FedSoc Forums
FedSoc Forums
✓

AI Meets Copyright: Understanding New York Times v. OpenAI

· 전사
01:04:52
Podcasts
RSS
Link

이 에피소드에 관해

Artificial intelligence is the most important technological tool being developed today, but the use of preexisting copyrighted works to train these AI systems is deeply controversial. At the end of 2023 the New York Times sued OpenAI and Microsoft, alleging that OpenAI's use of articles from the New York Times to train their ChatGPT large language model constitutes copyright infringement. An answer is due at the end of February, and it's expected the case will revolve on the question of whether the use of the copyrighted content of the Times was a fair use. The fair use analysis will likely turn on whether the use of copyrighted content to train a AI system "transforms" the work in a way which makes the use fair. The Supreme Court has spoken on this question twice recently, holding that Google's use of parts of Oracle's Java programming language to build the Android operating system was transformative, but that the licensing of a Andy Warhol work based on a photograph by Lynn Goldsmith was not transformative of Goldsmith's work. Also important and perhaps most on-point is a decision of the Ninth Circuit Court of Appeals that Google's Image Search system is transformative of the photographs it indexes and displays as thumbnails.

To help understand this case Professors Charles Duan from the American University Washington College of Law and Zvi Rosen of the Simmons School of Law at Southern Illinois University was joined by Steven Tepp of Sentinel Worldwide, who is also a Lecturer at the George Washington University School of Law and formerly of the U.S. Copyright Office. John Moran of Holland & Knight moderated the panel and provided additional perspective.
뉴스
정부
비영리
영어
미국
노골적인 콘텐츠
API를 사용하여JSON에서 이 에피소드를 가져옵니다.

이 에피소드 내

이 태그 수정하기

전사 🔗

WEBVTT
1
00:00:02.080 --> 00:00:06.400
Welcome to fedsock Forums, a podcast
of the Federal Society's Practice groups. I'm
2
00:00:06.440 --> 00:00:10.039
Ny kas Merrick, Vice President and
Director of Practice Groups at the Federal Society.
3
00:00:10.240 --> 00:00:14.640
For exclusive access to live recordings of
fedsock Forum programs, become a Federal
4
00:00:14.679 --> 00:00:20.679
Society member today at fedsoc dot org. Hello everyone, and welcome to this
5
00:00:20.719 --> 00:00:24.839
Federal Society virtual event. My name
is Emily Manning, and I'm an Associate
6
00:00:24.879 --> 00:00:29.000
director of Practice Groups with the Federal
Society. Today we're excited to host a
7
00:00:29.039 --> 00:00:35.240
discussion titled AI meets Copyright Understanding New
York Times v OpenAI. We're joined today
8
00:00:35.320 --> 00:00:40.320
by Charles Dwan V. Rosen,
Stephen M. Tepp, and our moderator
9
00:00:40.320 --> 00:00:44.960
today is John P. Moran of
Council at Holland and Knight. John has
10
00:00:45.000 --> 00:00:50.840
experienced litigating many patent, trademark and
trade secret cases and Federal District Court and
11
00:00:51.000 --> 00:00:55.280
argues appeals at the US Court of
Appeals for the Federal Circuit the US Court
12
00:00:55.280 --> 00:01:00.000
of Appeals for the Fourth Circuit.
He has prosecuted or directly supervised the prosecute
13
00:01:00.359 --> 00:01:06.680
of hundreds of patent applications and many
different technologies, including telecommunication systems and equipment
14
00:01:07.120 --> 00:01:15.480
robotics, artificial intelligence, imaging technology, nuclear reactor instrumentation, semiconductor devices and
15
00:01:15.560 --> 00:01:19.879
manufacturing processes and medical devices. If
you'd like to learn more about today's speakers,
16
00:01:19.920 --> 00:01:23.959
their full bios can be viewed on
our website FEDSOC dot org. After
17
00:01:25.000 --> 00:01:27.359
our speakers give their opening remarks,
we will turn to you the audience,
18
00:01:27.359 --> 00:01:30.480
for questions. If you have a
question, please enter it into the Q
19
00:01:30.560 --> 00:01:34.200
and A function at the bottom of
your zoom window, and we will do
20
00:01:34.239 --> 00:01:37.560
our best answer as many as we
can. Finally, I'll note that,
21
00:01:37.640 --> 00:01:41.079
as always, all expressions of opinion
today are those of our guest speakers,
22
00:01:41.319 --> 00:01:44.480
not the Federal Society. With that, thank you for joining us today,
23
00:01:44.519 --> 00:01:48.680
and John, the floor is yours. Thank you Emily for that introduction.
24
00:01:49.599 --> 00:01:53.840
Well, this afternoon's panel is Emily
indicated. We'll discuss the copyrighted issues raised
25
00:01:55.040 --> 00:02:00.480
in the complaint brought by The New
York Times against Microsoft and several Open AI
26
00:02:00.840 --> 00:02:06.400
entity. The complaint, which was
fought last December, has seven counts.
27
00:02:06.760 --> 00:02:12.960
Four of the counts are for copyright
infringement counts. There's a Digital Millennium Copyright
28
00:02:13.080 --> 00:02:20.439
Act count and a common unfair competition
by misappropriation of New York Times intellectual property.
29
00:02:21.080 --> 00:02:25.800
There's also a trademark dilution count,
which will not be addressed today.
30
00:02:27.199 --> 00:02:34.759
In support of the copyright infringement allegations, New York Times included in the complaint
31
00:02:34.919 --> 00:02:40.400
one hundred examples of outputs that are
nearly identicate and identical to the copyrighted New
32
00:02:40.479 --> 00:02:47.240
York Times content. Before we address
the copyright issues, a bit of technology
33
00:02:47.319 --> 00:02:53.960
vocabulary at a very high level may
help the discussion, so to begin with,
34
00:02:53.159 --> 00:03:00.639
the complaint focuses on the open aised
GPT models as it relates to today's
35
00:03:00.639 --> 00:03:09.719
discussion. The acronym stands for generative
pre trained transformer. The generative, which
36
00:03:09.759 --> 00:03:15.479
we will probably discussed today, indicates
that the model takes inputs from users and
37
00:03:15.599 --> 00:03:22.240
generates an output such as one hundred
examples of New York Times content and the
38
00:03:22.280 --> 00:03:28.479
complaint. The P for pre trained
indicates that the model is pre trained on
39
00:03:28.599 --> 00:03:32.719
the volume of data such as the
New York Times content, and lastly,
40
00:03:34.159 --> 00:03:39.800
the T for transformer very generally relates
to a way of processing data to provide
41
00:03:39.960 --> 00:03:46.639
context for the data, such as
a sequence of words. One note that
42
00:03:46.759 --> 00:03:52.479
the complaint uses the term embedded.
It appears that it uses that term in
43
00:03:52.560 --> 00:03:57.240
the non artificial intelligence sense, that
is, the sense to the stone is
44
00:03:57.240 --> 00:04:05.840
embedded in concrete, other than the
process of embedding converting words into corresponding numerical
45
00:04:05.960 --> 00:04:13.599
values. Those topics were probably those
concepts will probably discussed during the panel's discussion
46
00:04:13.680 --> 00:04:19.759
today. So as emily indicated to
discuss the New York Times allegations of copyright
47
00:04:19.800 --> 00:04:27.680
infringement, we have three experts.
Zev Rosen is an assistant professor, School
48
00:04:27.680 --> 00:04:33.319
of Law at Southern Illinois University.
It was the Abraham Chemistine Scholar in Residence
49
00:04:33.399 --> 00:04:40.600
at the United States Copyright Office and
is particularly renowned for his expertise on copyright
50
00:04:40.720 --> 00:04:47.199
law, especially in its historical development. Charles One is an assistant professor at
51
00:04:47.240 --> 00:04:54.079
the American University College of Law.
It was previously a post doctoral fellow at
52
00:04:54.160 --> 00:05:00.439
Cornarial Tech. Working with professional James
Grimmleman. Charles focuses his research on the
53
00:05:00.439 --> 00:05:08.079
public effects of technology policy, intellectual
property law, and primarily patents and copyrights.
54
00:05:09.720 --> 00:05:15.079
Steve Tapp is a president and CEO
of Sentinel Worldwide, which is an
55
00:05:15.079 --> 00:05:20.959
intellectual property consultancy. He's also a
lecturer in law at the George Washington University
56
00:05:21.800 --> 00:05:29.240
School of law. More recently,
Steve has co founded rights Cliques, which
57
00:05:29.279 --> 00:05:33.560
is a suite of software tools for
independent creators to register, manage, and
58
00:05:33.639 --> 00:05:40.839
enforce their copyrights. As Emily indicated, we'll start with opening statements, will
59
00:05:40.879 --> 00:05:45.560
begin with Zeve's, then Charles and
followed by Steve. So with that,
60
00:05:46.560 --> 00:05:50.199
I'll turn it over to z.
All right, everyone, good afternoon or
61
00:05:50.279 --> 00:05:57.600
morning, depending where you are there
are. This causes really fascinating because we're
62
00:05:57.639 --> 00:06:04.800
finally getting at a host of important
issues regarding AI and the Internet, and
63
00:06:05.000 --> 00:06:11.240
what we call traditional producer is a
cultural works we of New York Times or
64
00:06:11.319 --> 00:06:15.920
many other that I think will probably
be honest cot tales. There's a couple
65
00:06:15.920 --> 00:06:19.319
of legal issues I want to highlight, and I'd love some questions to follow
66
00:06:19.399 --> 00:06:26.199
this once my colleagues over opening statements. The initial issue here is really direct
67
00:06:26.279 --> 00:06:30.560
infringement, which is to say,
by taking all of these New York Times
68
00:06:30.560 --> 00:06:36.959
stories, they are effectively copying them
or creating derivative works of them. A
69
00:06:38.000 --> 00:06:42.600
derivative work is a work which recasts, adapts, or transforms and existing work.
70
00:06:45.000 --> 00:06:48.399
The scope of derivative works right is
frankly not terribly well understood because it
71
00:06:48.639 --> 00:06:54.399
usually dovetails of copying and is typically
going to be one or the other.
72
00:06:55.319 --> 00:07:01.639
But that is a big part of
what is being claimed here. The resolution
73
00:07:01.759 --> 00:07:06.279
of it is really, I think
going to be a question of a is
74
00:07:06.319 --> 00:07:12.079
one of those activities happening. I
tend to think it is, and honestly,
75
00:07:12.480 --> 00:07:15.480
I'm not quite positive how much even
opening I'm sure, I'm sure we'll
76
00:07:15.519 --> 00:07:19.879
disputed, but I think that's probably
an easier case. But at least some
77
00:07:20.000 --> 00:07:24.759
copying is occurring, although there's details
of that that I'm sure some of my
78
00:07:24.800 --> 00:07:28.680
colleagues will discuss. But then fair
use is going to be a major issue
79
00:07:28.759 --> 00:07:35.160
there, particularly whether or not it's
transformative. What transformative means is really a
80
00:07:35.199 --> 00:07:41.839
hard question, but too recent.
