, Chair of the Subcommittee on Courts,
Intellectual Property, and the Internet from the State of
California..................................................... 1
The Honorable Henry C. ``Hank'' Johnson, Ranking Member of the
Subcommittee on Courts, Intellectual Property, and the Internet
from the State of Georgia...................................... 2
The Honorable Jerrold Nadler, Ranking Member of the Committee on
the Judiciary from the State of New York....................... 4
WITNESSES
Sandra Aistars, Clinical Professor, Antonin Scalia Law School,
George Mason University
Oral Testimony................................................. 6
Prepared Testimony............................................. 9
Kristelia Garcia, Anne Fleming Research Professor of Law,
Georgetown University Law Center
Oral Testimony................................................. 26
Prepared Testimony............................................. 28
Joshua Landau, Senior Counsel, Innovation Policy, Computer and
Communications Industry Association
Oral Testimony................................................. 37
Prepared Testimony............................................. 39
Claire Laporte, Intellectual Property Fellow, Former Head of
Intellectual Property, Ginkgo Bioworks, Inc.
Oral Testimony................................................. 48
Prepared Testimony............................................. 50
LETTERS, STATEMENTS, ETC. SUBMITTED FOR THE HEARING
All materials submitted by the Subcommittee on Courts,
Intellectual Property, and the Internet, for the record........ 76
A statement from the Motion Picture Association, Inc., Apr. 10,
2024, submitted by the Honorable Darrell Issa, Chair of the
Subcommittee on Courts, Intellectual Property, and the Internet
from the State of California, for the record
QUESTIONS AND RESPONSES FOR THE RECORD
Questions submitted by the Honorable Darrell Issa, Chair of the
Subcommittee on Courts, Intellectual Property, and the Internet
from the State of California, for the record
Questions and responses for Kristelia Garcia, Anne Fleming
Research Professor of Law, Georgetown University Law Center
Questions and responses for Joshua Landau, Senior Counsel,
Innovation Policy, Computer and Communications Industry
Association
Questions and responses for Sandra Aistars, Clinical Professor,
Antonin Scalia Law School, George Mason University
Questions and responses for Claire Laporte, Intellectual
Property Fellow, Former Head of Intellectual Property,
Ginkgo Bioworks, Inc.
ARTIFICIAL INTELLIGENCE AND
INTELLECTUAL PROPERTY: PART III--
IP PROTECTION FOR AI ASSISTED INVENTIONS AND CREATIVE WORKS
----------
Wednesday, April 10, 2024
House of Representatives
Subcommittee on Courts, Intellectual Property, and
the Internet
Committee on the Judiciary
Washington, DC
The Committee met, pursuant to notice, at 10:12 a.m., in
Room 2141, Rayburn House Office Building, the Hon. Darrell Issa
[Chair of the Subcommittee] presiding.
Members present: Representatives Issa, Fitzgerald, Bentz,
Cline, Kiley, Lee, Johnson, Nadler, Lieu, Ross, Schiff,
Lofgren, and Ivey.
Also present: Representative Jackson Lee.
Mr. Issa. I am pleased to announce that for our purposes
the quorum is present. The Subcommittee will come to order.
Without objection, the Chair is authorized to declare a recess
at any time. We welcome everyone here today to the hearing on
Intellectual Property Protections for AI-Assisted Inventions
and Creative Works.
Most importantly, I will now recognize myself to recap
where we are on AI and intellectual property. I believe the
subcommittee, on a very broad basis, has been at the forefront
of Congress' examination of intellectual property and
artificial intelligence. Today's hearing titled again
``Artificial Intelligence and Intellectual Property Part III,''
shows the fact that we have, in fact, looked at this from three
different ways, but perhaps 30 or 40 different ways when you
look at the nuances that have been covered.
To cover just a few in our inquiry, recognizing that our
role in AI is different than other Committees. Many think that
this is about protecting against public disclosure. Although
the Constitution is within our jurisdiction, this Committee and
this Subcommittee will particularly, and has focused on Article
1, Clause 8 of the Constitution. A requirement to promote, a
requirement to reward so that we, in fact, continue to have
useful arts and inventions created. I make that point because,
in fact, one of the greatest threats that we have found in
earlier hearings is that if, in fact, we grant no protection
for patents, trademarks, copyrights that are somehow produced
by AI, then, in fact, could we find ourselves dismantling a
system that has made the United States the most innovative and
successful country in the history of mankind.
Today, we will continue, without a doubt, to explore AI in
a context of IP theft related to China. There are others who
steal IP, but nobody who does it as well as China. Very simply,
today, how we define authorship when AI is part of the creative
process? We particularly know that the Copyright Office has
made a bold statement that 100 percent AI generated will not be
copyrightable. In our research, we have discovered that, in
fact, there is no such thing at this time as 100 percent AI
generated. To generate a copyrighted work, there has to be an
input of what you want, an input of how you want it. The
Copyright Office has not opined on this, but we will today ask
you and others to opine on exactly that, where the line should
be, how close it should be.
I want to particularly draw attention in my opening
statement to an example of one of my heroes, one of the most
iconic piano players and, in fact, the Piano Man, Billy Joel,
who recently produced and is just beginning to make the public
aware, an AI generated music video, his music and his likeness.
In fact, it is him at three different ages, but none of the
characters underlying, none of the motions were, in fact, his,
but rather actors who made movements and then with AI, that was
hovered with Billy Joel very young, Billy Joel not so young,
and Billy Joel, still younger than he is today, but not young
at all. The fact is we have creative geniuses amidst us in song
and other arts who no longer have the ability on their own,
they may not have the vocal range, they may not, in fact, be
fit to tour and yet, they can still create. That is one of the
challenges we will face today.
I will say, without a doubt, this Committee intends to be
forward leaning on providing intellectual property protection
for these creations and finding a way to do it will not, in
fact, turning everything over to the machine and thus
denigrating or reducing the effectiveness of what you might
call old-fashioned creation. For that reason, the questions
will, in fact, be asking all of you to help us understand how
far we can go to continue to recognize that creations thought
of by man, assisted by machine, in fact, need to be protected.
With that, it is my pleasure to recognize my Ranking
Member, the gentleman from Georgia.
Mr. Johnson. Thank you, Mr. Chair, for holding this hearing
and thank you to the witnesses for appearing today. As most of
the people in this room already know, this is Part III in a
series of hearings on the impact of artificial intelligence on
intellectual property and innovation. I am proud to join my
Chair, Mr. Issa, in this on-going examination of the State of
innovation in America.
Artificial intelligence is changing how the world works
right before our very eyes and this Subcommittee has
considerations not just over the IP implications of the final
outputs of AI models, but also how these models are trained.
While some of the issues we discuss here today exist only as
hypotheticals for future iterations of generative AI
technologies, other issues are already real and before us,
impacting the way we do business and live our lives today. I am
looking forward to continuing our conversation in Part IV,
whatever the topic might be.
When we sit down behind the wheel of our car, lean over and
turn on the radio, we expect the song we hear to be a work
created by a human artist. The beat, the melody, the words, and
the riffs, we take for granted as having sprung from the mind
of a flesh and blood creator. What happens when that is no
longer the case? The question we will examine today is when a
work of art or original invention should no longer be worthy of
intellectual property protection. Generative AI can both
produce original content and assist the production of original
content that would clearly be eligible for IP protections if
created entirely by a human. An algorithm can spin out original
lyrics for a songwriter to put to music and it can analyze
large sets of data for a scientist developing a new chemical
compound. It is not debatable that these basic operations are
already reality and already we are beginning to see artists and
scientists seeking protection of such creations through
copyright and patent.
AI cannot be listed as an inventor on a patent. It cannot
own a copyright. The courts and the Copyright Office have,
respectively, already addressed those basic questions. The
question of how much AI involvement renders a product as no
longer having been created by a human being remains open. Both
the Copyright Office and the Patent and Trademark Office have
released guidance and are continuing to address the issue of
how we should, as a society, treat AI created or AI assisted
works of innovation.
I applaud President Biden, Director Vidal, and Director
Perlmutter for addressing these questions head on. From the
initial guidance it is clear that the question of how much AI
is too much for IP protection is far from a straight-forward
question. Yes, generative AI can write an essay, but the
nuances between AI assistance and 100 percent AI creation are
substantial. Moreover, innovations made with the assistance of
AI are the final product in a long chain of creators. AI is
typically developed by one set of people trained on the data of
many others which often includes others' copyrighted property
and then is deployed by a third set who become the model's
users. That model then turns around to be used in lucrative
ways like finding aberrations in data sets or making a better
autotune for a song.
I am looking forward to hearing from our witnesses how they
believe we should address these nebulous questions about the
inseparable nature of human and machine creativity and whether
existing copyright and patent laws provide any guidance. While
those conversations are ongoing, it is also time for Congress
to examine what our involvement should be and I am looking
forward to hearing from our witnesses what legislative tools
they believe agencies, inventors, and creators need to keep our
intellectual property system strong. We must remember that at
its heart, intellectual property protections exist to keep our
communities strong. Looking at my home district near Atlanta,
Georgia, patents drive the creation of new business and
emerging community leaders. Copyright protections protect up
and coming new artists and allow creators to make a living off
of their talents, contributing to the fabric, depth, and color
of the towns we call home, and both drive the growth of vibrant
cities. In fact, the U.S. Census Bureau last month named the
Metropolitan Atlanta area the sixth largest and third fastest-
growing region in the country. Because of this, we should
approach the question of IP protection for AI-driven works both
from a legal jurisprudence perspective, but also from a human
one by asking at every step what helps creators? We are more
likely to end up with a system that keeps working the way it
should no matter what revolutionary new technologies come our
way.
I look forward to hearing from our witnesses and I yield
back the balance of my time. Thank you.
Mr. Issa. I thank the gentleman. We now recognize the
Ranking Member of the Full Committee, Mr. Nadler, for his
opening statement.
Mr. Nadler. I would like to begin by thanking Chair Issa
and Ranking Member Johnson for holding this bipartisan hearing
to examine the scope of intellectual property protection for
artificial intelligence generated in AI-assisted works.
