-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
227 lines (222 loc) · 11.8 KB
/
index.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<meta http-equiv="X-UA-Compatible" content="ie=edge">
<title>AI Ethics Panel Notes</title>
<link rel="stylesheet" href="style.css">
<style>
html {
font-family: Arial, Helvetica, sans-serif;
max-width: 1200px;
margin: 0 auto;
}
img {
width: 100%;
height: auto;
}
</style>
</head>
<body>
<p>Please arrive by 7:45pm. Panel starts at 8pm sharp.</p>
<h2 id="8pm-welcome-and-introductions">8pm- Welcome and introductions</h2>
<p>I’ll ask each of you to introduce yourselves. 30 seconds or less each.
</p>
<h2 id="8-02-art-and-ai">8:02 - Art and AI</h2>
<p>(5 questions approx 3 minutes per question)</p>
<ol>
<li>
<p>Who owns the art etc generated by AIs and why?</p>
<ul>
<li>"It’s also odd to some lawyers that generative AI firms
are being sued and not those that compiled the dataset. In
the case of Midjourney, that would be the large-scale
Artificial Intelligence Open Network (LAION), based in
Germany. “If LAION created the dataset, then the alleged
infringement occurred at that point, not once the dataset
was used to train the models,” Eliana Torres, an
intellectual property lawyer with the law firm Nixon
Peabody, told <a
href="https://techcrunch.com/2023/01/27/the-current-legal-cases-against-generative-ai-are-just-the-beginning/"><em>Tech
Crunch</em></a> last month." <a
href="https://techmonitor.ai/technology/ai-and-automation/will-2023-be-year-of-ai-lawsuit">https://techmonitor.ai/technology/ai-and-automation/will-2023-be-year-of-ai-lawsuit</a>
</li>
</ul>
</li>
<li>
<p><strong>What is borrowing/what is copying?</strong></p>
<ul>
<li>My work is <a
href="https://rom1504.github.io/clip-retrieval/?back=https%3A%2F%2Fknn.laion.ai&index=laion5B-H-14&useMclip=false&query=orange+inflatable+sculpture+japan">in
the LAION dataset</a> If "orange inflatable
sculpture japan" is searched, my work is first hit with
aesthetic score of 7 and fill 4 of the top 10 results and
many more</li>
<li>Invisible bicycle and many other works are in the dataset
</li>
<li>With prompt fiddling I can get works that are close to mine
from the generators but I do not have enough images online
to make a full duplicate, but definitely see parts</li>
<li><img src="attachments/Screen%20Shot%202023-02-08%20at%206.31.18%20PM%20copy.png"
alt=""> My images in the red squares.</li>
<li><img src="attachments/Screen%20Shot%202023-02-08%20at%205.29.20%20PM%20copy.png"
alt="">My images in the red squares.</li>
<li><a
href="https://rom1504.github.io/clip-retrieval/?back=https%3A%2F%2Fknn.laion.ai&index=laion5B-H-14&useMclip=false&imageUrl=https%3A%2F%2Fwww.jimmykuehnle.com%2Fexhib%2Faichi_geidai%2FJimmy_Kuehnle_Big_Blob_inflatable_suit_viewer_participation_418.jpg">LAION
search result with multiple images of mine and my
name</a></li>
<li><a
href="https://fairuse.stanford.edu/overview/fair-use/what-is-fair-use/">https://fairuse.stanford.edu/overview/fair-use/what-is-fair-use/</a>
</li>
<li>The idea cannot be copyrighted but the expression of the
idea can. If AI uses too many parts of other artists' or
writers' specific expression of ideas, then it would
fall more into copying than borrowing.</li>
<li><a
href="https://petapixel.com/2023/02/02/ai-image-generators-can-exactly-replicate-copyrighted-photos/">Article
from Peta Pixel</a> that shows how some images are
retained in the diffusion models</li>
<li>Generated Image below<img
src="https://petapixel.com/assets/uploads/2023/02/Fnz3bW4aUAQsA5V-800x554.jpeg"
alt="image"></li>
</ul>
</li>
<li>
<p><strong>What are the limitations of AI?</strong></p>
<ul>
<li>currently a slight learning curve</li>
<li>big compute needed to make new models</li>
<li>AI doesn't have knowledge yet</li>
<li>Images relatively low resolution</li>
</ul>
</li>
<li>
<p><strong>How long will these limitations exist?</strong></p>
<ul>
<li>Not that long, the year over year, and even month over month
improvements are staggering.</li>
<li>Google's new Muse AI image generator just sped up its
text to image generation by a factor of 10, now takes about
a second an image, eventually will be faster than realtime -
<a
href="https://youtu.be/2AsoWS2t484">https://youtu.be/2AsoWS2t484</a>
from Two Minute Papers
</li>
<li>Google and Facebook already have video generators that are
currently crude but that won't last for long - <a
href="https://youtu.be/uzF6CTtjn-g">https://youtu.be/uzF6CTtjn-g</a>
from Two Minute Papers</li>
</ul>