The sort of formative case is the Acoff
81
00:07:42.160 --> 00:07:47.399
versus Campbell case about two live crews
Pretty woman, But I really think this
82
00:07:47.519 --> 00:07:53.800
case is going to be on fair
use, a conflict of the Google versus
83
00:07:53.800 --> 00:08:03.279
Oracle case, which found that the
Java program called Api eyes that their reimplementation
84
00:08:03.480 --> 00:08:09.160
or copying depending on your perspective,
I suppose into the droid operating system was
85
00:08:09.199 --> 00:08:13.920
fair use and transformative, and the
flip side Warhol versus Goldsmith holding that the
86
00:08:15.639 --> 00:08:20.680
Andy Warhol transformation into of Lyn Goldsmith's
photo of prints into an Andy Warhol silk
87
00:08:20.680 --> 00:08:28.439
screen of prints and then is licensing
to contiinast was not fair use and not
88
00:08:28.519 --> 00:08:33.720
transformative, So that's going to be
a bit of way. I suspect there
89
00:08:33.799 --> 00:08:39.279
was a motion to dismiss which is
pending, and that's all we have right
90
00:08:39.320 --> 00:08:41.799
now. We don't have, and
so fair use I don't think is ripe
91
00:08:41.879 --> 00:08:45.080
yet. I think you have to
wait for at leat summary judgment on that.
92
00:08:45.320 --> 00:08:50.559
But I kind of suspect if it
doesn't get dismissed and we have to
93
00:08:50.600 --> 00:08:56.679
get to that, we'll see on
that. You have to show direct infringement
94
00:08:56.720 --> 00:09:00.360
to get to any of the other
stuff except some of them more except for
95
00:09:00.879 --> 00:09:07.559
the last one for contributory infringement,
which is also alleged a lot of you
96
00:09:07.600 --> 00:09:09.799
us have to do, particularly with
Microsoft, which is not open AI,
97
00:09:11.759 --> 00:09:16.080
but as the complaint notes, is
kind of an alter ego, and we
98
00:09:16.159 --> 00:09:20.519
have all sorts of rights and open
AI, and we also own a lot
99
00:09:20.519 --> 00:09:26.600
of the IP. Contributory infringement is
defendant have knowledge of direction infringement and defendant
100
00:09:26.679 --> 00:09:37.519
materially contributing to that infringement from loaded
terms there knowledge knowledge you know question.
101
00:09:37.759 --> 00:09:41.600
We've had a conflict that happens a
lot in the context of the Internet and
102
00:09:41.639 --> 00:09:48.879
takedowns that whether knowing requires sort of
red flag or general knowledge. Where's a
103
00:09:48.919 --> 00:09:56.240
case of all the coximmunications and it
was very adequately flagging for users who repeat
104
00:09:56.279 --> 00:10:03.919
infringers. The details of that case
law are somehow twenty six years after the
105
00:10:03.960 --> 00:10:13.120
law was passed, still in flux
to a degree and material contribution. I
106
00:10:13.159 --> 00:10:16.240
think that that one is easier to
show here. I think it's going to
107
00:10:16.440 --> 00:10:18.440
hinge on knowledge, but I've been
wrong before. If that's the other key
108
00:10:18.480 --> 00:10:26.879
part vicarious liability. The case VET
I always teach when I talk about vicarious
109
00:10:26.919 --> 00:10:31.600
liability the case called Funavisia versus Cherry
Orchard, and it's a case VET held
110
00:10:31.600 --> 00:10:35.080
a out of fleet. I mean
it's contributory as well, but a case
111
00:10:35.080 --> 00:10:41.320
it held out of flea market that
was basically a hotbed of pirated media was
112
00:10:41.440 --> 00:10:45.759
engaging with vicarious liability because they had
a right to control the infringing activity.
113
00:10:45.799 --> 00:10:50.799
The infringing activity being the resale of
pirated discs installed at the flea market that
114
00:10:50.840 --> 00:10:54.960
were rented out and arriving a financial
commercial benefit from it, which was once
115
00:10:54.960 --> 00:11:00.519
again they were getting paid the fees
or restalls. Both of other I think
116
00:11:00.919 --> 00:11:07.919
are going to come out quite substantially. But other sort of pseudo IP issue
117
00:11:07.960 --> 00:11:11.440
I want to flag, if it's
really interesting, claims that New York Times
118
00:11:11.440 --> 00:11:18.919
owns Wirecutter, which is product recommendations, and they claim it is misappropriation because
119
00:11:18.960 --> 00:11:26.720
they are taking Wirecutter reviews and are
not getting affiliate payments. The large open
120
00:11:26.720 --> 00:11:31.720
AI are claiming this is preemptive by
a copyright lack which has at all legal
121
00:11:31.799 --> 00:11:37.840
or equitable rights that are within the
general scope of copyright are preempted. And
122
00:11:37.919 --> 00:11:41.639
on the other hand you have this
case is International News Service versus Associated Press,
123
00:11:41.679 --> 00:11:48.679
Supreme Court nineteen eighteen, holding the
misappropriation of hot news. In other
124
00:11:48.720 --> 00:11:52.600
words, news which was fresh out
of battle line the World War One was
125
00:11:52.799 --> 00:12:00.159
a misappropriation and a violation of common
law rights that was not preempted. The
126
00:12:00.200 --> 00:12:05.080
scope of VINS versus the AP nowadays
has been questioned for some cases have limited
127
00:12:05.080 --> 00:12:09.279
it, but it's not dead,
and so I think it's be interesting to
128
00:12:09.279 --> 00:12:13.600
see how that plays out. I
think, well, I'm looking forward to
129
00:12:13.759 --> 00:12:16.000
more of your questions. I think
we'll turn over Charles from more first head
130
00:12:16.039 --> 00:12:24.120
thoughts. All right, uh,
thanks thanks to me for that, you
131
00:12:24.159 --> 00:12:28.320
know, really excellent introduction to kind
of what the major issues are in this
132
00:12:28.440 --> 00:12:35.279
case and what the kind of key
doctrines of copyright and other law are that
133
00:12:35.360 --> 00:12:37.759
are at play. As you can
see, this case has a lot of
134
00:12:37.799 --> 00:12:39.879
things going on, and so you
know what I'll try to do is I'll
135
00:12:39.919 --> 00:12:43.399
try to number one, just go
through kind of what's been going on in
136
00:12:43.440 --> 00:12:46.600
the case specifically, so I'll talk
about particularly the motions to the SPISS that
137
00:12:46.639 --> 00:12:52.399
have been filed, and then take
a couple of guesses. As you mentioned,
138
00:12:52.679 --> 00:12:54.840
we don't know what the fair use
defense is going to look like,
139
00:12:54.879 --> 00:12:58.279
but I'll try to talk a little
bit about, you know what i'd expect,
140
00:12:58.159 --> 00:13:03.519
particularly, try to guess kind of
what I would imagine open AI would
141
00:13:03.559 --> 00:13:09.120
try to argue, and then maybe
leave with a couple of broader thoughts about
142
00:13:09.200 --> 00:13:15.639
kind of where this fits into the
larger debate over copyright and AI. So
143
00:13:15.960 --> 00:13:20.000
as far as the procedure of the
lawsuit goes, we started with a complaint
144
00:13:20.039 --> 00:13:24.919
in the lawsuit filed in December twenty
twenty three, and now we have on
145
00:13:24.960 --> 00:13:28.759
the table two motions to dismiss,
one from open Ai that was filed late
146
00:13:28.799 --> 00:13:35.159
February and one from Microsoft that was
filed about a week ago. These motions
147
00:13:35.159 --> 00:13:39.159
to dismiss don't go to the entirety
of the case, which is somewhat interesting.
148
00:13:39.200 --> 00:13:43.639
They only go to what the what
Microsoft and open Ai describe as sort
149
00:13:43.639 --> 00:13:48.080
of ancillary issues, and so the
ones that they talk about, in particular
150
00:13:48.200 --> 00:13:52.039
the contributory liability issue, they say
that there's a lack of knowledge, as
151
00:13:52.159 --> 00:13:54.879
Via mentioned, so they say that
that one should be dismissed. The common
152
00:13:54.960 --> 00:14:00.759
law misappropriation issue that's you mentioned this
hot news I in S is AP case
153
00:14:01.000 --> 00:14:03.639
kind of theory. They say that
that's preempted by the Copyright Act, that
154
00:14:03.720 --> 00:14:11.440
copyright protection being a federal law overrides
what the states, what state protection is
155
00:14:11.480 --> 00:14:13.559
given there, and they point to
know a number of cases that show that
156
00:14:13.600 --> 00:14:18.519
the I S versus AP doctrine is
kind of begrudgingly accepted at this point.
157
00:14:18.600 --> 00:14:22.919
That's kind of their argument for that. John mentioned also that there was this
158
00:14:22.000 --> 00:14:28.399
Digital Millennium Copyright Act count in the
complaint, and that one is actually somewhat
159
00:14:28.399 --> 00:14:33.320
interesting. So I'll just I'll go
into that just a little bit. Some
160
00:14:33.399 --> 00:14:37.399
of you may be familiar with the
Digital Millennium Copyright Act in terms of its
161
00:14:37.440 --> 00:14:43.279
anti circumvention provisions, the rules that
say that you're not allowed to kind of
162
00:14:43.320 --> 00:14:46.519
break digital rights management. This case
actually deals with a different part of the
163
00:14:46.600 --> 00:14:52.919
DMCA, Section twelve oh two,
which relates to copyright management information, so
164
00:14:52.039 --> 00:15:00.639
basically the inclusion of metadata like authors
or titles or copyright notices inside files,
165
00:15:00.679 --> 00:15:07.840
typically digital files. And so The
Times argues that in training these generative AI
166
00:15:09.039 --> 00:15:15.279
systems, Open AI and Microsoft removed
that information and as a result, violated
167
00:15:15.320 --> 00:15:18.320
section twelve h two. The difficulty
that they're going to face improving that,
168
00:15:18.559 --> 00:15:22.320
as Microsoft and open AI point out, is that in order to show a
169
00:15:22.399 --> 00:15:26.679
violation of the section, you have
to show what's called a double cienter requirement.