Generative AI may be in its nascency, but many models are
already capable of creating works that if created by a human
would be eligible for copyright protection. Inventors,
similarly, are using AI to aid in their discoveries. While AI
that is capable of creating an invention eligible for patent
protection under Section 101 of the Patent Act may, debatably,
not exist just yet, the involvement of generative AI in
innovative processes still raises the question of how much AI
involvement takes away the human element of discovery.
The question presented at this hearing is narrow at first
glance, simply how much human needs to be involved for a
creation to warrant intellectual property protection? When we
discuss generative AI, it is often in terms of broader
controversies, including ingestion, practical application, and
the replication of human works and styles. Discussions like the
ones this Subcommittee is embarking on today can launch us in
entirely unique and unexplored directions. How much AI is too
much to render something a human creation? Whether some uses of
generative AI are more acceptable as tolls than others? How we
could even tell a final product was created by machine rather
than by a person are all valid questions that ensue from this
narrow starting point.
While the technology may be new, the philosophical
questions presented here today are not. In 1637, Rene Descartes
debuted the first principle of his Principles of Philosophy,
cogito ergo sum. Therefore I am. Over 300 years later in 1950,
mathematician Alan Turing established what became known as the
Turing Test, to determine whether or not a computer is capable
of thinking like a human being.
The ability of AI to think and create has been examined
repeatedly throughout science fiction. ``Star Trek: The Next
Generation'' even examined the essential question of the rights
of highly advanced AI character called Data in its Season 2
episode Measure of a Man. While it did not, unfortunately,
delve into questions of IP protection with Data's creative
works, it is emblematic of a century's long interest in the
possibility of artistry, intelligence, and innovation by
nonhuman entities.
Many of the philosophical questions presented by AI in
science fiction are far enough in the future that they are not
concerns before the Judiciary Committee. The practical
questions of whether or not to extent intellectual property
protections to AI-assisted works is very much a question for
here and now. The Copyright Office and the Patent and Trademark
Office have begun addressing the question through draft
guidance to create basic rules of the road to artists and
creators seeking to protect works developed with the use of
generative AI.
Just over a year ago, the Copyright Office issued the first
guidance governing intellectual property claims for works
developed with the assistance of AI. The Office informed
artists that in these cases, copyright protection will only be
available for the part of a final product created by humans,
and it asks that the creators identify which parts of each work
of authorship were developed using AI. President Biden, in last
fall's Executive Order on AI, directed the USPTO to develop
guidance as well for governing the use of AI in claimed
inventions before the Patent Office. Just two months ago, the
PTO issued and initial, ``inventorship guidance and examples
for AI-assisted inventions.'' Like the Copyright Office, the
question under the PTO guideline is not whether or not AI was
used in the creation or invention, but how much human
innovation and involvement there was in the creation of the
invention.
I applaud the Copyright and Patent Offices for taking much
needed first steps to clarify what should and should not
receive intellectual property protection as generative AI
becomes increasingly and indelibly incorporated in our society
and how we creative. Patents and copyrights are two very
different types of intellectual property, and it is important
that we acknowledge this reality as we continue conversations
about how AI can and should be used in their development.
There is no such thing as a one-size-fits-all approach to
innovation and we would be remiss if we failed to recognize
this. Many questions remain unexplored in both the patent and
copyright realms and I am looking forward to hearing from our
witnesses about what they think is missing from the guidance
and from these conversations overall. For example, the guidance
at both agencies leaves open the question of how to maintain
candor and honesty in applications for IP protection and I am
particularly looking forward to hearing from our witnesses
regarding how we can enforce the AI rules we create.
AI undoubtedly will have enormous impact on our IP system,
and it is imperative that we, as the Judiciary Committee,
consider how best to ensure that our innovation protections
remain strong even as technologies change.
I look forward to hearing from our witnesses and I yield
back the balance of my time.
Mr. Issa. I thank the Ranking Member. All other Members
will have the ability to have their opening statements placed
in the record.
It is now my pleasure to introduce our distinguished panel
of witnesses.
Starting with Ms. Sandra Aistars, who is a Clinical
Professor at George Mason University, the Antonin Scalia Law
School, where she leads the school's Arts and Entertainment
Program. Professor Aistars is also a Senior Fellow for
Copyright Research and Policy and a Senior Scholar at the
Center for Intellectual Property and Innovation Policy. Prior
to joining the Scalia Law School, she was the Chief Executive
Officer of the Copyright Alliance, a nonprofit, public interest
organization that advocates on behalf of artists and creators
and I had the opportunity to see her in that role.
Next, we have Ms. Garcia. She is the Anne Fleming Research
Professor at the Georgetown University School of Law. Professor
Garcia's work focuses on intellectual property law through the
lens of law and economics. She previously taught at the
University of Colorado Law School, was a Fellow at George
Washington University Law School, and spent nearly a decade
working in the music industry. Welcome.
We next have Mr. Joshua Landau. Mr. Landau is a Senior
Counsel for Innovation Policy at the Computer and
Communications Industry Association where he focuses on patent
issues. He previously worked in private practice at WilmerHale
where he represented clients in patent litigation, counseling,
and prosecution in proceedings before the District Courts and
the PTAB.
Last, we have Ms. Claire Laporte. Ms. Laporte is the
Intellectual Property Fellow at the Ginkgo Bioworks, a
biotechnology company focusing on cell programming. She
previously served as the company's head of intellectual
property and design and IP strategy and oversaw its
implementation. Before joining, Ms. Laporte was a trial
attorney in a private practice where she focused on patent and
trademark secret cases.
We welcome all our witnesses and pursuant to the
Committee's rules would you please rise to take an oath? Raise
your right hand if you would. Do you solemnly swear or affirm
under penalty of perjury that the testimony you are about to
give is true and correct to the best of your knowledge,
information, and belief so help you God? Please be seated. Let
the record reflect that all witnesses answered in the
affirmative.
If you haven't seen this a thousand times on CSPAN and many
of you have testified before, let it be said that every part of
collateral material, including your full written copy of your
oral statement, will be placed in the record. So, in that case,
we would ask that you do the best you can to remain within the
five-minutes so that we can get to the questions that we all
have because I know today you have answers for our questions.
So, with that, we go to Professor Aistars.
STATEMENT OF SANDRA AISTARS
Ms. Aistars. Thank you, Chair Issa, Ranking Member Johnson,
Ranking Member Nadler, Members of the Subcommittee, thank you
for the opportunity to testify. I do not represent anyone on
these issues, and I am appearing here on my own behalf. My
recent scholarship is focused primarily on visual arts, but my
comments this morning are applicable to copyright more broadly
as well.
First, I believe that when determining whether to register
claims of copyrights or creative works made by humans with the
use of generative AI or GAI for short, we should look to the
acts of the human, not at the outputs of the GAI. If a human
has met the minimal creativity requirement set out by the
Supreme Court in Feist, the question is whether the interaction
with the GAI is undermining a claim to human authorship for
extending authentic human vision for a work. Applying this
approach, at least some creative works authored by humans using
the assistance of GAI will be protectable by copyright.
Second, liability for training of GAI must be resolved, but
denying IP protection to otherwise protectable works that
humans who use GAIs is counterproductive. It will relegate
legitimate human works to the category of synthetic data.
Denying copyright to humans who create works with the
assistance of GAIs if the work is otherwise sufficiently
original to qualify for protection will do nothing to instill
respect for IP rights and those who develop and train GAIs. To
the contrary, doing so will merely disenfranchise creative
workers from being able to claim copyright in expressive works
based on the media or tools they choose to work with,
relegating their works to the category of synthetic data and
foreclosing to these human artists the opportunity to control
or be compensated for use of their works.
Third, turning otherwise protectable GAI assistive creative
works into synthetic data exploits human creators. Creating a
class of unprotectable synthetic works would be a windfall to
GAI companies because such works would immediately become
available for GAI training without the need for any permissions
from the human creator who used the GAI to make the creative
work. This exploits creative workers on both the input side by
not protecting copyright in the initial materials the GAI is
trained on and on the output side by not protecting copyright
and expressive works created by humans using the GAI in their
authorship.
My main point is that copyright law should continue to
function in a technology-neutral fashion. The minimalist spark
of creativity test for originality from Feist can readily be
adapted to fix the GAI authorial workflow. The proper inquiry
in the GAI context is whether the human author has used the GAI
as an artist uses any tool or material in their artmaking
practice. Has the artist deployed GAI or engaged it
authentically and in their own voice
in a manner that demonstrates the artist is being true to their
creative vision. I described this more fully in my written
testimony.
The author should not be asked to dissect creative works
element by element by element to prove control or
foreseeability over the precise operation of GAI tools. The
Copyright Office currently requires authors to disclose and
disclaim more than de minimis use of GAI in creative works.
This approach causes problems for creators. The Copyright
Office guidelines do not marry well with the artmaking
practices of visual artists and other creators because they
require the creator to exercise control and foreseeability over
a device that they have not manufactured, trained, or deployed
into the marketplace. Instead, the relevant inquiry should be,
as I have described in my testimony, whether the creator is
executing their own intellectual conception. Are they engaging
the GAI authentically and in their own voice? Are they staying
true to their own creative vision?
I thank you for your attention and I look forward to your
questions this morning.
[The prepared statement of Ms. Aistars follows:]
[GRAPHICS NOT AVAILABLE IN TIFF FORMAT]
Mr. Issa. Thank you. Ms. Garcia.
STATEMENT OF KRISTELIA GARCIA
Ms. Garcia. Thank you, good morning, Chair Issa, Ranking
Member Johnson, Ranking Member Nadler, and the Members of the
Subcommittee for inviting me to testify today about the
copyrightability of works that are generated in whole or in
part by artificial intelligence. My name is Kristelia Garcia. I
am a law Professor at Georgetown where I teach and research on
issues relating to copyright law.
In my written testimony, I make three primary assertions.
First, is that works wholly or substantially generated by AI do
not merit copyright protection. Second, works only partially
generated by AI may merit copyright protection. Third, most
relevant for your purposes, that there is no call for
legislative action at this time.
I would like to use my time today to present a couple of
examples that illustrate and support these determinations. My
first example goes to works wholly or substantially generated
by AI. So, let's say that I direct ChatGPT with a simple prompt
to write a Congressional testimony and I don't give it anything
else. I don't give it a topic or the points that I would like
to make, any conclusions to reach. The resulting work then can
hardly be said to have been authored by me in the sense that I
couldn't possibly predict, nor control, what the AI is going to
write, nor could the resulting writing be said to be the
product of my original intellectual conception. I can also tell
you from disappointing personal experience that it is not
especially good at it either.