</li>
<li>
<p>What about the AI Lawsuits?</p>
<ul>
<li>The abstracts of the lawsuits seem to misunderstand the
technology and therefore may not succeed.</li>
<li>Jurors may rule against the generators anyway.</li>
</ul>
</li>
</ol>
<h2 id="-8-25-ai-and-bias">~8:25:- AI and Bias</h2>
<p>(2 questions- approximately 3 minutes per question)</p>
<ol>
<li>
<p>Can we eliminate bias in AI?</p>
</li>
<li>
<p>Given that fairness is subjective (even mathematically), how do
we decide what is fair and unbiased?</p>
</li>
</ol>
<h2 id="-8-31-ai-and-manipulation">~8:31 - AI and Manipulation</h2>
<p>(3 Questions-approximately 3 minutes per question)</p>
<ol>
<li>
<p><strong>How effective is AI at manipulation (politics
etc)?</strong></p>
<ul>
<li><a
href="https://www.nytimes.com/2023/02/07/technology/artificial-intelligence-training-deepfake.html">The
People Onscreen Are Fake. The Disinformation Is
Real.</a> - New York Times article about state actors
creating deepfakes to spread misinformation</li>
<li>Tom Cruise example - needed a body double and voice actor
but now could be more easily done</li>
</ul>
</li>
<li>
<p><strong>Who should be held responsible?</strong> - The software
is already available and open source, ultimately the actors are
responsible. For state actors sanctions and diplomacy and
deterrence will likely be the unfortunate solution as with
nuclear weapons.</p>
</li>
<li>
<p><strong>Can we identify manipulative AI (deepfakes etc)?</strong>
- for now yes. It will not be long until the detection takes
much longer than a campaign cycle. Eventually it will likely be
almost undetectable without massive compute power.</p>
</li>
</ol>
<p>~8:40 - Questions from Audience</p>
<p>~9:15 — What is the future of AI and Ethics?</p>
<p>(Approximately 5 minutes)</p>
<p>~ 9:20 I’ll ask each of you to give a short closing remark (30 seconds).
Then I’ll give a session closing remark, thank the audience and staff
</p>
<p>9:30 we are done!</p>
<p>I’ll leave you with this article on AI Lawsuits</p>
<ul>
<li>Stable Diffusion Getty <a
href="attachments/1-1-stable-diffusion-complaint.pdf">Lawsuit
Filing</a>
<ul>
<li>"Stability downloaded or otherwise acquired copies of
billions of copyrighted images without permission to create
Stable Diffusion," - This could easily be circumvented
by having the AI look as a screen of images via a camera
without ever copying the images.</li>
<li>"Ultimately, it is merely a complex collage tool."
- Collage can already be protected by fair use.</li>
<li>"By training Stable Diffusion on the Training Images,
Stability caused those images to be stored at and
incorporated into Stable Diffusion as compressed copies.
Stability made them without the consent of the artists and
without compensating any of those artists." -
Misunderstanding of the diffusion process and compression
</li>
<li>"All AI Image Products operate in substantially the
same way and store and incorporate countless copyrighted
images as Training Images." - They are not copying the
images or storing them.</li>
<li>The best argument is the financial harm to the artist when
typing a name in gets a style. This could be avoided by
removing names from the training set or prompts.</li>
<li>The lawsuits are going to have a tough hill to climb and
then they will have to litigate the next technology</li>
<li>"The resulting image is necessarily a derivative work,
because it is generated exclusively from a combination of
the conditioning data and the latent images, all of which
are copies of copyrighted images. It is, in short, a
21st-century collage tool." p20 again - not all the
images in the training are copyrighted so this is going to
be a hard line of argument</li>
</ul>
</li>
<li><a
href="https://techmonitor.ai/technology/ai-and-automation/will-2023-be-year-of-ai-lawsuit">https://techmonitor.ai/technology/ai-and-automation/will-2023-be-year-of-ai-lawsuit</a>
</li>
</ul>
<p><img src="attachments/inflatable_orange_fin_for_suit.jpg" alt="">
original
prompt file name inflatable_orange_fin_for_suit.jpg</p>
<p><img src="attachments/Pasted%20image%2020230208173651.png" alt=""></p>
<p>Prompt - inflatable_orange_fin_for_suit aichii japan sculpture rugakusei
</p>
<p><img src="attachments/Pasted%20image%2020230208180638.png" alt=""></p>
<p><img src="attachments/Pasted%20image%2020230208182932.png" alt=""></p>
<p><img src="attachments/Pasted%20image%2020230208182954.png" alt=""></p>
<p><img src="attachments/Pasted%20image%2020230208183003.png" alt=""></p>
</body>
</html>