170
00:15:28.240 --> 00:15:31.559
Number one that's open AI knew that
it was removing and number two that
171
00:15:31.639 --> 00:15:35.919
it knew that the result of removal, or at least should have known that
172
00:15:35.960 --> 00:15:41.399
the result of removal would be further
infringement. And so open and Microsoft argue
173
00:15:41.480 --> 00:15:45.559
that number one, there's no efforden
is that this information was actually removed during
174
00:15:45.600 --> 00:15:48.559
the training process, but number two
that they won't be able to satisfy the
175
00:15:50.000 --> 00:15:54.279
knowledge requirements. They also raise a
time bar question. They say that there's
176
00:15:54.320 --> 00:16:02.440
a three year period look back period
that limits the extent that the copyright allegations
177
00:16:02.480 --> 00:16:06.080
can go back. That's actually a
case that's being considered by the Supreme Court
178
00:16:06.159 --> 00:16:10.399
right now. It was just argued
a couple of weeks ago, and so
179
00:16:10.799 --> 00:16:14.840
that's just another argument that they bring
up. But this again doesn't get to
180
00:16:14.919 --> 00:16:19.480
the substitutive questions that you mentioned,
the questions of whether or not there actually
181
00:16:19.720 --> 00:16:26.200
is copyright infringement in the training of
these systems. Using the New York Times
182
00:16:26.200 --> 00:16:33.360
and other articles, there are three
points that the Times identifies as where the
183
00:16:33.399 --> 00:16:37.639
infringement could occur. Number One,
they say that the collection of the articles
184
00:16:37.720 --> 00:16:41.960
to make the training data use to
train these AI systems, that was an
185
00:16:42.000 --> 00:16:45.720
infringement because you were making a lot
of copies in order to collect them.
186
00:16:45.120 --> 00:16:51.919
Second, they allege that the model
itself all of the data parameters I think
187
00:16:51.960 --> 00:16:59.480
one point seven trillion numbers that make
up the the GPT systems that's somewhere embedded
188
00:16:59.519 --> 00:17:03.879
in there is all of the information
necessary to replicate a lot of the articles,
189
00:17:03.879 --> 00:17:07.480
and therefore the model itself is a
potential infringement. And third, they
190
00:17:07.519 --> 00:17:12.119
say that when you use the model
in such a way that it generates that
191
00:17:12.559 --> 00:17:17.200
generates infringing content, that that use
is sort of a public performance. It
192
00:17:17.240 --> 00:17:23.160
allows you to get the information out
in order to show that these in order
193
00:17:23.200 --> 00:17:26.680
to show infringement, what the Times
would have to show is number one,
194
00:17:26.680 --> 00:17:30.799
that this is copyrightable subject matter.
There are some interesting questions there because you
195
00:17:30.839 --> 00:17:33.920
know, a lot of the information
that's being drawn is factual, and so
196
00:17:33.079 --> 00:17:37.799
maybe there'll be that issue that comes
out. Generally factual information is not considered
197
00:17:37.839 --> 00:17:41.000
copyrightable, but you know, like
I said, that's probably not going to
198
00:17:41.039 --> 00:17:45.440
be the lead argument. There also
are questions of what exactly counts as an
199
00:17:45.480 --> 00:17:48.839
infringing act. You know, is
something that's internally inside the model that nobody
200
00:17:48.880 --> 00:17:52.680
could actually see or understand, is
that an infringement. There's actually sort of
201
00:17:52.680 --> 00:17:56.119
an interesting question about that. But
again, the large issue that open ai
202
00:17:56.200 --> 00:18:00.720
and Microsoft intend to raise, in
fact, they say that they are they're
203
00:18:00.759 --> 00:18:06.839
actually very excited they say in their
in their motions of dismiss to litigate this
204
00:18:06.920 --> 00:18:11.799
issue is the fair use question,
assuming that it is an act of copying
205
00:18:11.880 --> 00:18:14.839
or it is a derivative work to
do all of those things I just mentioned,
206
00:18:15.400 --> 00:18:18.759
does the fair use doctrine permit it? And so courts have used the
207
00:18:18.799 --> 00:18:22.240
fair use doctrine a variety of situations, Google versus Oracle and the software context.
208
00:18:22.759 --> 00:18:27.480
On the Warhol case that was an
artistic use. On the Campical case
209
00:18:27.480 --> 00:18:30.240
that was parody. Oh, fairiuse
is sort of this jack of all trades
210
00:18:30.279 --> 00:18:36.200
doctrine ends up being used in all
sorts of places. In that line,
211
00:18:36.559 --> 00:18:38.759
there are a number of cases that
will help open AI quite a bit,
212
00:18:40.039 --> 00:18:44.920
although possibly to a limited extent.
There was a case over Google Images where
213
00:18:44.960 --> 00:18:48.039
Google had collected a bunch of images
and was displaying them using an image search
214
00:18:48.079 --> 00:18:53.720
engine. Courts said that because Google
had really downsampled them and they didn't really
215
00:18:53.759 --> 00:18:59.839
serve as replacement, that database of
images was fair use. There's a case
216
00:18:59.839 --> 00:19:04.160
called Eye Paradigms in which a company
made a plagiarism detection program and there was
217
00:19:04.160 --> 00:19:08.160
a question of whether or not the
inputs to the plagiarism detection program, which
218
00:19:08.200 --> 00:19:12.559
were basically the essays that were being
detected, whether or not that was an
219
00:19:12.599 --> 00:19:15.680
infringe And again the court said,
you know, this is sort of a
220
00:19:15.720 --> 00:19:22.039
new tool, the actual articles aren't
retrievable. Similarly, with the Google Books
221
00:19:22.119 --> 00:19:25.440
case, it was alleged that Google
scanning of a bunch of books to create
222
00:19:25.480 --> 00:19:30.880
the Google Book search engine was copyright
infringe and again a court said, I
223
00:19:30.880 --> 00:19:34.880
think the second Circuit said that this
was fair use on the grounds that the
224
00:19:36.039 --> 00:19:40.759
mass scanning of books to provide a
service that didn't really replicate the value of
225
00:19:40.759 --> 00:19:48.480
the books themselves was allowable and as
a result not a copyright infringement. Courts
226
00:19:48.559 --> 00:19:53.279
usually applied this four factor test in
which they look at the the nature of
227
00:19:53.319 --> 00:20:00.240
the copyrighted work, the nature of
the use, the purpose in character of
228
00:20:00.279 --> 00:20:03.920
the use, the amount that was
used. And finally, and probably this
229
00:20:03.039 --> 00:20:07.160
is the most important factor by many
measures, the economic impact on the market
230
00:20:07.200 --> 00:20:11.359
for the original copyrighted work. And
I think that that's going to be the
231
00:20:11.359 --> 00:20:15.640
most interesting one to follow in this
case, because, on the one hand,
232
00:20:15.079 --> 00:20:22.440
these are incredibly valuable systems. Right
Artificial intelligence has huge potential in terms
233
00:20:22.599 --> 00:20:30.559
of business uses, commercial uses,
uses for individual consumers. It can be
234
00:20:30.680 --> 00:20:37.480
used as the platform for many other
technologies. Is that part of the market
235
00:20:37.839 --> 00:20:42.200
that inheres in the copyright that The
New York Times has in all of its
236
00:20:42.279 --> 00:20:45.240
articles, or that a novelist has
in all the novels that are used in
237
00:20:45.279 --> 00:20:48.359
training. Right, the novelists or
the New York Times, they would say,
238
00:20:48.480 --> 00:20:52.200
yes, the point of our articles
it is to provide information, and
239
00:20:52.279 --> 00:20:57.079
that information is being used to provide
a valuable service through things like SHATGBT,
240
00:20:57.400 --> 00:21:03.039
and as a result, there should
be some cut of that open AI of
241
00:21:03.079 --> 00:21:04.960
course, would argue the other way
around. They would say, look,
242
00:21:06.000 --> 00:21:10.279
this is a completely different service.
It doesn't serve as a replacement for the
243
00:21:10.319 --> 00:21:15.359
original articles. It's transformative in the
way that Svia mentioned, and transformative something
244
00:21:15.400 --> 00:21:18.960
that the courts have really looked at, so that I think is going to
245
00:21:18.960 --> 00:21:22.400
be sort of the parameters of debate. We do have these cases about mass
246
00:21:22.440 --> 00:21:26.519
text and data mining which are different
from this case, but you know,
247
00:21:26.039 --> 00:21:32.319
provide some basis for understanding where fair
use goes and that transformativelopment and what effects
248
00:21:33.119 --> 00:21:38.160
the availability of these systems has on
those copyrighted works. I think is going
249
00:21:38.200 --> 00:21:47.279
to be really important and something really
to watch as this case progresses. All
250
00:21:47.359 --> 00:21:52.200
right, I think that makes it
my turn. So let me begin by
251
00:21:52.200 --> 00:21:56.640
saying thank you to the Federalist Society
for inviting me today, and Emily for
252
00:21:56.759 --> 00:22:00.799
organizing this panel, and of course
the On for his kind introduction, and
253
00:22:00.839 --> 00:22:06.440
my fellow panelists for their opening remarks. Let me note that my remarks are
254
00:22:06.519 --> 00:22:10.640
my own and do not necessarily reflect
the views of any client or employer.
255
00:22:11.440 --> 00:22:15.759
I want to begin by putting this
case on others like it into a broader
256
00:22:15.799 --> 00:22:19.440
perspective. Those of us who've been
working in copyright law and policy over the
257
00:22:19.440 --> 00:22:25.039
past thirty or so years, and
I'm afraid my gray hair gives that away,
258
00:22:26.480 --> 00:22:30.079
have seen history repeat itself over and
over. First, a new technology
259
00:22:30.079 --> 00:22:34.920
comes along and makes it easier than
ever to copyright, to obtain copyrighted works.
260
00:22:36.480 --> 00:22:41.759
Now, in the ideal scenario that
development is mutually beneficial, creators can
261
00:22:41.799 --> 00:22:47.680
reach new audiences expanded audiences more easily, and the widespread availability of creative works
262
00:22:47.759 --> 00:22:53.160
drives demand for the technology. Everybody
wins. But in practice, the operators
263
00:22:53.200 --> 00:22:59.160
of the technology in the past have
often made choices that allocate to themselves the
264
00:22:59.200 --> 00:23:03.039
lions share of the income. In
some cases, those choices have included willfully
265
00:23:03.119 --> 00:23:10.319
tolerating infringement on platforms, knowing full
well that creators, especially independent creators,
266
00:23:10.880 --> 00:23:15.680
like both the means and tools to
achieve meaningful vindication of their rights. So
267
00:23:15.839 --> 00:23:22.359
here we are with generative AI systems
built on large language models. It's deja
268
00:23:22.400 --> 00:23:26.680
vu all over again. Such systems
require massive volumes of works in order to
269
00:23:26.680 --> 00:23:33.240
be capable of producing the commercially valuable
outputs the designers seek to market. That
270
00:23:33.319 --> 00:23:37.400
fact is not in dispute, but
how those works are obtained is a commercial
271
00:23:37.480 --> 00:23:41.759
choice. There is nothing in the
nature of the technology that requires those works
272
00:23:41.799 --> 00:23:48.440
to be scraped without notice, without
authorization, or without compensation. Yet that's
273
00:23:48.480 --> 00:23:53.680
precisely what's happened. In the public
policy sphere, people willing to defend those
274
00:23:53.720 --> 00:23:59.880
decisions often try to create a false
dichotomy between the massive, unauthorized scraping and
275
00:24:00.039 --> 00:24:04.079
the existence of generative AI. The
reality is that there are companies that have
276
00:24:04.160 --> 00:24:11.279
built generative AI systems unlicensed materials.