Now, the work that would result from my getting a simple
prompt to write Congressional testimony doesn't merit copyright
protection for a number of reasons. First and foremost, because
AI is not human, it doesn't meet the statutory definition of an
author. You may recall hearing about that case a few years back
of the Macaque monkeys who were denied a copyright in their
selfies. We have the same situation here. No human, no
copyright. We should recall further that the photographer in
that case, himself a human, was also denied a copyright in the
monkeys' photos, and this is because despite befriending the
monkeys and giving them a camera, the court found that the
photographer had not sufficiently contributed to the ultimate
images in that he could neither predict, nor importantly
control, and which did not then reflect his own original
intellectual conceptions.
Even if we could overcome this doctrinal improbability, why
would we? The AI in my example wasn't incentivized by copyright
protection to write the testimony. It was just responding to my
direction as it has been programmed to do. If we did
nonetheless grant a copyright here who would we grant it to?
The AI is not a legal person, so it can't enforce the
copyright, nor can it be enforced against. Giving the copyright
to the programmer of the AI is also ill advised since, among
other things, the program has already enjoyed copyright
protection on the code that operates the AI.
Could we instead give the copyright to me, the user who
made a simple prompt, also no, because I merely contributed an
idea and copyright doesn't protect ideas. It only protects
expression.
My second example goes to partially generated AI work. So,
what if instead of merely prompting an AI to write
Congressional testimony, I instead direct it with multiple
iterative prompts. I give it a topic and I give it the points
that I would like to make, conclusions I would like to reach,
tweak it back and forth, and give it a time limit. Then, I take
the resulting testimony, select a few sentences, delete the
rest and drop those sentences into a draft I have written up
separately. Now, this example involves a lot more human
intervention and contribution such that the resulting testimony
might be said to be imbued with my own original intellectual
conceptions.
First, in this case, I am likely to get, and I should get a
copyright because, I am a human. I can be incentivized by the
protections that copyright offers. Second, I have arguably made
sufficient interventions, contributions, and exercised
sufficient control over the project such that we can view the
resulting work as a product of my creativity and efforts.
So, I will circle back to where I began to conclude that I
see no call for legislative action at this time where the
copyrightability issues currently presented by AI-generated
work can be competently handled via a combination of statutory
interpretation and Federal litigation. Now, I can see that the
Copyright Office's disclose and disclaim approach in which a
human author may have to disclose the involvement in AI and
then disclaim portions of the work attributed to the AI
necessarily involve some tricky line drawing, but there is
nothing revolutionary or disruptive about a case-by-case
approach to copyright registration and we have only seen four
cases so far. The Copyright Office also received over 10,000
comments as a result to a call on these questions. In other
words, we are still very much in a learning phase, so we don't
know what type of legislative intervention, if any, would be
helpful or called for at this time.
I will also note that the courts have proven adept at
interpreting and applying copyright law in the context of so-
called disruptive or novel technologies, sometime shifting to
digitization of books to APIs. None of these technologies
necessitated a new categorical rule and AI doesn't either.
Thank you for your time and I look forward to your
questions.
[The prepared statement of Ms. Garcia follows:]
[GRAPHICS NOT AVAILABLE IN TIFF FORMAT]
Mr. Issa. I thank you.
I might note that we have a long history of wanting to
prevent excessive litigation. I know that may not be welcomed
by everyone on the bar, but that is another part of what we do.
So, that second part of what you said was most interesting,
that the courts could take care of it with litigation.
Mr. Landau?
STATEMENT OF JOSHUA LANDAU
Mr. Landau. Chair Issa, Ranking Member Johnson,
distinguished Members of the Subcommittee, thank you for
inviting me to testify today.
CCIA has been at the forefront of technology policy issues
for more than 50 years, and our members are at the forefront of
artificial intelligence today. They make the chips that AI runs
on. They develop leading-edge AI models, and they use AI to
solve problems for their customers.
It's critical that we protect American innovation while
preventing the abuse of intellectual property protections over
AI output. If we look to existing law, we can find the balance
that we need.
In 1966, the Register of Copyrights was facing a difficult
issue. Computer technology was becoming more widespread and
sophisticated. People were beginning to apply for copyright
protection for creative works made by computers. The Register
wrote,
It is certain that both the number of works proximately
produced or ``written'' by computers and the problems of the
Copyright Office in this area will increase.
Fortunately, the Register also had an answer--and it was a
good one. The Register identified the determinative question:
Whether the ``work'' is basically one of human authorship, with
the computer merely being an assisting instrument, or whether
the traditional elements of authorship in the work . . . were
actually conceived and executed not by man but by a machine.
The Register's approach 60 years ago remains the right
approach today. The same inquiry would apply to patentable
inventions--was the invention one conceived of by human
ingenuity with the AI operating as an assistive tool, or was
the invention generated exclusively by the AI system? When a
machine is responsible for the traditional elements of
creation--the original expression of an artistic work or the
generation of an invention--then intellectual property rights
are inappropriate. When the machine simply supports human
authorship or invention, then copyright or patent protection
should remain available.
With that guiding principle in mind, many questions
regarding AI output and IP protection become easier to answer.
Complete human creations remain protectable; completely AI
creations should not. There will be difficult questions in the
center of that continuum, where humans and AI collaborate to
create and invent, but the answers there lie in existing law--
in the law of inventor-ship and authorship--and in the
Register's focus on whether a human was truly the creative or
inventive entity.
Those difficult questions will require attention. For
example, if an inventor uses AI as a tool to develop a new
invention, when does it cross the line from collaboration to
the human simply taking the AI's work as their own? The
existing law of inventorship can guide us. Someone who provides
background information or explains the state-of-the-art isn't
an inventor, nor is someone who simply suggests a desired goal
and doesn't contribute to the solution of the problem.
The courts are suited well to developing these basic,
longstanding principles to address future uses of AI. Beyond
the availability of IP protection for AI output, the AI tools
will continue to impact other areas of the law. For example, in
patent law, the question of obviousness rests on whether a
person of ordinary skill with the relevant prior art available
could have created the invention. This is fundamentally similar
to how many AI systems operate.
The availability of AI as a tool to aid an invention, thus,
raises the level of ordinary skill in the art. If Albert
Einstein, me, and my six-year-old son could each ask an AI tool
for the answer to a problem and we all get the same answer,
it's hard to argue that any of us has invented anything that we
should be allowed to patent.
Outside of the legal system, there are potential policy
problems if we make AI output eligible for IP protection.
Already, more than half of U.S. patents are issued to foreign
inventors. These inventors can and do assert their patents
against American companies. The PTO was even forced to change
its trademark rules to deal with the flood of inaccurate and
potentially fraudulent trademark applications from China.
If AI output is patentable, a foreign adversary might
employ the same tactic--flooding the Patent Office with AI-
generated inventions. This will either overwhelm the office's
resources or result in even more cases of foreign entities
weaponizing patents against the American economy. In either
case, granting patents for AI output presents significant
concerns.
Finally, we should bear in mind that the constitutional
purpose of copyright and patent protection isn't to provide
economic rewards. It's to promote the progress of science and
useful arts. The rewards are a means to that end. AI doesn't
require the same economic incentives people may.
AI will, undoubtedly, become a part of the ordinary process
of invention and creation, but as the Register of Copyrights
recognized almost 60 years ago, and as the courts and agencies
have continued to recognize today, it's essential that we
ensure AI remains an adjunct to human creativity, not a
replacement for it. This approach will protect American
creativity and innovation, including in the field of AI.
I thank the Subcommittee for its time today and I look
forward to your questions.
[The prepared statement of Mr. Landau follows:]
[GRAPHICS NOT AVAILABLE IN TIFF FORMAT]
Mr. Issa. Thank you. Ms. Laporte?
STATEMENT OF CLAIRE LAPORTE
Ms. Laporte. Thank you Chair Issa, Ranking Member Johnson,
and the Members of the Subcommittee, I'm honored by your
invitation to speak today.
My name is Claire Laporte. I'm testifying today on behalf
of Ginkgo Bioworks, a biotechnology company in Boston. We're
publicly traded, and we work to use biology to solve problems
in industries ranging from agriculture to chemicals, to human
therapeutics, and we're heavily invested in developing and
using AI in our work.
I'm going to use my five minutes to make three basic
points.
The first point: Is quite uncontroversial. That is that
biotechnology is a critically important technology for our
national security and global competitiveness. That is one thing
that pretty much every country around the world agrees on. So,
I'm not going to dwell on that now.
The second point: There is no need to panic over the role
of AI, at least in the biotechnology space. We are far from
having anything that is solely AI-generated, and we don't know
what that would even look like in the future. I've got three
subpoints here.
The first one is that, in biology, AI is really just the
latest step in a gradual technological evolution. Biology has
been working hand-in-hand with computer science for decades,
and, in fact, we have a whole field that you can get a Ph.D. in
called computational biology that teaches you how to use these
kinds of tools that we've been working on for a long time.
The second subpoint in this part of my point is that, in
biology, you don't need to worry about AI developing things
that are fanciful or untrue, which I know is a concern in the
copyright space. A text-based AI system like ChatGPT can come
up with stuff that is crazy or false, but biology operates in
the real world. So, there is a real biophysical constraint on
the outputs.
The output of the AI will be used to make a physical thing,
which, then, has to be tested by a human--usually, that has
figured out how to test that thing. The results either do work
or they don't work. If they don't work, you discard them. If
they do work, that is a helpful development along what is
usually a multistep, very complex pathway in developing
something that is extremely difficult.
Third, AI has to be prompted by humans. We are very, very
far away from anything in which you could ask an AI to do
something like cure cancer, or whatever. That is an immense,
multistep problem. It goes off in many different directions.
There's absolutely no way that can happen.
In fact, solving engineering problems is extremely
difficult and humans have to be carefully trained in their
discipline to be able to direct AI tools, at least in biology,
to create anything worthwhile. Again, it always needs to be
tested to see if it works. Because if it's sort of fanciful or
ridiculous, it won't.
So, the takeaway here for my second point is that AI can be
transformative by helping us do research faster, cheaper, and
more predictably, but it is definitely not steering itself to
come up with biological inventions.
The third point: There is no need to complicate our already
complicated patent system by adding new inventorship
regulations, and by doing so, adding yet another complication
and step to our already overburdened litigation process.
The Patent Office has recently released new guidelines on
AI and inventorship that I think are likely to cause exactly
that kind of difficulty and complication. Under current U.S.
inventorship law, conception--having the inventive idea--is the
touchstone of inventorship. That can only be done by a human.
The AI is simply crunching through unimaginably large
datasets to come up with possible answers, based on things that
people have done before. The creativity is in the person
crafting the prompt and getting the AI to sort of row and
winnow through that mass of data in a particular direction.
The Patent Act expressly provides that ``patentability
shall not be negated by the manner in which the invention was
made,'' but the new regulations are doing essentially that. You
can have an invention that, if invented by old-school methods,
would be patentable, but if invented with the help of AI, might
not be. That is a completely nonsensical result.
Our patent system and our patent litigation system are both
ways too complicated as it is. I think that at this point, we
are not in a position where we should be adding more complexity
that potentially hurts our global competitiveness in comparison
to other countries that are not so concerned about these issues
of inventor-ship. So, my view is, if we've invented something
great using AI, let's celebrate that, not penalize it.
Thank you and I look forward to answering the questions.
[The prepared statement of Ms. Laporte follows:]
[GRAPHICS NOT AVAILABLE IN TIFF FORMAT]
Mr. Issa. I thank all our witnesses.
We now go to Mr. Fitzgerald for the first round of
questions.
Mr. Fitzgerald. Thank you, Mr. Chair.
One hundred and forty years ago, the Supreme Court, in
issuing its opinion on the copyrightability of the photograph
stated, quote,
The human intelligence can produce nothing without material
assistance. And whether it's a tool, a machine, or another's
hand, it is always the thought of the artist which directs the
instrument which guides and inspires the material means.
I think that opinion is kind of incredibly important for our
discussion today because, just as artists embrace the camera as
a creative tool, they have also begun to embrace the use of
generative AI. Just as human involvement is required in the
copyright of a photograph, so, too, must a human be involved,
obviously, in the use of AI as a creative tool.
I think what we are discussing here today is the amount of
involvement required, right, to grant a copyright. So, the
Copyright Office in its semiannual ruling in Zarya of the Dawn
denied copyright because of the generative AI tool used,
Midjourney. It did not allow the artist to exercise ultimate
creative control; instead, comparing it to a client who may
just hire an artist to possibly create an image for them.
Ms. Aistars, did the Copyright Office get it right in this
case? What should ultimate creative control look like? What do
you think about that?
Ms. Aistars. Thank you for the question.
I believe the relevant inquiry should be whether the work
is the artist's own intellectual conception, as you referred us
to the decision in the Cirone case, Bruegel v. Cirone,
emphasized that an artist engages with the tools that they use
and that's what we should be looking to.
What the Copyright Office is doing in asking artists to
look at their work and to pick the work apart, and to
demonstrate control over a tool and explain how much is being
used, rather than focus on the quality of the engagement, is
problematic to me.
I think the inquiry should be a qualitative test because
that is what Feist and Cirone are asking us to do already.
Those are the Supreme Court cases that we should be focusing
on. So, again, the relevant inquiry should be whether the work
is of the artist's own intellectual conception and at least
minimally creative, just as it always has been.
Mr. Fitzgerald. So, some will say that the Copyright Office
is more focused on real input of the artist versus the output
by the program itself. How do you find balance--
Ms. Aistars. So, I think they're actually doing the
opposite, actually.
Mr. Fitzgerald. Yes.
Ms. Aistars. They're looking at the--they're not looking at
the input of the artist. They're looking at the output of the
program. They're asking the artist to pick apart the end result
and say: Disclose and disclaim, where we see the AI tool making
contributions.
What I'm saying is let's talk about the artist's
intellectual conception. Let's talk about what the artist has
done in their creative process. What matters here, is the
artist's intellectual conception, the artist's process. Has the
artist made at least minimally creative contribution? What's
that spark of creativity that Feist asks us to look for? Where
is that from the artist?
Mr. Fitzgerald. Yes. A lot of times--it's been referred to
many times, but Jackson Pollock would create his paintings,
obviously, by some type of creative process where he would
approach the canvas in the way that he does, but the final
product is still copyrightable because the artist has exercised
this ``modicum of creativity,'' which is the standard under
copyright law.
Why do you think copyright be denied simply because of the
AI-generated output is random or unpredictable based on how the
artist creates?
Ms. Aistars. So, I don't think that there is a randomness
necessarily implied by using a GAI tool any more than there is
randomness implied by using any other tool or media by an
artist. I think we should look at what the artist's process is.
When an artist decides to create a work, an artist, typically,
has a preconception for a work, an idea in their mind. What do
they want to convey to the world?
They approach their work with that spirit in mind, and
then, they decide what tools are they going to pick up. Are
they going to pick up a paint brush? Are they going to pick up
the engagement with an AI, right? What is their creative
process going to look like, as they move through that
engagement, whatever that tool is?
I guess what I'm saying to the Committee is we should have
a sort of technologically neutral approach as to what tool the
artist is choosing. We should, instead, be asking: What is the
artist doing?
Mr. Fitzgerald. I'm out of time, but thank you.
I yield back, Chair.
Mr. Issa. I thank the gentlelady for her thorough answer to
that question.
[Laughter.]
We now go to the gentleman from Georgia, Mr. Johnson.
Mr. Johnson. Thank you, Mr. Chair.
Professor Garcia, because generative AI models are trained
on mass corpuses of human creators' works, AI outputs are
inextricably linked to human creativity. Can you walk us
through how pending litigation about the fair use doctrine
might or might not affect whether AI-generated outputs are
copyrightable?
Also, give us a little more color to your opinion that
legislative intervention is not really appropriate at this
particular time. That causes me to be a little queasy about
ceding legislative authority to the courts or to regulators.
Ms. Garcia. Thank you for the question.
I'll start with the first part of the question, which asked
what effect this pending litigation around whether or not the
training--let's call it data--is going to constitute copyright
infringement. I think that it could have important
ramifications for this question; namely, because if it turns
out that fair use does not apply, and that the use of
copyrighted works without authorization to train these large
language models constitutes copyright infringement, that will
mean that the products produced by those artificial
intelligence platforms would not be copyrightable, because we
don't copyright works that are infringed. Infringing works are
not copyrightable.
So, that would be significant, regardless of who we decided
we might want to assign the copyright to, including to the user
who just gives a prompt. So, I think that would be significant.
To clarify and talk a bit further about my suggestion that
we don't need legislative intervention at this time, I really
want to emphasize ``at this time,'' because I don't think we
know yet what we want.
I say this because we have these four generative-AI-related
registration determinations from the Copyright Office. Two of
them, only two of which deal with the more interesting question
of partially generated AI works, right, where we have this
disclaim, disclose and disclaim approach that we've been
mentioning.
I think that neither of those examples are quite the
example that I want. So, I would like to see more, so that we
get a better idea of sort of the things that Professor Aistars
was alluding to, which is: How do we determine, if we are going
to be looking at inputs as opposed to outputs, what is our
standard going to be?
I agree that it's going to necessarily be subjective and a
bit case-by-case. If we want legislative guidance around what
should be involved in those subjective analyses, I just think
we don't know yet what to ask for.
Mr. Johnson. Thank you.
Professor Aistars, do you have anything to add to what Ms.
Garcia has said, or anything that you want to comment on about
what she has said?
Ms. Aistars. So, if we were going to talk about training of
the AI tools and fair use applicability to those tools, we
would have a very lengthy conversation, and Chair Issa would
cut me off. So, I will restrain myself from answering those
questions.
I do think there are some very important questions that
need to be answered about the training of tools, and I,
personally, am very doubtful that the courts will be able to
find that the training of those tools is excusable with fair
use arguments.
At some point somehow, we will resolve how those tools are
being trained, and we will in some way somehow move on to the
conversation we're having today, which is: How do users use
those tools and what happens with the artwork, the creative
works, that are generated through the use of those tools by
users?
Professor Garcia and I are trying to have that conversation
sort of separating the training of the datasets issue from
that.
Mr. Johnson. OK. Thank you. Mr. Landau, 20 seconds.
Mr. Landau. Fundamentally, it comes back to the idea, was
the human essential creator of the artistic expression or was
it the computer? So, if the--
Mr. Johnson. How would you tell?
Mr. Landau. That's a good question. There's some difficulty
in telling, but it's the same kind of question we've asked
for--
Mr. Johnson. It's going to be the honor system?
Mr. Landau. No, I think it's the same kind of question
we've asked for decades at this point as to what portion of a
work is protectable and what isn't. I think of the Nike Jumpman
logo. There was litigation over that because somebody else
reproduced a similar photo, the idea of a basketball player
floating in the air, but it wasn't infringement because it was
their own expression of that idea. So, that expression is the
key. If you're just giving the AI the idea of ``Give me a
picture of a basketball player floating in the air,'' that's
not enough to generate a copyright for you.
Mr. Johnson. Thank you. I yield back.
Mr. Issa. Thank you.
I will note, by the way, that we have had a series, and
will have a series more, of roundtables. So, if I ask you to
stay as close as possible to that zero number, it's only
because we have members coming in and out for a short time
today. I suspect many of you will be invited back to those
roundtables in the near future.
We, then, go to the gentleman from Virginia, Mr. Cline.
Mr. Cline. Thank you, Mr. Chair.
I want to followup on the question of you, Mr. Johnson, and
particularly Ms. Garcia, with regard to the legal landscape for
patent inventorship and a concern that the PTO guidelines that
have been issued are not statutory; they are simply advisory.
The courts can go in a completely different direction, if they
wish. You don't see any concern about the fact that the
guidance leaves unclear the process the PTO will use to make
determinations about human and AI inventorship? For example,
the PTO doesn't typically question or examine whether a patent
application correctly listed the inventors. It doesn't require
any evidence about inventorship to be submitted. So, what are
the ramifications to both the examination process and the
resulting patent? Can the courts take this in a fundamentally
different direction than the guidelines?
Ms. Garcia. Thank you for the question.