Those that choose to do otherwise are not
277
00:24:11.359 --> 00:24:15.920
engaged in a crusade for the betterment
of humanity. They are commercial enterprises trying
278
00:24:15.920 --> 00:24:19.960
to avoid paying for critical inputs.
As the chairwoman of the Federal Trade Commission
279
00:24:21.000 --> 00:24:26.200
recently said very plainly, firms cannot
use claims of innovations as an excuse for
280
00:24:26.319 --> 00:24:32.279
law breaking. So let me turn
to the particular legal issues in this case,
281
00:24:32.359 --> 00:24:37.400
and I'm going to focus on the
direct copyright infringement issues. One would
282
00:24:37.440 --> 00:24:41.680
think that in a circumstance when computers
were and our programmed to crawl the Internet
283
00:24:41.720 --> 00:24:47.480
and copy literally billions of works,
the largest copying effort in history, that
284
00:24:47.559 --> 00:24:51.440
it would be beyond serious contention that
the reproduction right of those works has been
285
00:24:51.480 --> 00:24:56.599
implicated, And yet open AI and
others are trying to put that exact matter
286
00:24:56.640 --> 00:25:00.960
in dispute. Anyone who is even
a passing understanding of how computers operate knows
287
00:25:00.960 --> 00:25:04.920
that computers must make copies in order
to process what has been input into them,
288
00:25:06.960 --> 00:25:10.000
and the New York Times evidence shows
that by inputting a certain set of
289
00:25:10.000 --> 00:25:14.960
prompts, substantially, if not strikingly, similar copies of their original works will
290
00:25:15.000 --> 00:25:18.759
be output by open AI, So
it seems self evident that copies of the
291
00:25:18.799 --> 00:25:22.000
original works must be in the computer
memory in order for that to happen.
292
00:25:23.599 --> 00:25:29.279
Still, open AI argues to the
contrary, if the internal operation of the
293
00:25:29.279 --> 00:25:33.119
system were transparent to the public,
we would have real insight into the facts
294
00:25:33.119 --> 00:25:37.480
of how it operates. But despite
the name, it was given Open AI,
295
00:25:37.559 --> 00:25:41.240
like other generative AI systems, is
in fact locked up tight. Perhaps
296
00:25:41.279 --> 00:25:45.839
some of this will come out in
discovery, but in any event, I
297
00:25:45.880 --> 00:25:51.240
am deeply skeptical that there's any serious
argument other than that the unauthorized scraping of
298
00:25:51.279 --> 00:25:56.720
copyrighted works does implicate the reproduction right, which means, as my fellow panelists
299
00:25:56.720 --> 00:26:00.839
have already said, the real action
in this case be in the fair use
300
00:26:00.960 --> 00:26:04.920
argument. As has already been noted, the fair use assessment in the context
301
00:26:04.920 --> 00:26:11.240
of the first factor is likely to
involve consideration of whether opening eyes copying constitutes
302
00:26:11.279 --> 00:26:17.359
a transformative use. While the term
transformative has been part of copyright jurisprudence for
303
00:26:17.359 --> 00:26:21.799
a very long time, it was
given special significance by the Supreme Court decision
304
00:26:21.880 --> 00:26:26.400
in the two Life Crew case Campbell
Vis's A Cuff Rose to give the proper
305
00:26:26.400 --> 00:26:30.480
caption in nineteen ninety four. Since
that time, lower courts has struggled to
306
00:26:30.519 --> 00:26:36.480
apply this term, sometimes resulting in
extreme results, such as when Google's forbatim
307
00:26:36.519 --> 00:26:40.920
copying of tens of millions of books
was held to be highly transformative by the
308
00:26:40.920 --> 00:26:45.599
Second Circuit. Fortunately, the Supreme
Court had occasion to revisit this doctrine in
309
00:26:45.640 --> 00:26:52.000
twenty twenty two in Warhol Foundation versus
Goldsmith and articulated a much more reasonable and
310
00:26:52.039 --> 00:26:56.839
workable framework. So I think the
lower court fair use decisions that pre date
311
00:26:56.920 --> 00:27:03.599
Goldsmith are now of questionable applicability.
The first factor begins with a contrast between
312
00:27:03.680 --> 00:27:07.519
nonprofit use, which is favored,
and commercial use, which is disfavored.
313
00:27:07.200 --> 00:27:11.400
Prior to Goldsmith, some courts were
finding that a transformative use not only negated
314
00:27:11.400 --> 00:27:17.880
the commerciality but essentially overtook all the
other fair use factors as well. But
315
00:27:17.960 --> 00:27:21.240
a Goldsmith, the Supreme Court was
much more measured, holding the weight of
316
00:27:21.240 --> 00:27:26.319
commerciality against fair use can be lessened
by the degree to which the use is
317
00:27:26.319 --> 00:27:30.200
transformative, that is, has a
further purpose or different character. It's a
318
00:27:30.200 --> 00:27:37.119
sliding scale, not a Boolean analysis. So what constitutes transformative use? Some
319
00:27:37.200 --> 00:27:41.480
courts had gone so far in finding
any new element of the use to be
320
00:27:41.519 --> 00:27:45.960
transformative that many commentators wondered what if
anything, was left of the statutory right
321
00:27:47.200 --> 00:27:52.640
to authorize the creation of derivative works. As they mentioned, the Goldsmith Court
322
00:27:52.680 --> 00:27:56.720
wrote, to make transformative use of
an original must go beyond that required to
323
00:27:56.759 --> 00:28:03.039
qualify as a derivative use that has
a distinct purposes justified because it furthers the
324
00:28:03.039 --> 00:28:06.880
goal of copyright, namely to promote
the progress of science and the arts,
325
00:28:06.920 --> 00:28:14.519
without diminishing the incentive to create.
Quoting author's gildmersus Google, the court continued,
326
00:28:14.960 --> 00:28:18.799
the more the appropriator is using the
copied material for new transformative purposes,
327
00:28:19.079 --> 00:28:23.000
the less likely it is that the
appropriation will serve as a substitute for the
328
00:28:23.039 --> 00:28:29.200
original or its plausible derivatives. So
we see that substitution is a key part
329
00:28:29.200 --> 00:28:33.079
of the analysis, one that has
the power to disqualify use from transformative status.
330
00:28:33.880 --> 00:28:40.279
And the Times has shown that OpenAI
is capable of preproducing substantially similar copies
331
00:28:40.319 --> 00:28:45.000
of its works. That's powerful evidence
on the substitution effect that AI will have
332
00:28:45.039 --> 00:28:52.640
to contend with when this case moves
forward. The Supreme Court in Goldsmith summarized
333
00:28:52.640 --> 00:28:56.480
its first factor analysis with the passage
quote. The first factor considers whether the
334
00:28:56.599 --> 00:29:00.519
use of a copyrighted work has a
further purpose or different character, which is
335
00:29:00.559 --> 00:29:04.519
a matter of degree, and the
degree of difference must be balanced against the
336
00:29:04.559 --> 00:29:10.079
commercial nature of the use quote quote. Those who are sympathetic to the AI
337
00:29:10.200 --> 00:29:12.359
side will point to the creation of
new works by the use of the resulting
338
00:29:12.400 --> 00:29:18.680
system. But this misses a key
step. Generative AI systems do not create
339
00:29:18.720 --> 00:29:22.720
images or text or anything on their
own initiative. Human interaction is required in
340
00:29:22.759 --> 00:29:27.119
the form of prompts, and without
that human interaction, the resulting product would
341
00:29:27.160 --> 00:29:33.319
have no protectable creative expression. So
it isn't the AI system that generates new
342
00:29:33.319 --> 00:29:37.839
works. It is the human users. And it's been good law for a
343
00:29:37.960 --> 00:29:41.599
very long time that the copyer of
works may not stand in the shoes of
344
00:29:41.640 --> 00:29:47.319
the users of those copies to justify
the initial copying. That, of course,
345
00:29:47.359 --> 00:29:51.640
is Michigan Document Services versus Princeton University
Press. We can talk about that
346
00:29:51.680 --> 00:29:55.759
if it comes up in the Q
and A. As for the remaining fair
347
00:29:55.880 --> 00:29:59.319
use factors, it seems to me
they all favor the copyright owner in varying
348
00:29:59.359 --> 00:30:03.519
degrees. The works were largely creative, although perhaps some news articles marginally less
349
00:30:03.559 --> 00:30:08.119
so still I think it cleared.
The second factor favors the Times. The
350
00:30:08.279 --> 00:30:14.119
entirety of every work was copied,
so the third factor is strongly against fair
351
00:30:14.240 --> 00:30:17.920
use, and the fourth factor,
harmed to the current or potential market,
352
00:30:18.039 --> 00:30:22.799
very strongly favors the Times. Again, substitution is probably the strongest evidence of
353
00:30:22.839 --> 00:30:26.079
harm, followed closely by the existence
of a licensing market for the same use,
354
00:30:26.799 --> 00:30:30.799
both of which we have here.
So if open AI is going to
355
00:30:30.799 --> 00:30:33.920
prevail on fair use, it seems
to me they will need a very strong
356
00:30:33.960 --> 00:30:37.559
finding on the first factor in order
to overcome not only the commercial nature of
357
00:30:37.599 --> 00:30:41.200
the use, but all the other
factors as well. On the law,
358
00:30:41.240 --> 00:30:45.759
this seems unlikely. I can only
imagine a finding of fair use frankly if
359
00:30:45.759 --> 00:30:49.920
the court is taken in by the
cool factor arguments. But it would be
360
00:30:49.960 --> 00:30:55.960
a travesty for creators works to lose
essentially all protection in the aihe and it
361
00:30:56.000 --> 00:31:00.000
would certainly harm the incentive to create, which is what the Goldsmith Court kept
362
00:31:00.079 --> 00:31:07.680
certainly in mind. Thanks, we
have a couple of questions before I go
363
00:31:07.799 --> 00:31:11.079
there. I think you addressed this, Steve, and that is the substitution,
364
00:31:11.319 --> 00:31:18.839
the competitiveness in the market, and
in particular that particular fact which the
365
00:31:19.160 --> 00:31:23.640
complaint I believe goes to some links
to try to create a competitive market.