I want to preface my answer by clarifying that I'm really
speaking to the Copyright Office and the guidance that it's
released thus far relating to registration determinations
around copyrightable material. So, I don't want to speak out of
turn to my more patent-focused colleagues.
Yes, I share the concern that right now it sounds like,
comparably, on both the patent and the copyright side, the
guidance is not the strongest guidance. It's not really giving
us a lot of clarity right now.
I think what I am, rather, saying is not that we'll not
need any legislative intervention. It's just that I wouldn't
know what to ask you for yet.
Like I said, the Copyright Office at least recently issued
a public call for comments on copyrightability in AI, and it
got over 10,000 responses, which is more, much more than
normal. It will be interesting to see what the outcome of that
review looks like and where people are landing.
I'm hoping that we've got lots of answers from different
artists as to what they think their creative processes look
like and where they think the line should be drawn. Then, my
thought would be, once we have more guidance on that, we would
be able to get something legislatively that would enforce
whatever the Copyright Office has determined.
Mr. Cline. I want to come back to copyright but let me give
Mr. Landau and Ms. Laporte the opportunity to respond to that.
Ms. Laporte. I'll give it a try. I have a lot of concern
about the complications in the system that can come from making
inquiry into the role of AI in biological inventions, anyway,
as something that becomes very important.
I'm a patent litigator. I did that for decades. I can tell
you right now that every case is now going to have a new layer
based on what the new guidance that the Patent Office has a new
layer of litigation about what tools did you use; how much AI
did you consult?
Also, what about the old computational biology tools?
Because the new guidance, by the way, does not define AI. So,
it could be anything.
So, we already have layer after layer after layer of
complication in patent litigation, and I worry that it just
makes the system unworkably expensive. It makes it a
competitive disadvantage for us.
Mr. Cline. Well, let's play devil's advocate. Could AI be
used to read through millions of patents to determine if any
were granted by mistake? If so, what would happen then?
Ms. Laporte. Well, I think that once the Patent Office
empowers AI searching for examiners, it may improve patent
quality. So, I actually think that the use of AI in engineering
disciplines is a very significant benefit, both for examination
and for invention purposes. It stands to help us get ahead
competitively, and I think we should just be careful not to do
anything that's going to inhibit the use of it.
Mr. Cline. Well, what if it's a foreign adversary that
wants to eat away at our patents and they employ AI to review
U.S.-granted patents? Could that lead to more patent challenges
in court? To what extent is China's utilization of AI in its
patent system influencing practices within the U.S. patent
system, especially as it relates to the biotech industry?
Ms. Laporte. So far, I don't think that we're seeing much
impact at all. The things that have to be done for China to
challenge patents are very cumbersome and expensive and one-
off. So, if you want to challenge a granted U.S. patent, you
have to challenge it in the PTAB, which is not an easy or
inexpensive process.
So, I'm actually not so concerned about that, particularly
because there isn't a competitive reason for that to occur in
most cases. Certainly, I think that AI is going to raise the
stakes on pretty much everything that happens, because it's
going to increase the accuracy and quality of what is done in
examination and in the inventing process to begin with. I think
that we should welcome that shift, and I don't think that there
is so much of a risk of sort of massive amounts of anything
coming from China, that it merits making a change in policy as
a result.
Mr. Cline. Thank you. I yield back.
Mr. Issa. I thank the gentleman.
We now go to the Ranking Member of the Full Committee, Mr.
Nadler, for his questions.
Mr. Nadler. Thank you, Mr. Chair.
In one sense, this hearing has been very interesting. You
are, essentially, advising us to do nothing, which sometimes is
Congress' best fallback position.
[Laughter.]
Let me ask you, Professor Garcia, in your testimony you
outline many legal reasons why predominantly AI-generated works
do not warrant copyright protection. Some may argue, however,
that denying AI-generated works protection will only
incentivize applicants to conceal their use of AI altogether.
How do you think we should balance the legal reasons to deny
protection with that practical policy concern?
Ms. Garcia. That's certainly one of among other moral
hazards to such a policy. I think that the down sides of
issuing copyright or wholly or substantially generated AI works
outweigh the possible consequences of folks covering up AI's
involvement. I might be an optimist, but I'm hopeful that the
documents that applicants have to sign saying they've submitted
truthful applications might deter some of that.
On the other side, if we really started issuing copyrights
to all the outputs that an AI generates, we would have a
massive influx of copyrights, right, because it can in a matter
of a half hour produce thousands of works. Pretty soon no one
can draw a cat in a tub because I put cat in a tub and got
8,000 different versions of cat in a tub and there's none left.
That to me, along with the fact that like who would we enforce
against if one of those cats in a tubs infringes another
person's work, seem to me like sort of bigger and more imminent
issues that people being dishonest on their applications.
Mr. Nadler. Thank you. Professor Aistars, do you agree?
Ms. Aistars. I generally do agree with that analysis.
There's also the backstop of prohibitions against including
inaccurate information with knowledge in a copyright
registration form. So, since we're all on knowledge that we are
supposed to, at least for the time being, disclose and disclaim
AI use that will be helpful to us. The Unicolors decision, the
Supreme Court, doesn't really help registrants all that much in
this case.
Mr. Nadler. Thank you. Mr. Landau, you have argued that if
AI output is patentable, foreign adversaries could flood the
Patent Office with AI-generated output overwhelming the agency.
Triaging and reviewing applications requires tremendous
resources even if the applications are ultimately rejected. So,
why should the PTO's AI guidance have any impact on the
potential flood of applications?
Mr. Landau. So, that the PTO's AI guidance will mostly
operate for the people who are not trying to trick the PTO. It
will mostly operate for people who are in good faith trying to
get their innovation through the office. The problem is that
even if the guidance is in force and the PTO at this point, at
least in the guidance, has nothing in there that affirmatively
asks, the examiner has to decide to ask was AI used in some
way? Even having that affirmative ask up front in the
application data sheet as part of the application process would
be a good step forward in putting the applicants on record.
There are tools that they used in the trademark context
about five years ago where they were receiving this flood of
fraudulent applications. They put into place a task force. They
increased their random audits to provide a backstop to try and
detect these fraudulent applications.
Mr. Nadler. Thank you. Ms. Laporte, some commentators have
raised concern the current U.S. conception law; that is, the
point at which an invention is deemed conceived, could if too
narrowly construed deprive human inventors of legitimate patent
rights in their inventions. They caution that this would put
the United States at a disadvantage to the rest of the world.
Is this a reasonable concern? If so, how should Congress seek
to prevent this misapplication?
Ms. Laporte. I think that it is a reasonable concern to
think that if we overregulate by requiring AI disclosures and
things of that kind that we could actually inhibit legitimate
inventions. There are so many steps and so much human
creativity that is actually involved in directing AI toward an
engineering task that the risk of inhibiting invention is the
greater risk that we should focus on.
Mr. Nadler. So, again, we shouldn't do anything here?
Ms. Laporte. Well, it would be a good thing to get rid of
the current Patent Office guidelines because those open up a
Pandora's Box of potential litigation complications that can
cause problems for inventions that are perfectly legitimate and
meet all the statutory criteria, the many statutory criteria
that invention needs to meet to be patentable.
I think adding another layer puts us at a big competitive
disadvantage because again, if somebody makes an invention in
an old-timey way, it's patentable, but if we make it in a
modern way using AI, then it's not patentable? Why is that a
good resolution? It doesn't make any sense. It puts us at a
competitive disadvantage because other economies are not
worried so much about who the inventor is. They're just worried
about making sure that they patent inventions that should be
patented. That's the focus we also should have.
Mr. Nadler. Thank you. My time is expired. I yield back.
Mr. Issa. I thank the gentleman. We now go to the gentleman
from California, Mr. Kiley.
Mr. Kiley. There is a very important distinction I think
between the two areas we have been talking about today;
copyrights and patents, which you alluded to in your testimony,
Ms. Laporte, which it is a different paradigm of creativity
versus usefulness. With a copyright you are really talking
about it is all about creativity whereas an invention is
measured against its usefulness in the real world. So, that
argues for treating these two areas in a different way and
perhaps having a more search and inquiry into the active
creation and copyright as opposed to patents where there is
just a different standard.
We already have in our system a lot of complications,
ambiguity, and confusion when it comes to patent subject matter
eligibility that has been introduced by the courts. We know
that this has in some ways chilled invention and investment in
emergent fields because of the uncertainly of being able to be
rewarded for your invention.
So, I am concerned about what this guidance and this whole
new inquiry of sort of looking under the hood of the inventive
process is going to mean as far as exacerbating those problems
and thereby limiting invention, limiting usefulness. I am
trying to imagine what discovery will even look like when you
now have this new tool to invalidate a patent saying, oh, no,
you crossed the line of using too much AI in such a way as to
invalidate it and what that is going to look like when we are
having these inquiries into the process of invention.
So, I wanted Ms. Laporte, to give you another opportunity
to sort of draw out what the implications of this might be. You
just said it is going to reduce your competitiveness. Do you
see this maybe leading inventors or companies to forego the
most efficient or most effective process in developing a drug,
or whatever the case may be, for fear that this will then lead
to a more vulnerable patent?
Ms. LaPorte. Sorry. I had it on. Yes, I do think there is
that risk. If you have to worry that your patent might be
invalid because of the fact that you used the most efficient
modern tools to develop your invention, that is of course a
disincentive both to using it and to disclosing it if you do
use it. So, to me that imparts a level of inefficiency to our
system that I do worry puts us at a competitive disadvantage.
To get a patent you have to meet a lot of statutory
criteria. You pointed to utility. It also has to be novel; it
has to be nonobvious, it has to be enabled, it has to be
adequately described. Of course, the process of description is
something that's very important because you need to--at least
in the biotech context you can't get a patent without
describing how you tested the thing that you invented and that
it worked.
So, there are so many hurdles already to patentability,
which is reasonable and valid. There should be. We shouldn't be
able to grant a patent unless it's a good patent, but once it
has actually met that standard, let's have that be patentable.
Mr. Kiley. So, there was some discussion; I think it was
from Mr. Landau actually, related to the fact that you just
cited of nonobviousness, how that is a requirement for
patentability and how the availability of new tools such as AI
is sort of baked into the nonobviousness inquiry because and
ordinary person, a personal of ordinary skill is going to have
access to AI tools.
So, if the availability of AI tools is already baked into
the nonobviousness inquiry, which is something that is its own
separate requirement, then why is it that we need to further
inquire as to the extent which there was a human-AI combination
or too much of one, too much of the other in the process?
I'll give you a hand, Mr. Landau, if you would like a
chance to opine on that.
Mr. Landau. Sure. So, the difference is that our laws say
that humans can be inventors and nonhumans cannot. It's the
same as in the copyright context. So, we do need to inquire
into who the actual inventor is. I don't think this is a
particularly new thing. We have had inventorship disputes and
litigation for probably as long as we have had a patent system.
It's an ordinary part of patent litigation; it's an ordinary
part of patent prosecution that you swear that you were the
inventor. So, I don't think that it's a huge ask of an
applicant to say, do you swear that you truly conceived this
invention?
Mr. Kiley. Ms. Laporte, did you have anything to add to
that?
Ms. Laporte. I think that the fact that an invention has to
be nonobvious means that as everybody begins to use AI it will
improve patent quality by giving much greater access to
examiners to things that they can look at to understand what
the closest prior art is. So, yes, I think that it has the
potential to improve patent quality and that we do not need as
a result of that to include the additional sort of layer of
bureaucracy and complication that is now being imposed by the
current Patent Office guidance. I think that should be
rescinded.
At this point, at least when it comes to biological
inventions, there's absolutely no chance that you can just
prompt AI to create something, or even have it create something
unprompted, as Stephen Thaler claims.
So, I am very, very skeptical about the need for this
additional regulation and I do worry that it hurts us in the
long run given the fact that we're the only country that
focuses on it in this way.
Mr. Kiley. Thank you. My time is expired.
Mr. Issa. I thank the gentleman.
We now recognize the gentlelady from North Carolina, Ms.
Ross, for her questioning.
Ms. Ross. Thank you, Mr. Chair.
Thank you to all the witnesses for sharing your expertise
with us.
Many companies and startups in the Research Triangle Area
in North Carolina, which I represent, utilize AI to accelerate
innovation, as Ms. Laporte has talked about, for instance to
sort through large data bases of genes and speed up analyses of
proteins and molecules. They use AI as a tool to assist with
the development they have done for decades. This assistance
could make an enormous difference for people with cancer or for
kids with rare diseases, and so many others that are waiting
for treatments for their conditions to come to market quicker
and more cheaply. The bottom line is that use of AI in biotech,
pharmaceutical, and medical research and development can help
people who are suffering get help faster. I firmly believe that
IP protection should be extensive enough to incentivize
investors to utilize new technologies like AI to accelerate the
research and bring their innovations to market as quickly as
possible.
The USPTO's guidance released last year affirms this
stance, but we have heard some problems with it, potentially.
It does provide that inventors may obtain patents for
inventions as long as the human contribution to the product is
significant. I am grateful for all your expertise. I would like
to stay with Ms. Laporte for my line of questioning because I
really want to focus on patents and the sciences.
You have talked about what is going on in the U.S. and
alluded to what is going on in other countries, but I would
like you to share any statutory or regulatory approaches in AI
and IP in other countries that might be either considered or
avoided by the United States, and whether we need to do
anything to remain competitive on a global stage.
Ms. Laporte. Well, I'm only admitted to practice in the
United States, and so I think as to details I should probably
get back to you after the fact.
I will say, however, that having directed global litigation
sometimes I would be involved in being U.S. counsel for a
company that was also being sued or was suing in many other
jurisdictions around the world that our system is uniquely
complicated and that puts a tax on innovation and the use of
innovation that we don't need. So, I do think that what you
said in your opening statement is exactly right, that we need
to focus on how we can most incentivize the use of the very
best tools to come up with the best inventions.
Just to add to that a little bit, the way that antibodies
were developed--many of them are manufactured in your
district--the way that antibodies have traditionally been
developed is by immunizing a mouse with a target antigen. Then,
you screen the antibodies that the mouse's body generates to
find the ones that meet what you want. Under the current PTO
Office guidelines, you might have to, perhaps, disclose that
the mouse had some inventive role. It's just not clear to me
that there's any benefit of trying to figure out how many
angels can dance on the head of a pin when you have a useful
invention that meets all the statutory criteria.
Ms. Ross. OK. Well, to followup on that, some commentators
advocate for a requirement that applicants formally document
inventor contributions including details about the
contributions of AI. I guess that could have also applied to
other things in computational biology. They argue that this
would complement already common practices like maintaining a
lab notebook.
What do you think of this idea and is your company
documenting its use of AI in a manner like this?
Ms. Laporte. Certainly, we keep good records about what the
science we do and how we do it, but that inquiring into the
scientific practices and strategies of a company is something
that inhibits the use of the patent system, because of course
the specific approaches that you take, the mental ways that you
think about approaching a problem, those are all actually trade
secrets or creative things that are not things that are
disclosed. When you make an invention, you have to say how to
make and use it. You do not need to say all about what your
creative thought process was that went into how you make and
use it. So, we should not try to expand the scope of those
requirements.
Ms. Ross. I very much appreciate your responses. Thank you
and I yield back.
Mr. Issa. I thank the gentlelady. I will take the liberty
of going next to the gentlelady from California, Ms. Lofgren,
for her questions.
Ms. Lofgren. Thank you very much. This is a very
interesting and useful hearing and some of the things you have
talked about hadn't thought about before.
I think the Copyright Office is making good steps forward
and trying to figure out where is that line. I think we all
agree we need to protect the humans here in terms of their
creativity, but where is the line? If you give 15 prompts is
that enough? I think that is something that they are going to
be able to come back to us with and that will be very useful.
In terms of the Patent Office, Ms. Laporte, I think you
have made a powerful case about the guidance, and it causes me
some questions. For example, Google unfolded every protein in
the human body and posted it for every scientist in the world
to use. Would that violate the standards if people use that AI-
generated information? In the case of fusion right now people
are using AI to understand the actions of plasma interacting
with targets to advance fusion. Would the use of AI to
understand plasma run afoul of the guidance that the PTO has
given? Also, the Google question.
Ms. Laporte. Well, I'll have to defer on the plasma
question because I'm not sure I fully understand the basis of
that, but we can I'm sure get at this issue through the use of
the tools relating to protein folding.
So, we now have very useful tools that can take a sequence
and tell you what expected structures of the protein is going
to be. Those are tools that are now universally used by
everybody including I will assume patent examiners if a
relevant question comes up because they're publicly available.
So, that to me just becomes part of what somebody now with
ordinary skill in the arts is definitely going to use it.
Those things are public. There's nothing that gives one
person an advantage over another in inventing as a result of
using a publicly available computational tool. So, I can't
imagine any good reason why we wouldn't want everybody to use
those who's trying to create something new and useful.
Ms. Lofgren. So, the answer really is the guidance would
not apply if the AI-generated information is available to
everybody. Wouldn't that be true of any use of AI? Because
anybody could use the tool.
Ms. Laporte. I may have misinterpreted your earlier
question. So, in terms of whether that should be disclosed, it
seems to me that now since it is a very commonly used tool it
might not even occur to people to disclose that. However, I do
think that part of what's happening, in biology at least, and
probably in other engineering disciplines, is that, in addition
to there being publicly available AI tools there are also
companies that are developing their own private AI.
Ms. Lofgren. Oh, I get it. Yes.
Ms. Laporte. So, there may be some significant differences
there in terms of whether we have a level playing field.
Ms. Lofgren. Right. I have a question--Ms. Garcia, maybe it
was you; I can't remember who said that if the courts find that
the use of the data for machine learning was infringing, that
any use of AI would also then be infringing. Did I
misunderstand that?
Ms. Garcia. What I said is if the training of the AI on
copyrighted works is found not to be fair use such that it
constitutes--
Ms. Lofgren. Correct.
Ms. Garcia. --copyright infringement, any output of that
language model trained on infringing content would be
uncopyright-able because we don't copyright infringing work.
Ms. Lofgren. OK. So, I basically did understand it.
Ms. Garcia. Yes.
Ms. Lofgren. OK. So, one of the things that I agree that we
ought to tread carefully that the PTO and Copyright Office are
working diligently. Here is the question, whether we ought to
deal with remedies. Because I believe that if there was
infringement on the data collection for learning, there ought
to be compensation, but I do worry if there were injunctive
relief so that AI in America was enjoined, that would put us at
a tremendous disadvantage versus China and others.
Do you think we should consider whether we should direct
the remedy toward monetary compensation as opposed to shutting
down AI in America?
Ms. Garcia. Certainly, from an innovation standpoint I
think that a remedy aimed at monetary damages makes a lot more
sense. I'd also be hopeful that in an addition to or alongside
that we might see efforts toward careful efforts of very well-
crafted efforts toward some of what I call a liability rule or
setting up a licensing scheme, right, to have these all over
copyright where we have collectives.
Ms. Lofgren. Right. Mandatory licensing of some sort.
Ms. Garcia. Compulsory licensing, yes.
Ms. Lofgren. Thank you. Mr. Chair my time is expired.
Ms. Garcia. Thank you.
Mr. Issa. I thank the gentlelady.
We now go to the gentleman from California if he is ready,
Mr. Schiff.
Mr. Schiff. Thank you, Mr. Chair.
There is no doubt that AI has the potential to change our
economy, our political system, our day-to-day life in
potentially very good ways and also very disruptive ways. AI
has opened a world of opportunity for creators, whether that is
utilizing binary technology to isolate vocal tracks or generate
unique compositions that artists can incorporate into their
work, but we have to balance technological advancement with the
protection of individual creators' rights, ensuring that
progress does not come at the expense of the creative
community. After all, human creativity powers AI and makes
these advancements possible.
As has been raised during the hearing today the rapid
development of generative AI technologies has outpaced existing
copyright laws which has led to widespread use of creative
content to train generative AI models without consent or
compensation. As you are indicating it might violate the
copyright laws.
This is why yesterday I introduced the Generative AI
Copyright Disclosure Act which will further innovation and
safeguard the contribution of creators, ensuring that at a
minimum they are aware of when their work contributes to AI
training data sets. Similar, to proposals put forward in the
EU, my bill mandates transparency from companies by requiring
they disclose a summary of copyrighted works used to train AI
systems, ensuring creators are informed and have the tools to
advocate for credit or compensation where it is due.