366
00:31:23.920 --> 00:31:30.599
And how does that get them out
of or diminish the Google Books copying case.
367
00:31:33.599 --> 00:31:37.720
Well, I'm not sure the degree
to which Google Books is still good
368
00:31:37.799 --> 00:31:42.559
law after Goldsmith. Not the Goldsmith
overturned that decision in any way, but
369
00:31:42.799 --> 00:31:51.440
the analysis was so different that it's
hard to make that case any longer.
370
00:31:51.480 --> 00:31:57.880
That the Google books really is a
useful president when you look at this at
371
00:31:57.920 --> 00:32:04.400
ai generative systems in the most general
way and kind of departing for a moment
372
00:32:04.440 --> 00:32:07.680
from the particulars of the law,
but in a matter of social policy,
373
00:32:08.359 --> 00:32:17.920
we have systems where people's work product
was copied without their knowledge, permission,
374
00:32:19.640 --> 00:32:25.880
or compensation, and it's being used
to design and implement technology that will then
375
00:32:25.960 --> 00:32:30.839
put many of those people out of
business entirely or substantially reduce their income by
376
00:32:30.880 --> 00:32:37.200
competing directly with them. There's also
questions about, you know, well,
377
00:32:37.200 --> 00:32:40.839
why did you copy this work versus
that work? Does any individual work have
378
00:32:42.279 --> 00:32:49.559
particular value? Works were chosen because
they are valuable for what they are,
379
00:32:50.559 --> 00:32:54.599
and so there is value there.
And I think at the highest level of
380
00:32:55.079 --> 00:33:02.519
public policy analysis, value was taken
and value has been created in companies that
381
00:33:02.559 --> 00:33:12.160
are already valuing themselves at over trillion
dollars, and there's got to be a
382
00:33:12.200 --> 00:33:19.960
way to compensate the people whose works
were taken to build those systems. It's
383
00:33:20.000 --> 00:33:23.319
beyond this panel talk about exactly how
that works, going back to the particulars
384
00:33:23.359 --> 00:33:28.839
of the copyright law, the fact
that this goes back to get to the
385
00:33:28.880 --> 00:33:34.160
reproduction right. It seems self evident
to me that copies were made, and
386
00:33:34.200 --> 00:33:39.319
I've just made the fair use analysis
that I think is appropriate. So I'm
387
00:33:39.640 --> 00:33:45.039
I wouldn't be surprised to see the
times prevail in this case. But time
388
00:33:45.079 --> 00:33:52.160
will tell. So we have a
question that goes to ask quickly on I'm
389
00:33:52.200 --> 00:33:55.960
sorry, go ahead, Yes,
I just wanted to say I think it's
390
00:33:55.960 --> 00:34:00.960
slight different tack on Google books.
I think it's very much of a piece
391
00:34:01.000 --> 00:34:04.720
of Google image Search, and I
think it has to be read as fact
392
00:34:04.759 --> 00:34:07.480
specific. I mean, whether or
not it's good law after Warhol, I
393
00:34:07.519 --> 00:34:14.079
don't know, but I really think
that the focus on stipid view and directing
394
00:34:14.119 --> 00:34:19.079
people to booksellers and gootle books,
and the focus on directing people to websites
395
00:34:19.079 --> 00:34:23.039
and Google image search with with key
too decisions in terms of a market effect
396
00:34:23.079 --> 00:34:27.679
and transformation. If the use was
not a substitute in any way, at
397
00:34:27.760 --> 00:34:31.320
least in the courts view, but
rather a channeling use of sorts and so
398
00:34:31.440 --> 00:34:37.039
I do think it has to be
a difference there where there's no channeling whatsoever
399
00:34:37.199 --> 00:34:40.679
here. But like Steve said,
we'll see, well so if I can
400
00:34:40.800 --> 00:34:45.199
just kind of jump in a very
quickly. So one of the arguments that
401
00:34:45.480 --> 00:34:49.719
is sort of previewed in the in
the motions of the Smiths is this question
402
00:34:49.840 --> 00:34:58.440
of to what extent the actual substitutional
use is actually relevant. And so open
403
00:34:58.480 --> 00:35:02.239
Aye and Microsoft both point exhibit jay
of the complaint, which is where you
404
00:35:02.280 --> 00:35:07.320
get sort of the original duplication or
memorization as they call it, in the
405
00:35:07.320 --> 00:35:14.079
industry of articles. And as those
examples show, in order to get a
406
00:35:14.519 --> 00:35:19.559
in order to prompt chatterp to produce
a New York Times article, you first
407
00:35:19.559 --> 00:35:22.280
have to give it the first half
the article, essentially. And so their
408
00:35:22.400 --> 00:35:24.960
argument then is that at that point, if you already have the first half
409
00:35:24.960 --> 00:35:28.679
of the article, it's not much
of a substitution to say that you get
410
00:35:28.679 --> 00:35:30.599
the second half, especially when the
second half is available in other places on
411
00:35:30.639 --> 00:35:35.960
the Internet. And so that's an
interesting argument. I haven't heard that before,
412
00:35:36.000 --> 00:35:40.199
and it's sort of unique to this
sort of situation, but I think
413
00:35:40.199 --> 00:35:44.360
that what it does mean is that
it's going to be a little more complicated,
414
00:35:44.360 --> 00:35:49.360
I think to prove the substitution element
compared to some of the other cases.
415
00:35:49.400 --> 00:35:52.199
That doesn't mean that, you know, I think that's it's a slamm
416
00:35:52.199 --> 00:35:53.960
don win for them. It certainly
is the case that you can get a
417
00:35:54.039 --> 00:35:59.880
large chunk of copyright material. It's
just that it's not the normal way that
418
00:36:00.039 --> 00:36:05.280
one would expect to retrieve it.
There's a lot of focus on that this
419
00:36:05.440 --> 00:36:12.079
is a bug rather they want it
would legally shouldn't be significant, but has
420
00:36:12.119 --> 00:36:15.840
come up before, so who knows. Yeah, my understanding is that there
421
00:36:15.880 --> 00:36:22.079
also is an amount of research on
just how to stop at a computer level
422
00:36:22.360 --> 00:36:23.599
of the sort of thing from happening. You know, is there a way
423
00:36:23.599 --> 00:36:28.400
that they can just put filters at
the end, or can they use some
424
00:36:28.440 --> 00:36:31.840
sort of special training processes. That
is sort of an interesting research question that
425
00:36:31.880 --> 00:36:35.400
I know is open. I think
that the difficulty is that a lot of
426
00:36:35.440 --> 00:36:39.719
times articles are available online and slightly
different variations, like somebody might admit a
427
00:36:39.719 --> 00:36:46.039
paragraph, and that's why it's hard
for it's hard to simply detect identical copies.
428
00:36:46.800 --> 00:36:49.719
But it's sort of interesting that,
you know, there may be a
429
00:36:49.800 --> 00:36:52.920
technical solution in addition to the legal
approach, and there's a question of which
430
00:36:52.960 --> 00:36:55.440
one you think should lead. Right. On the one hand, maybe we
431
00:36:55.480 --> 00:36:59.719
want to say that the law encourages
people to develop these sorts of technologies that
432
00:37:00.000 --> 00:37:02.840
on void duplication. On the other
hand, maybe we say, maybe we're
433
00:37:02.840 --> 00:37:07.239
worried that by imposing the hammer of
the law too early, we prevent people
434
00:37:07.280 --> 00:37:13.360
from coming up with these sorts of
technologies that can satisfy better middle grounds,
435
00:37:13.400 --> 00:37:15.760
and we end up, you know, one of the one of the remedies
436
00:37:15.800 --> 00:37:20.960
that's asked for is destruction of chat
GPT, and so there's you know,
437
00:37:21.000 --> 00:37:23.840
there is a question of whether or
not it would be premature to be applying
438
00:37:23.880 --> 00:37:29.480
this before we really know what the
sort of opportunities the technology ultimately look like.
439
00:37:30.599 --> 00:37:34.119
So that's what I was referencing earlier
when I said that some people want
440
00:37:34.159 --> 00:37:37.800
to look at this as copyright versus
AI, and you can't have both.
441
00:37:38.880 --> 00:37:45.639
Had had open Ai chosen to license
the materially used, that no one would
442
00:37:45.679 --> 00:37:51.199
be asking to destroy chat GPT,
nor is anyone asking to destroy AI systems
443
00:37:51.239 --> 00:37:57.239
that were built on licensed material.
So putting that aside for a moment,
444
00:37:57.880 --> 00:38:02.559
the substitute effect had several layers to
it. First and foremost, the fact
445
00:38:02.559 --> 00:38:09.840
that you can get that material out
of chat GPT is strong evidence that the
446
00:38:10.119 --> 00:38:15.000
copyrightable material is somewhere in the chat
GPT system, the open AI system,
447
00:38:15.320 --> 00:38:22.280
and therefore the reproduction right is implicated. Then on the question of whether the
448
00:38:22.320 --> 00:38:29.679
output is infringing and the substitution effect. There we kind of have two or
449
00:38:29.719 --> 00:38:37.159
three different layers in traditional copyright analysis. Obviously, substantial similarity is an access
450
00:38:37.519 --> 00:38:46.960
are the two problems to proving copying
and infringing reproduction. So the fact that
451
00:38:47.000 --> 00:38:53.920
it may be difficult or unusual to
get chat GPT to put that out doesn't
452
00:38:53.960 --> 00:38:58.599
mean it's less infringing. It just
means it may take a little more work,
453
00:38:58.639 --> 00:39:05.400
but it's still infringing. More generally, more, there's a question about
454
00:39:05.920 --> 00:39:15.239
the degree to which these generative AI
systems will overall all diminish demand for the
455
00:39:15.280 --> 00:39:21.400
copyrighted works produced by the people who
produce the works that were used to build
456
00:39:21.440 --> 00:39:29.039
the system, and that admittedly departs
somewhat from a direct infringement analysis to more
457
00:39:29.079 --> 00:39:34.840
public policy question of is it right
that the people whose works were taken without
458
00:39:34.920 --> 00:39:38.800
notice, without permission, and without
compensation should be told, sorry, you're
459
00:39:38.840 --> 00:39:45.199
out of lock, and these new
systems that we built with all the things
460
00:39:45.239 --> 00:39:53.719
we took from you are now going
to put you out of business. We
461
00:39:53.760 --> 00:40:00.320
have a question from the audience concerning
generally to the panelists together, and the
462
00:40:00.400 --> 00:40:07.360
specific question is relevance of sega versus
accolade and the concept of intermediate cop copying.