This bill is a step and only a first step in this direction
and I welcome the opportunity to work with Members of the
Subcommittee to champion innovation while safeguarding the
rights and contributions of creators. I am very grateful to
have the support of many leading organizations in the creative
industry including the Human Artistry Campaign, Recording
Industry Association of America, SAG-AFTRA, and IATSE, to name
a few.
Now that I have finished the commercial announcement, one
primary concern many of these creative groups have been that
generative AI models will replace writers in the writing room,
actors on set, or artists in the recording studio.
Let me start if I can, Professor Garcia, while new
technologies like AI offer a breath of advancements in the
production space to utilize or partner with what can creators,
many of whose livelihood depends on copyright protection,
expect in terms of AI authorship decisions from the Copyright
Office? What do you anticipate?
Ms. Garcia. Thank you for the question. I think that given
the guidance that we've seen so far, which as I mentioned
previously is a really small set, right--we've got four such
generative AI-produced works, registration determinations. If
the work is wholly or substantially generated by AI, they can
expect that the Copyright Office will not grant protection.
Where it's partially AI-generated the Copyright Office, as
we've been discussing, appears to be taking a disclose and
disclaim approach, which is easier said than done. With the
notion being that what would happen if you wanted to secure a
copyright registration in a work for which an artist had used
generative AI, they may have to first disclose that they've
done so, and then second, make a good-faith attempt to describe
how much of the work was done by generative AI.
Here a really important distinction to make that I've been
seeing from the questions and from testimony today is the
difference between using AI as a tool, much like a paintbrush,
and generative AI, which is actually creating something new. I
think it's the latter that really adds challenges. Artists
should not be concerned about using AI as a tool to aid in
their creativity, only when they're using generative AI to
substitute for their creativity.
Mr. Schiff. Where do you draw the line of human versus AI
authorship and how do you think that the specific material an
AI model is trained on should be considered? I would throw that
open to the whole panel.
Ms. Garcia. Sure. So, I'll start. I wish I knew exactly
where to draw the line. I'll say again that I'd love to see
more of these applications come in so we can see what people
are doing with the generative AI, to get a sense of it. If I
had to give one criterion now it would be to consider whether
the final product reflects the original intellectual conception
of a human. If so, then I think I would lean toward granting
copyright protection.
Mr. Schiff. That is very hard to define, isn't it?
Ms. Garcia. Yes.
Mr. Schiff. Would others care to weigh in?
Mr. Landau. Sure. Creativity is hard to define as well, so
I think that this is part and parcel of copyright is how do we
say that something is creative? How do we define creativity? In
the context of AI, I agree with Professor Garcia completely
that it's the human using it as a tool?
I think it was the Ranking Member that mentioned autotune,
which is its own form of AI and has been used for--I don't know
if it's decades now, but for quite a while. We've seen similar
tools in the patent context. Those tools are fine. The
generative AI is where it can step over that line. I don't
think we have equivalents in the patent space yet, certainly
not the tech side, and I'm hearing not really on the bio side,
where you just say give me this and it spits something out. If
we get to that point, then that will become much more an
important question on patents. Right now, it's probably not.
Mr. Schiff. Thank you, Mr. Chair.
Mr. Issa. I thank my colleague. I might note I also thank
him for the submission of his bill. I know that yours and Mr.
Cline's bills; his is forthcoming, are going to be taken up by
this Subcommittee. So, I appreciate your thoughtful work on
that.
We now go to the gentleman from Maryland who has been so
patient, Mr. Ivey.
Mr. Ivey. Thank you, Mr. Chair.
Let me followup on that line of questioning, because I am a
little unsure about how we are trying to draw a line with
respect to what is creative, when AI is a tool as opposed to
when it is actually creating new work I guess of some sort.
Because I just came back from a conference, and they were
talking about generative AI is not creative and that it is
actually a tool.
So, Professor Garcia, let me come back to you because you
talked a little bit about that line that you were drawing. Does
that vary from the type of work we are talking about or where
would you draw the line with respect to not only that
distinction, but other types of works? Let's say like art,
music, biotech, and investigation? How do you differentiate
between there?
Ms. Garcia. Thank you for the question, and I think that
you've identified one of the primary challenges, particularly
in the copyright space is that copyright as an umbrella statute
tries to cover so many different types of works. The use of
generative AI and the use of AI tools looks dramatically
different from visual arts to music to film and so forth.
Again, I've tried to identify in my written testimony a few
things like predictability and control, which I recognize can
to some extent also be applied in the use of AI as tool, which
I've suggested is OK for copyright-ability purposes.
Here's an example I'd give: One of our colleagues had
mentioned Jackson Pollock earlier, where if he just took a
paintbrush and splashed it onto a canvas, he wouldn't be able
to predict exactly what it would look like. I would argue
differently to giving a simple prompt to AI to like, make a
painting he would have had more input, right? He picked the
color of the paint and the force with which he splashed and the
angle at which he splashed it. So, we could see Pollock in that
work as opposed to me just telling Midjourney like make a
painting.
Mr. Ivey. You have got a better eye for Pollock than I do,
I guess.
[Laughter.]
Mr. Ivey. OK. Others on the panel, what are your thoughts
on that?
Ms. Aistars. So, if I may, I think there may be some
misunderstanding in terms of how artists actually engage with a
tool like Midjourney and also how artists develop and engage
with tools, any tools over time. So, that's why I say we start
with the fact that it is an artist's intellectual conception.
Then when we translate that idea over to the context of a GAI
tool, which is in fact a two-way interaction between the human
and the GAI--I'm suggesting that intellectual conception. What
we're looking for next then is whether the artist is continuing
that intellectual conception in a manner that demonstrates that
the artist is staying true to their creative vision. So, the
artist isn't just--
Mr. Ivey. Let me just pause you there. OK. So, I think I
believe it was here--the prompt was to create a trailer for a
science fiction movie of a Martian invasion or something, and
it came up like that. I guess that is a generative AI tool. So,
the creative vision--
Ms. Aistars. Right. I would argue that this is not an
intellectual conception and that is not a sufficient direction
from the prompter to have engaged the tool with an authentic
creative voice that is conveying something of the artist--
Mr. Ivey. All right. So, let me ask you a followup
question: Let's say the person who gave the initial prompt then
says change this and then change that. At what point does it
move from being--does it become a creative product for the
artist as opposed to just a generative AI that would not get
that copyright protection?
Ms. Aistars. Yes, so that's actually not how the tools
work. So, you're not saying like change this, change that like
you might say as a creative art director to a client's
commissioning an artist to do something. You're actually often
as the user of the tool giving prompts that say I want to use,
this equivalent of this f-stop the equivalent of bond paper
this media.
Mr. Ivey. Let me make it really easy.
To test out ChatGPT I said generate a sermon by Martin
Luther King. It came up in two seconds. If I then took that
document--and a couple of them were explicit quotes, but mostly
not. If I then took that document and edited it out, or even
better I said generate a speech of Glenn Ivey and then made
changes to it, some editing, wouldn't that at that point become
a creative process that I had generated?
Ms. Aistars. So, I would argue no, you're actually saying
to the tool infringe somebody's copyright.
Mr. Ivey. Well, it would have been me, right?
Ms. Aistars. Pardon me?
Mr. Ivey. It would have been me, my second--
Ms. Aistars. Oh, you said Martin Luther King. Sorry.
Mr. Ivey. The second was me, Glenn Ivey. Right here.
Ms. Aistars. OK. So, in that case you're asking to make a
derivative work of your own work so our inquiry would really
not be so much whether you are entitled to a copyright on the
first work, but whether you've made sufficient changes to that
second work to be entitled to it.
Mr. Ivey. Right.
Ms. Aistars. So, it's really not an AI issue as much as it
is whether you've made changes to your second work that are
sufficient for a copyright in your second work.
Mr. Ivey. Well, I see I have run over my time, but thank
you, Mr. Chair. Thank you.
Mr. Issa. I thank you. You did it very well.
Mr. Ivey. Thank you.
Mr. Issa. I recognize myself and I am going to pick up
where Mr. Ivey left off. We all agree that if the Beatles
remaster one of their works it is still copyrighted, right?
I get all yeses. Beatles remaster but they use AI, is it
still copyrighted?
[No response.]
Mr. Issa. I am getting mostly yeses.
Ms. Garcia, you had earlier talked about--to that point a--
let me make sure I get it right here. I put down remastered on
your point because you were talking about original works that
went into or were ingested into, if you will, AI. You said
something that was interesting. Ms. Lofgren, I thought was
going to go there. She didn't quite get to the point I had,
which was if you have ripped off the work and then you create
something, isn't the underlying copyright still there?
In other words, this new product, which is a derivative
with copyrighted material in the extent that this what Mr.
Schiff has been working on. To the extent that you are in fact
taking earlier copyright work, and that those works are less
than 70 years past the life of the author, aren't we still
seeing a modicum of copyright in that product, meaning that you
may have ripped it off, but it doesn't mean that there wouldn't
be some potential protection against just everybody else using
it themselves?
Ms. Garcia. Well, as I understand what you've described,
this would be--if you're using an AI that's been trained on
infringing material let's say, or material that's found to not
be fair use, so it infringes.
Mr. Issa. No, no, no.
Ms. Garcia. No.
Mr. Issa. Again, you use the term fair use, and I am
drifting off what you said.
Ms. Garcia. OK.
Mr. Issa. Let's assume it isn't fair use. Then it is in
fact copyrighted material. Let's not assume the word ripped off
for a moment. Let's simply say it is contained. To the extent
that libraries are built using copyrighted material--Ms.
Aistars, I am going to use the Getty Library, OK, all that
massive amount of video and still pictures. To the extent that
this is ingested and then something is created (A) Getty is
going to tell me, as they have, hey, you owe me. You don't have
the right to steal it, but you owe me. To the extent that the
output would not be given a copyright--if they are given no
copyright, then where are we going to get the revenue from for
that?
In a sense, don't we have an obligation to look at--and I
am going to go particularly to Ms. Aistars for a reason. To the
extent that we are using the video visual art that we have so
much of we have an obligation I will say to have an inducement
of value if we want to have those derivative creations.
Otherwise, there won't be those derivative creations if you
have to pay on the input and you can't monetize on the output.