463
00:40:08.360 --> 00:40:14.559
Can any of you guys speak to
that? Yeah, so this is
464
00:40:14.719 --> 00:40:19.800
also I was mentioning that there is
this sort of question of whether or not
465
00:40:20.239 --> 00:40:27.000
sort of like purely internal use would
count as an infringement. And there are
466
00:40:27.000 --> 00:40:31.639
a couple of cases that say that
at least that's an interesting question. So
467
00:40:30.400 --> 00:40:37.639
so the say atholate case and the
reverse engineering cases they raise that question.
468
00:40:37.639 --> 00:40:42.760
I think you probably remember this is
the cartoon network case, right, which
469
00:40:42.800 --> 00:40:46.639
talks about transitory copying. So you
have a number of you have a number
470
00:40:46.639 --> 00:40:52.480
of cases that support the sort of
idea that maybe like internal copying or copying
471
00:40:52.519 --> 00:41:00.719
that is precedent to some other public
use might not receive the same analogy assists.
472
00:41:00.480 --> 00:41:04.960
And so that's obviously going to be
relevant here because you know to the
473
00:41:05.039 --> 00:41:07.920
extent that the allegation is that the
copying is inside the model. The model
474
00:41:07.960 --> 00:41:14.239
is not something anybody can see,
and so that might fall within those doctrines.
475
00:41:14.239 --> 00:41:17.159
Similarly, the training process, again
that's not something anybody actually sees,
476
00:41:17.719 --> 00:41:23.639
and as a result, those aren't
those potentially aren't necessarily the sorts of things
477
00:41:23.639 --> 00:41:30.159
that are of concern to copyright law. In view of those cases, it's
478
00:41:30.199 --> 00:41:32.119
really kind of you know, what's
going on the output end that has the
479
00:41:32.119 --> 00:41:36.679
commercial impact. That's where the fair
use question is going to really come up.
480
00:41:36.760 --> 00:41:38.760
That's where the market effect is going
to come up. That's where transformativeness
481
00:41:38.840 --> 00:41:44.079
is going to come out. Yeah, that that that's that that's at least
482
00:41:44.079 --> 00:41:50.199
part of kind of where where those
cases fit in. Go ahead, see
483
00:41:50.239 --> 00:41:52.960
I will say, I mean my
view of both Accolae case and Cartoon Network
484
00:41:53.039 --> 00:41:59.800
that they boiled down to if you're
doing some internal copying for what otherwise illegally
485
00:42:00.199 --> 00:42:02.880
and non copying use. Now it
will say where Cartoon Network with plenty of
486
00:42:02.920 --> 00:42:08.840
copying, but it was all licensed
and and in Siga xacol it it's interoperability
487
00:42:08.840 --> 00:42:14.760
copying and just that you could reverse
engineer the interface for the game console.
488
00:42:15.320 --> 00:42:17.800
I think those cases basically boiled down
to, well, yeah, if you're
489
00:42:17.840 --> 00:42:22.119
if you're if not actually infringing in
any way. It's just a way of
490
00:42:22.119 --> 00:42:27.239
getting to that conclusion. And with
the other words, totally legal intermediate copying
491
00:42:27.280 --> 00:42:29.480
is okay. But I don't think
that's the case here. The whole problem
492
00:42:29.559 --> 00:42:34.639
if it chat GPT is spitting out
lots of copyrighted content and it's copyright content
493
00:42:34.679 --> 00:42:37.119
in and out, I think that's
the problem. Well, I mean that
494
00:42:37.199 --> 00:42:38.280
that's sort of the question of the
case, right, if it turns out
495
00:42:38.280 --> 00:42:40.800
that that is fair use. Right, if it turns out that, you
496
00:42:40.800 --> 00:42:44.519
know, they accept the argument that
this is a bug and you know,
497
00:42:44.880 --> 00:42:47.079
you know deminimus spitting out of exact
copy. You know, I don't know
498
00:42:47.199 --> 00:42:51.280
exactly how the court would rule,
but they say that that's okay, then
499
00:42:51.320 --> 00:42:54.000
the precedent copying should be okay as
well. We don't want we don't we
500
00:42:54.000 --> 00:42:58.960
don't want basically copyright law to be
focused on trivialities in a sense. So
501
00:42:59.280 --> 00:43:01.840
you know, kind of the question
is it's sort of a standardfall together situation
502
00:43:02.079 --> 00:43:08.440
where if the outputs end up being
problematic not not protected under fair use,
503
00:43:08.800 --> 00:43:13.480
then the whole thing probably wouldn't be
fair use either. But if it is
504
00:43:13.639 --> 00:43:17.440
then chances are the earlier copying is
okay as well in view of those presents.
505
00:43:19.400 --> 00:43:22.559
I think that's a little bit too
narrow in one respect, Sega versus
506
00:43:22.559 --> 00:43:30.199
Accolade was predicated on the fact the
ruling was predicated on the fact that the
507
00:43:30.239 --> 00:43:36.000
resulting products that Accolade wanted to create
were not competing with what they copied in
508
00:43:36.079 --> 00:43:42.599
any way, not direct substitutes or
even trying to create competing products as alternatives,
509
00:43:43.159 --> 00:43:45.880
but rather games rather as oppose to
the operating system, which is what
510
00:43:45.920 --> 00:43:53.079
they copied, games that would simply
work on that operating system. Whether generative
511
00:43:53.119 --> 00:44:01.719
AI systems are outputting substantially similar material
to what was copy or even just generally
512
00:44:01.800 --> 00:44:12.000
competing with creators, it's much more
of a market effect than we had in
513
00:44:12.039 --> 00:44:17.320
any way in Sega Versus Accolade,
and I have to on the Cartoon Network
514
00:44:17.320 --> 00:44:22.400
case. I don't really see much
of an analogy there to any part of
515
00:44:22.440 --> 00:44:30.039
this case that dealt with you know, it was all three aspects of the
516
00:44:30.079 --> 00:44:39.039
Cartoon Network case were and remain outlying
outlier decisions on both the temporary copy issue
517
00:44:39.360 --> 00:44:45.840
and the public performance issue, as
well as the authorizing the copy or the
518
00:44:50.000 --> 00:44:53.400
I can't think of the word anyway, none of those if this is not
519
00:44:53.480 --> 00:44:58.800
about public performance, this is not
about temporary copies. Those copies are in
520
00:44:58.840 --> 00:45:00.960
there for a long period of time. I think that you actually doualize that
521
00:45:00.960 --> 00:45:04.920
as a public performance. Interestingly,
I'm not exactly sure why, but I
522
00:45:04.920 --> 00:45:08.039
think that was actually the place.
Well, yes, but it's not the
523
00:45:08.039 --> 00:45:13.800
same sort of issue where oh,
there's only one copy versus ten thousand copies,
524
00:45:14.360 --> 00:45:20.559
And ironically the ten thousand copies is
not a public performance, said Cartoon
525
00:45:20.559 --> 00:45:24.199
Network. In the event, of
all the holdings of Cartoon Network, that
526
00:45:24.280 --> 00:45:30.239
element is surely the weakest. After
the Supreme Court decision in Areo, which,
527
00:45:30.280 --> 00:45:37.079
while not explicitly overruling Cartoon Network,
articulated very clearly an analysis at odds
528
00:45:37.119 --> 00:45:42.480
with the Second Circuits analysis of public
performance in Cartoon Network. I don't see
529
00:45:42.480 --> 00:45:52.880
any way that that's good law.
Do any of you have a view of
530
00:45:53.320 --> 00:46:00.679
the likelihood of a licensing solution to
this or do you think it's going to
531
00:46:00.760 --> 00:46:06.960
resolve on by the court Supreme Court? Probably I think the heavily licensed I
532
00:46:07.000 --> 00:46:09.119
think that I mean not saying I
don't want to say inevitably, you never
533
00:46:09.199 --> 00:46:13.719
know this could be litigated for a
dozen years. But they'll note in the
534
00:46:13.840 --> 00:46:20.000
complaint that New York Times is when
was heavily used sources for open AI,
535
00:46:20.960 --> 00:46:27.559
but other sources, notably Associated Press
or others mentioned in the complaint, had
536
00:46:29.239 --> 00:46:31.639
been paid for licensing. So I
think it's there, But I think,
537
00:46:34.719 --> 00:46:37.280
well, I mean, the old
line is that litigation is negotiation by other
538
00:46:37.360 --> 00:46:40.039
means. I'm not sure it's always
true, but I think it is true
539
00:46:40.039 --> 00:46:45.360
here. Yeah, I mean I
think that you know, there there are,
540
00:46:45.679 --> 00:46:49.320
as Steve mentioned, there are a
number of companies that are out there
541
00:46:49.360 --> 00:46:52.159
developing generative systems that are that are
that are based on that are that are
542
00:46:52.199 --> 00:46:58.320
based on licensing. The difficulty,
of course, which is which I'm sure
543
00:46:58.320 --> 00:47:00.400
opening I would be happy point to
on believe that they point to this in
544
00:47:00.519 --> 00:47:06.599
some of their papers, is that
a lot of the reason why these systems
545
00:47:06.679 --> 00:47:12.239
work particularly well is just the breadth
of content, the fact that they have
546
00:47:13.840 --> 00:47:20.000
large quantities, tremendously large quantities of
data to work off of, which would
547
00:47:20.039 --> 00:47:22.559
be substantially harder to obtain by a
licensing. And so, you know,
548
00:47:22.599 --> 00:47:29.719
while I think it's correct that this
isn't a question of just AI versus licensing
549
00:47:29.760 --> 00:47:32.639
that you that you can have both
at the same time. You aren't changing
550
00:47:32.639 --> 00:47:37.320
the parameters of what the technological development
looks like, right, and that could
551
00:47:37.400 --> 00:47:39.519
be a good thing. I'm not
necessarily saying that it's a bad thing if
552
00:47:39.519 --> 00:47:46.119
computer scientists are forced to work with
a different set of legal parameters. There
553
00:47:46.159 --> 00:47:52.440
could be value in specifically working with
licensed copies. But one thing to keep
554
00:47:52.480 --> 00:47:57.679
in mind is that there is also
a pretty good chunk of poly domain information
555
00:47:57.719 --> 00:48:02.000
that's available that had been the basis
for training of artificial intelligence for many years,
556
00:48:02.239 --> 00:48:05.960
and one of the reasons that companies
have moved away from that is that
557
00:48:05.960 --> 00:48:10.639
it produced very strange systems. A
lot of that material is hundreds of years
558
00:48:10.639 --> 00:48:16.679
old material that's sufficiently out of copyright, or material that is very strange to
559
00:48:16.800 --> 00:48:23.360
use as training data. The en
run email data base was a common one
560
00:48:23.400 --> 00:48:28.280
that was used, and as people
have pointed out, that ended up producing
561
00:48:28.400 --> 00:48:34.800
systems that incorporated all sorts of very
strange biases and so's you know. I
562
00:48:34.840 --> 00:48:37.400
think that it's on the one hand, yeah, you can't have a world
563
00:48:37.480 --> 00:48:43.480
in which licensing is prominent. You
do end up with a different technological environment,
564
00:48:43.639 --> 00:48:47.679
and I think it's it's an interesting
question of what that environment looks like
565
00:48:47.880 --> 00:48:51.440
and how you really feel about it, what it does to the true actually
566
00:48:51.440 --> 00:48:54.280
of the technology. Well, Charles, I think you just made a great
567
00:48:54.320 --> 00:48:59.159
case for the value of the copyrighted
works that were copying in order to build
568
00:48:59.480 --> 00:49:05.320
generative AI systems. Does it take
a little longer? Might it cost more
569
00:49:05.920 --> 00:49:10.199
to compensate creators for using their works
to build your system? Well? Yes,
570
00:49:10.280 --> 00:49:15.960
of course, the same way that
it would take a lot longer if
571
00:49:16.119 --> 00:49:21.800
the AI companies didn't have Nvidia chips
because they didn't want to pay for them,
572
00:49:22.519 --> 00:49:25.960
their processing capacity would be greatly diminished. But they need them and they
573
00:49:27.000 --> 00:49:30.920
want them, so they pay for
them. And I would suggest that the
574
00:49:30.039 --> 00:49:38.360
social aspect of this of telling creators
you're the one input into generative AI that
575
00:49:38.440 --> 00:49:43.079
no one has to pay for.