Wouldn't you agree that at least, in principle, that is one of
the challenges that this body, that isn't supposed to do any
legislating here according to all four of you, probably has to
look at? I am asking you because it is sort of more your end of
the business.
Ms. Aistars. Yes, so I began by saying that we need to
resolve the liability issue for training so that we aren't
relegating these classes of--
Mr. Issa. Yes, and we can't resolve it here today, and that
is one of the reasons that I am simply saying to the extent
that it is not fair use, that in fact there has to be some sort
of revenue process to the original creators. Now, those revenue
processes are unlikely to be on the ingest side. They are more
likely to be on the output side. That is one of the challenges
that this Committee is looking at and anticipating is that
copyrighted--if we don't grant a copyright to output material,
then we don't have the ability to compensate the input and it
favors the case of saying, well, it is fair use then. Why don't
you think about that for a question? I am going to come back to
you.
Ms. Aistars. OK.
Mr. Issa. I want to go over to the other side really quick.
There was a lot of discussion--and you--Ms. Laporte, you used
the word prompt a couple of times. One of the questions, since
you do not think the current guidance from the PTO is good--it
is one of the things that you asked us to do during this
hearing is to maybe undo it.
Isn't it practice typically in patents that when you are
describing your invention and you have all your little charts,
maybe not always in your regime--but in electronics, we almost
always--we show--in my era we showed a slide ruler and an
abacus, but today they show the computer, they show other
instruments. In a sense isn't what the Patent Office should be
saying is tell us your prompts and tell us the AI library that
you used so that, in fact, somebody of ordinary skill can (1)
duplicate and (2) verify whether or not it falls under
obviousness based on a limited number of prompting, or a
relatively large amount of prompting and how the AI model would
have dealt with it?
Ms. Aistars. So, we're not patenting our AI tools. We're
patenting specific genetic constructs that we make using the
tools. So, the outcome of what we do is likely to be a DNA
sequence of some kind, which--
Mr. Issa. No, I understand that. The reason I was asking
though is that you have to disclose to somebody of ordinary
skill.
Ms. Aistars. You have to disclose how to make and use the
invention. It doesn't mean that you have--
Mr. Issa. Right. You also have an obviousness test. So, the
Patent Office is concerned about large use of generative AI
ultimately creating patents that are applied for that are
basically machine-created and therefore obvious to anyone who
has the machine.
So, back to the question which was fairly--I wanted to make
it fairly narrow. Shouldn't the Patent Office, rather than the
tests that they are giving, simply say that as we have done in
patents for generations, the tools used are disclosed often? In
this case, the prompts requested to the extent that they are
digital prompts--a musician who is playing an electronic piano
can't really define his prompts. You can define the prompts you
gave as part of the process, couldn't you?
Ms. Aistars. Not necessarily. I think particularly given
the fact that we are all likely to be developing proprietary AI
systems using proprietary data, and while my company is very
interested in trying to make sure that there's as much data
sharing as possible, we still all develop things along slightly
different pathways.
Mr. Issa. Well, I appreciate that, but that is one of the
questions I have, which is my obligation is to make sure that
all industries disclose that which allows someone of ordinary
skill, not someone of ordinary skill with your special sauce
that isn't disclosed, to duplicate the process later. That
becomes a very important point in granting of patent is we are
granting patents not so you can be enriched, but so 20 years
later society shall be enriched.
Ms. Aistars. Absolutely, but the thing is that what we--the
invented thing is going to be a biological construct, not
software, not AI. We are using AI to solve a biological
problem. What we're patenting is the solution to the biological
problem. So, you can absolutely use that thing 20 years after
the patent is applied for. It will be open source. Anybody will
be able to use that thing. It's easy to make. The things that
we make are typically--well, sometimes they're very hard to
make, but often the--like if I make a new enzyme, for example,
it's easy to be able to teach somebody how to make and use that
enzyme. I do not need to teach them how the AI that I prompted
and what data I used and everything else. That is irrelevant to
being able to make and use the invention.
Mr. Issa. I appreciate that. Briefly, have you had time to
come back with how I should deal with this conundrum that Ms.
Garcia gave us if it is not fair use, but we are also not
granting a copyright on this derivative work, even if it is a
derivative work that is limited to 70 years from its input
materials, how do we in fact incentivize those new creations?
Ms. Aistars. Yes, so in my testimony I have been telling
you to grant a copyright on works so long as the author
demonstrates minimal creativity in the works. We've been doing
so. I've also suggested that human authors should be
compensated and have control of their works when they're
ingested into an AI data training set. So, that's for another
hearing.
I wanted to be clear though, I think there may be some
misunderstanding in terms of how I view the copyright guidance.
I think we're thinking about filling out our registration
applications like lawyers think about filling out registration
applications. So, to us we look at the application and we say
well, what's easy? Like let's put a number on it. Let's put a
bright-line test. How much, right? Quantitatively how much is
too much GAI use?
That's not an artist's process when an artist is creating a
work. An artist is not sitting there saying like, OK, I've used
this many prompts, I've engaged this way, right? An artist is
in the flow working and creating the image that they're
creating and they're not stopping to think that way. So, I'm
trying to suggest how to make a test that is intuitive to an
artist and asking you to think qualitatively about how the
artist is using the tool rather than quantitatively.
Mr. Issa. On that note--
Ms. Jackson Lee. Mr. Chair?
Mr. Issa. Yes. The gentlelady from Texas.
Ms. Jackson Lee. Mr. Chair, you have been a gracious Chair,
I understand, and I do know the protocols of--
Mr. Issa. Would the gentlelady like a short time to be
waived in at the discretion of the Chair at this time?
Ms. Jackson Lee. I would be overwhelmed if I was given
that--
Mr. Issa. I am an overwhelming supporter. The gentlelady is
recognized.
Ms. Jackson Lee. Well, I want everyone to know that Chair
Issa--and I thank the very gracious Ranking Member Mr. Jackson,
but Chair Issa and I have been on these matters for a long
time. So, the only thing that I will pose, not to prolong, is
the question of sourcing and--which it looks as I was listening
sourcing was permeated throughout--the artist is in the flow.
That means that the artist is not necessarily sourcing. They
are just in the flow.
So, can someone just take a stab at the random sourcing
artist versus the more precise engineer when it comes to AI? I
have some legislation that I would like this Committee to look
at. I am not on the Committee, but have a deep abiding
commitment and interest to this area. The whole idea of
sourcing in the AI world, would someone care to comment on how
meticulous we have to be? What does that do to the college
student that is now facing that, particularly the grad student
that is now facing that? Without any boundaries what do we have
now? What world are we living in now on sourcing? I hope I have
tried to be--I have professors here. I will yield to whoever
wants to take a stab at that.
Ms. Aistars. So, I'll just say I feel very strongly that
there needs to be liability for the GAI companies who train
their tools. They need to appropriately source materials to
train their tools and that artists whose creative works are
used to train GAI tools should have compensation and control
over the use of their works. So, that's my position on training
GAI tools.
This particular hearing we weren't asked to talk about that
issue and so my view is that liability for training of GAIs has
to be resolved, but that denying IP protection to otherwise
protectable works to humans who use GAIs is counterproductive,
because that will relegate the work that they create using GAIs
to this category of synthetic data. That will harm the
legitimate human creators of works by harming them on the input
side, by using their work to train the machine, and then
harming them again on the output side by not giving them the
ability to get a copyright on their work if they have otherwise
met the originality requirements in creating that work. So,
that would be my position.
Ms. Jackson Lee. I don't know if there is one or the other
that wanted to answer, and I realized what the topic of the
hearing was, I figured I should get in the door before it got
closed on me.
Anyone else want to just comment?
Mr. Landau. Yes, I think if I understood you--your question
correctly, you are asking about sourcing on the output side so
that you would know what portion of your work comes from the AI
versus what portion comes from the person. I think that's
actually a really important question because at the end of the
day, if you can't point to something that was sourced by the
AI, then you have surpassed any threshold. It is your work at
that point. It's when you can point to something and say, yes,
that part right there, that came from the AI. That's when you
should not be receiving copyright over at least that portion of
the work.
Ms. Jackson Lee. Very important.
Mr. Issa. I thank the gentlelady.
Ms. Jackson Lee. He was very kind and the professor was
trying to finish her sentence.
Mr. Issa. Please. I didn't mean to cut you off.
Ms. Jackson Lee. Yes.
Ms. Garcia. No, no. Sorry. I was going to add that I think
I agree with Professor Aistars' assessment that even though it
is not the topic of this particular hearing, an answer on the
training data is important because I also empathize with the
inclination to say, well, if we can't compensate them on the
inputs; I think we probably should, then maybe let's compensate
on them on the outputs.
(1) The challenge statutorily, right, is that our current
definition for derivative works doesn't allow those outputs to
be copyrightable, right, because they can't point to a
particular preexisting work; it's just billions of works.
(2) They are themselves not copyrightable.
(3) Most importantly, they're not authorized. So, all those
things would have to be true to get a proper derivative work
copyright.
Mr. Issa. With that, I am going to ask all the witnesses to
give us three--I thank the gentlelady--three indulgences.
Ms. Jackson Lee. Thank you.
Mr. Issa. (1) Answer questions that will be submitted from
any Member that was here or unable to be here including Ms.
Jackson Lee.
(2) Please revise and extend thoughts that you have as a
result of today's hearings and--Ms. Garcia and the others,
including areas that might be tangential to this, but clearly
cover this Subcommittee's jurisdiction and AI. So, don't be
inhibited by the limitation of this because we can't have you
on every panel, we can't have you at every roundtable, but we
want your input. So, if you will do all that for us, we would
appreciate it.
Additionally, the Ranking Member and myself are always
available for ex parte-type events if you have additional ideas
or requests. This is a challenge for us because as all of you
said, you don't want us to do anything except perhaps undo what
has already been done. We see a lot of potential wisdom in
that, but we also are monitoring this in a way in which we want
to make sure that we prevent harm from developing.
With that, not having gone back into my Billy Joel example
to any of you, which would have been my next question, we stand
adjourned.
[Whereupon, at 12:05 p.m., the Subcommittee was adjourned.]
All materials submitted for the record by Members of the
Subcommittee on Courts, Intellectual Property, and the Internet
can
be found at: https://docs.house.gov/Committee/Calendar/ByEvent
.aspx?EventID=117077.
[all]