The coders they get paid, developers,
576
00:49:43.840 --> 00:49:49.559
the chip makers they get paid.
The power utilities that provide all that electricity
577
00:49:49.599 --> 00:49:54.760
to the data farms they get paid. But the creators, sorry, you're
578
00:49:54.840 --> 00:50:00.039
out of lock. That I think
there's an it looked at me in in
579
00:50:00.119 --> 00:50:05.559
that context. It's a pretty difficult
argument to sustain, but I think that
580
00:50:05.760 --> 00:50:12.960
generally correct this idea that if there's
some value out there, then possibly we
581
00:50:13.000 --> 00:50:15.199
want to make sure that there's compensation
with the value. The difficulty here,
582
00:50:15.239 --> 00:50:19.880
of course, is that copyright doesn't
cover every possible value of a work.
583
00:50:20.400 --> 00:50:24.159
Right, you have factual databases which
simply have no copyright protection because facts are
584
00:50:24.159 --> 00:50:29.880
not copyrightable, and so there is
sort of this larger question of, you
585
00:50:29.920 --> 00:50:31.960
know, how do you deal with
the problem value. Similarly, a lot
586
00:50:32.000 --> 00:50:37.039
of these systems use private data,
right, they use individual personal data.
587
00:50:37.079 --> 00:50:40.039
That's data that's not copyright protectable at
all, right, and that is value.
588
00:50:40.119 --> 00:50:43.599
So you know, one way you
could solve as by trying to create
589
00:50:43.599 --> 00:50:46.400
some sort of property right in private
information, but that actually causes a lot
590
00:50:46.440 --> 00:50:52.119
of other problems, as you know, a fair amount of scholarship that's out
591
00:50:52.159 --> 00:50:57.679
there. So I think that you
want to separate out the questions of how
592
00:50:57.679 --> 00:51:04.280
do you deal with the value propose
of information from the copyright questions, and
593
00:51:04.880 --> 00:51:07.840
if you're talking about just sort of
the value propositions, maybe once you're really
594
00:51:07.840 --> 00:51:10.320
asking for some sort of regulatory agency, which you know, we can debate
595
00:51:10.360 --> 00:51:15.840
that, but then you also have
to have the second question of is copyright
596
00:51:15.880 --> 00:51:22.880
the correct vehicle for doing that given
that copyright is historically limited to certain things
597
00:51:22.960 --> 00:51:27.159
which are a little bit different from
what exactly is going on with artificial intelligence
598
00:51:27.159 --> 00:51:32.239
systems. The fact versus the idea
versus expression, the fact versus expression dichotomies,
599
00:51:34.119 --> 00:51:37.239
and of course the fair use doctrine, these all play into kind of
600
00:51:37.239 --> 00:51:43.039
what the boundaries of copyright law are
that don't make value exactly aligned with the
601
00:51:43.119 --> 00:51:46.559
with the rights provide unfederal law.
I do agree there are elements of these
602
00:51:46.599 --> 00:51:52.800
issues that go beyond copyright law,
and you're seeing all sorts of proposals in
603
00:51:52.800 --> 00:52:01.239
those veins, from privacy protections to
state and federal legion on issues of publicity
604
00:52:01.320 --> 00:52:07.760
rights, narrowing it to the copyright
issues, And because that is the focus
605
00:52:07.800 --> 00:52:14.760
of this panel, I do think
that of course collections of data can be
606
00:52:14.840 --> 00:52:19.199
protectable as compilations by their selection,
arrangement coordination, as I'm sure you know.
607
00:52:20.639 --> 00:52:24.320
And then many of the works that
were taken were taken precisely because they
608
00:52:24.360 --> 00:52:36.079
are valuable expression of contemporary American society, and we don't want people aren't going
609
00:52:36.119 --> 00:52:40.079
to pay for AI systems that speak
in old English or forsooth or what have
610
00:52:40.199 --> 00:52:46.679
you. So when I speak of
the value. I'm speaking both generally beyond
611
00:52:46.719 --> 00:52:52.400
copyright, but also in the context
of the fourth factor, where there's harm
612
00:52:52.480 --> 00:53:00.960
to the current and potential market or
for the works that were copied. I
613
00:53:00.000 --> 00:53:06.679
have a question from the audience's question
of possible to the panel, I'm tired
614
00:53:07.119 --> 00:53:12.800
is tirty? Is is there a
place in this discussion for Sony safe harbor?
615
00:53:15.239 --> 00:53:21.840
Yeah? For Sony you said,
right, yes, yeah, So
616
00:53:21.960 --> 00:53:24.920
I actually pulled along from Sony.
Everyone reads Sony for this first half,
617
00:53:24.960 --> 00:53:31.280
which is capable of substance of you
know, subject of commercially significant non infringing
618
00:53:31.440 --> 00:53:37.880
uses. Of course, in Sony
the court held that time shifting was okay,
619
00:53:37.159 --> 00:53:40.920
but librarying was probably not. And
everyone forgets to the second half.
620
00:53:44.320 --> 00:53:46.840
There's room for it, but I
don't think it gets you there in this
621
00:53:47.000 --> 00:53:53.199
case. I think that clearly open
Aye had well. Versually is an interesting
622
00:53:53.280 --> 00:53:57.599
question. It's a capable of non
infringing uses, I mean through it depends
623
00:53:57.639 --> 00:54:04.679
how you define it. What what
point is non fringing? And I mean
624
00:54:05.079 --> 00:54:08.000
actually a lot of the most of
dismiss was on sexual limitations, but a
625
00:54:08.039 --> 00:54:13.519
hepstink discovery rule was going to bring
in the ingestion of a lot of that
626
00:54:13.599 --> 00:54:16.280
as well. Potentially of course,
then you know we might be waiting to
627
00:54:16.280 --> 00:54:21.719
find out what win our shop versus
Neely holds, or if anything is held
628
00:54:21.760 --> 00:54:27.320
in that case. This is a
three year sexual limitations case. Yes,
629
00:54:27.639 --> 00:54:31.920
so on the Sony case. This
is for those of you who weren't familiar
630
00:54:31.960 --> 00:54:37.639
with it. This is Sony Betamax
issue from the early nineteen eighties when the
631
00:54:37.840 --> 00:54:45.639
movie studios sued claiming infringement by virtue
of early video cassette recording of broadcast television.
632
00:54:45.280 --> 00:54:50.320
And as we said there was there
were two findings. One was that
633
00:54:50.840 --> 00:54:58.800
recording a show that was broadcast over
free broadcast television strictly for the purpose of
634
00:54:58.880 --> 00:55:05.440
watching it later at a more convenient
time is a fair use, but keeping
635
00:55:05.480 --> 00:55:10.199
libraries of those is not. And
then second, in terms of the contributory
636
00:55:10.280 --> 00:55:19.239
liability of the manufacturer of the Betamax
machine, Sony in that case, that
637
00:55:20.079 --> 00:55:27.039
the existence of commercially significant, substantial
noninfringing uses that the court would not impute
638
00:55:27.159 --> 00:55:30.920
knowledge of the infringement, and knowledge
being one of the two problems of contributory
639
00:55:30.960 --> 00:55:37.239
liability. The Supreme Court later in
Rockster made very clear that there is no
640
00:55:37.400 --> 00:55:40.880
safe harbor. This question was framed
as a safe harbor, and Sony that
641
00:55:42.000 --> 00:55:45.440
is not the correct framing of the
law. There is not a safe harbor.
642
00:55:47.000 --> 00:55:52.079
If there are commercially significant, non
infringing uses, then Sony is still
643
00:55:52.119 --> 00:55:57.519
good law that the court will not
impute knowledge for that reason. That doesn't
644
00:55:57.559 --> 00:56:00.719
mean the court might not impute knowledge
for another reason, in which it did
645
00:56:00.760 --> 00:56:06.679
in Brockster, and that reason being
in that case that the defendant had induced
646
00:56:06.840 --> 00:56:12.320
the infringements by the users of the
Groxter system. So all of that,
647
00:56:12.719 --> 00:56:15.800
of course is in the context of
contributory liability, which is a doctrine of
648
00:56:15.840 --> 00:56:22.159
secondary or third party liability in US
law. But there are direct infringement issues
649
00:56:22.239 --> 00:56:27.440
here that I think need to be
resolved first, and that's where the fair
650
00:56:27.559 --> 00:56:31.480
use comes in. The fair use
argument comes in which Sony has nothing to
651
00:56:31.519 --> 00:56:37.639
say about in this context. Yeah, so I I know we're pretty close
652
00:56:37.639 --> 00:56:39.119
to out of time. But if
I'm just say a couple of things about
653
00:56:39.159 --> 00:56:44.400
Sony, so Sony is simultaneously irrelevant
and also highly relevant to this case.
654
00:56:45.239 --> 00:56:49.239
So the reason that's not terribly relevant
is that Sony dealt with this contributory infringement
655
00:56:49.280 --> 00:56:52.880
issue and here the allegation is direct
infringement, right, And so as a
656
00:56:52.880 --> 00:56:58.199
result, by doctrine it doesn't help
us other than for the time shifting and
657
00:56:58.239 --> 00:57:01.280
the question of what counts as fair
use. But the way it is relevant
658
00:57:01.360 --> 00:57:06.920
is that Sony is within a line
of cases that deal with the intersection between
659
00:57:06.920 --> 00:57:09.760
copyright law and technology. Right.
So you have a new technology, the
660
00:57:09.840 --> 00:57:15.199
VCR, which doesn't itself say much
about any particular copyright work, but is
661
00:57:15.239 --> 00:57:21.960
a vehicle by which infringements can occur. And similarly, you had cases about
662
00:57:22.039 --> 00:57:24.079
xerox machines, photographs. You can
go all the way back to the printing
663
00:57:24.119 --> 00:57:28.519
press if you really want, and
you recognize that you have these general purpose
664
00:57:28.519 --> 00:57:35.840
technologies that can enable copyright infringement,
that can enable easier copying. How do
665
00:57:35.880 --> 00:57:39.519
you deal with that sort of balance? And really the answer comes down to
666
00:57:39.639 --> 00:57:45.239
what exactly are the boundaries of the
copyright? Right. If we say that
667
00:57:45.280 --> 00:57:49.119
copyright involves any sort of copying at
all, then of course all of these
668
00:57:49.119 --> 00:57:52.239
things are infringement, and you wouldn't
be able to have printing process, photographs,
669
00:57:52.320 --> 00:57:55.320
xerox machines, any of these sorts
of things. But of course there
670
00:57:55.360 --> 00:57:58.880
is a fair use doctrine. That
fair use doctrine exists for a reason.
671
00:57:58.920 --> 00:58:04.440
It's to set a bound around what
the copyright right protects and to allow for
672
00:58:04.480 --> 00:58:07.079
technological innovation beyond that space. And
so in a sense, what this case
673
00:58:07.119 --> 00:58:12.599
really is all about is what exactly
is that line. Where does the line
674
00:58:12.800 --> 00:58:19.719
fall such that an act that colloquially
we would call copying falls outside of copying
675
00:58:19.760 --> 00:58:22.760
and falls within the space of permissible
innovation. That's a line that's been pretty
676
00:58:22.760 --> 00:58:27.800
important throughout history and a lot of
different technologies. And so in that sense,
677
00:58:27.880 --> 00:58:31.039
this case is not terribly new.
It's simply an iteration of that same
678
00:58:31.159 --> 00:58:36.840
problem that has come up many many
times for copyright law and for a lot
679
00:58:36.880 --> 00:58:42.760
of other areas of law, how
it intersects with new technologies. So that's
680
00:58:42.880 --> 00:58:46.199
the false dichotomy of we wouldn't have
the printing pressor's xerox machines if we didn't
681
00:58:46.199 --> 00:58:51.400
have fair use, because there's no
such thing as licensing. That's of course
682
00:58:51.480 --> 00:58:57.039
not the case, Nor do I
believe that there's some sort of penumber around
683
00:58:57.079 --> 00:59:06.519
Sony or other copyright andechnological innovation cases. And what I do agree with is
684
00:59:06.559 --> 00:59:13.000
that copyright has always been a law
that has reacted to a variety of factors,
685
00:59:13.039 --> 00:59:17.320
developments in the marketplace, consumer preferences, and technological evolution. Absolutely,
686
00:59:19.599 --> 00:59:23.159
and the court looks at some of
the particulars and fair use has a key
687
00:59:23.280 --> 00:59:29.440
role in that, of course.
So we have decisions like Rockster, as
688
00:59:29.440 --> 00:59:32.920
I mentioned, Ario, another one
I mentioned, and many others where the
689
00:59:34.000 --> 00:59:37.400
court said, no, this is
a model based on infringement and we're not
690
00:59:37.440 --> 00:59:42.079
going to permit it. We have
others where the court, like in Sony,
691
00:59:42.199 --> 00:59:46.000
said this is inherently meant to be
an innocent business that could be used
692
00:59:46.039 --> 00:59:50.760
for something afarious, but we don't. We're not going to hold the manufacturer
693
00:59:50.840 --> 00:59:55.360
or the device secondarily liable just because
it could be used in a bad way.
694
00:59:57.519 --> 01:00:00.760
In the case before us, these
generatives AI systems that made a conscious
695
01:00:00.800 --> 01:00:07.519
decision to copy copyrighted works by the
billions. I think it's a lot more
696
01:00:07.679 --> 01:00:17.800
like Ario and Rockster than it is
like the innocent and business models Aily.
697
01:00:17.880 --> 01:00:22.800
I know we're up against the hour, and I have just one question for
698
01:00:22.880 --> 01:00:25.360
the panel as a whole, and
the question is what if we remove the
699
01:00:25.400 --> 01:00:30.880
technology from this discussion and just put
it into a business sense that for some
700
01:00:30.920 --> 01:00:36.320
reason the company starts a business and
happens to go into the New York Times
701
01:00:36.400 --> 01:00:40.239
archives and carts it all off into
their warehouse, and then they adhire a
702
01:00:40.360 --> 01:00:45.239
whole bunch of employees for people who
come into the front counter and ask them
703
01:00:45.320 --> 01:00:50.039
questions, and they go back and
skirt around the New York Times information and
704
01:00:50.039 --> 01:00:55.599
give them an answer or or an
asset that's directly competing with what the New
705
01:00:55.719 --> 01:01:01.000
York Times uses is archives for Is
is that a valid analogy? Is it,
706
01:01:01.079 --> 01:01:06.920
you know, technologically and misplaced or
does that make a difference in the
707
01:01:06.960 --> 01:01:12.239
discussion. If you remove the technology
from what's actually happening with the copyright,
708
01:01:12.280 --> 01:01:17.039
it works, John, I think
I think you're a little description. It's
709
01:01:17.119 --> 01:01:22.159
kind of close to a library,
which is okay, but you do get
710
01:01:22.199 --> 01:01:24.559
into some But if you tweak it
in two ways. One, if you
711
01:01:24.599 --> 01:01:28.199
say, what if they give copies
of Times that are clothes I cart it
712
01:01:28.239 --> 01:01:32.320
off instead of just giving them summaries, or even if like a copy cut
713
01:01:32.360 --> 01:01:37.639
and paste out sentences instead of a
ransom note style answer, I think you're
714
01:01:37.679 --> 01:01:42.760
getting closer to the fat current here. And then the flip side is what
715
01:01:42.920 --> 01:01:49.599
if they grab the New York Times
from you know that from how offer presses
716
01:01:50.320 --> 01:01:54.280
in the East Coast? Why are
summaries of a story so West coast and
717
01:01:54.320 --> 01:01:59.159
print from there? And of course
that's ions versus AP, which was held
718
01:01:59.159 --> 01:02:04.760
to be young awful. So I
don't think you can make it technology independent
719
01:02:04.800 --> 01:02:08.559
because all the technology dictates some of
illegal questions. I mean, to make
720
01:02:08.599 --> 01:02:14.199
things even worse. They do talk
about hallucinations, the problem in which sometimes
721
01:02:14.239 --> 01:02:16.400
you will ask for a New York
Times article from chatchipg, I'll give you
722
01:02:16.400 --> 01:02:20.639
something completely made up that has nothing
to do with that. So the analogy
723
01:02:20.679 --> 01:02:23.519
really is a library reference desk that
gives you, every once in a while
724
01:02:23.639 --> 01:02:27.480
an exact copy of an article that
I found, and every once in a
725
01:02:27.519 --> 01:02:32.199
while gives you totally useless factual information. It's a little hard to draw analogies
726
01:02:32.239 --> 01:02:37.920
to that point, I think,
but it does, and I think it
727
01:02:37.000 --> 01:02:40.039
kind of shows why it's a little
difficult to work with analogies, and sometimes
728
01:02:40.079 --> 01:02:44.039
it really is worth just figuring out
exactly what's going on with the technology.
729
01:02:44.079 --> 01:02:51.159
Yet it's at its proper levels.
I'll just add that what you described was
730
01:02:51.199 --> 01:02:54.880
a lot like a library. If
it's a nonprofit organization, you didn't stipulate
731
01:02:54.920 --> 01:03:01.159
which way that organizations operating nonprofit ormercial. But the point I want to make
732
01:03:01.280 --> 01:03:12.679
is Congress has taken pains to enact
specific statutory exemptions for libraries to provide library
733
01:03:12.679 --> 01:03:19.719
patrons with a certain degree of service
without going so far that it implicates the
734
01:03:20.000 --> 01:03:24.599
incentive to create in the first place. And libraries are not entitled to just
735
01:03:25.440 --> 01:03:30.119
take copies of everything that they want
to have in their collections without paying for
736
01:03:30.159 --> 01:03:36.119
them, without permission and so on. And in fact, in the online
737
01:03:36.159 --> 01:03:39.559
context, some of these issues are
being litigated right now. There's an entity
738
01:03:40.239 --> 01:03:46.039
that calls itself the Internet Archive that
has created a website where it just lets
739
01:03:46.079 --> 01:03:51.760
people take copies of copyrighted works,
and they want to call themselves the library,
740
01:03:52.519 --> 01:03:57.719
but they don't qualify under any of
the library exceptions, and thus far
741
01:03:57.760 --> 01:04:01.320
in the pending litigation they've been quite
unsuccessful. It's on appeals, so we'll
742
01:04:01.320 --> 01:04:03.800
see where it goes. But I
don't I wouldn't put a lot of money
743
01:04:03.800 --> 01:04:13.559
on them. Over to you,
Emily all right on behalf of the Federal
744
01:04:13.599 --> 01:04:16.239
Society. Thank you all for joining
us for this great discussion today. Thank
745
01:04:16.280 --> 01:04:19.880
you also to our audience for joining
us. We greatly appreciate your participation.
746
01:04:20.440 --> 01:04:25.599
Check out our website fedsoc dot org
or follow us on all major social media
747
01:04:25.639 --> 01:04:29.880
platforms at fedsoc to stay up to
date with announcements and upcoming webinars. Thank
748
01:04:29.920 --> 01:04:33.320
you once more for tuning in,
and we are adjourned. Thank you for
749
01:04:33.400 --> 01:04:38.639
listening to this episode of FEDSOC Forums, a podcast of the Federal Societies Practice
750
01:04:38.639 --> 01:04:42.920
Groups. For more information about the
Federal Society, the practice groups, and
751
01:04:42.960 --> 01:04:46.199
to become a Federal Society member,
please visit our website at fedsoc dot or org

과거 에피소드 검색

FedSoc Forums의 과거 에피소드를 검색하세요.
이 검색 창을 웹 사이트에 삽입
면책 조항: 이 페이지에 포함된 팟캐스트와 작품은 The Federalist Society에서 가져온 것입니다. 이 팟캐스트는 소유자의 재산이며 Listen Notes, Inc.와 제휴하거나 보증하지 않습니다.
PROMOTED
The best Podcast API to search all podcasts and episodes. Trusted by 15,200+ companies and developers.
The best Podcast API to search all podcasts and episodes. Trusted by 15,200+ companies and developers.
PodcastAPI.com

추천
API를 사용하여 이 에피소드 가져오기