1 of 28

David Atkinson

Bluesky

Intersecting AI on

Intersecting AI on

“GenAI Exceptionalism” Book Preorder Notification Signup

Intersecting AI on

2 of 28

Why It’s Important

  • If it’s fair use, then GenAI companies don’t have to pay for most songs, movies, images, videos, books, essays, code, and other material it trained on.
  • Fair use = no copyright infringement

​

  • Copyright law was created pursuant to Congress’s express constitutional power…

“To promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries”

  • Progress must be thought of in decades and centuries, not the sugar high of months or years.

2

3 of 28

The Most Cited Cases

  • Functional Cases
    • Author’s Guild v. Google (“Google Books”)
    • Author’s Guild v. HathiTrust (“HathiTrust”)
    • A.V. v. iParadigms (“iParadigms”)
    • Perfect 10 v. Amazon.com (“Perfect 10”)
    • Kelly v. Arriba Soft Corporation (“Kelly”)

3

Criticism or commentary

Search or indexing utility

Software interoperability

4 of 28

The Most Cited Cases

  • Functional Cases
    • Google Books
    • HathiTrust
    • iParadigms
    • Perfect 10
    • Kelly

4

Plagiarism checker Turnitin.com

5 of 28

The Most Cited Cases

  • Functional Cases
    • Google Books
    • HathiTrust
    • iParadigms
    • Perfect 10
    • Kelly

5

Plagiarism checker Turnitin.com

Indexing for image search

Query for information about the books. Also provided attribution, a link to buy the book, and an option to opt out.

All of these served a single function. Turnitin.com will never create an infringing image, for example.

6 of 28

Expressiveness Didn’t Matter

Kbvkj kjbdsoiw iuwebc chcie kdjcb cuegwn lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum.

​

6

7 of 28

The Most Cited Cases

  • Interoperable Cases
    • Sega Enterprises v. Accolade (“Sega”)
    • Sony Computer Entertainment v. Connectix (“Sony”)
    • Google v. Oracle America (“Oracle”)

7

8 of 28

The Most Cited Cases

  • Interoperable Cases
    • Sega
    • Sony
    • Oracle

8

Reverse-engineered code for its functional requirements. The people then only used the functional code and provided all the creative/expressive elements for the games.

9 of 28

The Most Cited Cases

  • Interoperable Cases
    • Sega
    • Sony
    • Oracle

9

Reverse-engineered code for its functional requirements. The people then only used the functional code and provided all the creative/expressive elements for the games.

Google’s copying targeted a widely used, utilitarian aspect of computer code. Google was not targeting the expressiveness of the code.

10 of 28

How the Facts Differ From GenAI

  1. GenAI companies are interested in the expressive elements!
    1. The quality of the expressiveness matters!

10

11 of 28

How the Facts Differ From GenAI

  1. GenAI companies are interested in the expressive elements!
    1. The quality of the expressiveness matters!
    2. The expressive elements (the words, sentences, images, sounds) are stored within the models.

11

“Write a song about moving from Philadelphia to Bel Air”

​

***There are an infinite number of ways the model could have replied!***

12 of 28

How the Facts Differ From GenAI

  1. GenAI companies are interested in the expressive elements!
    1. The quality of the expressiveness matters!
    2. The expressive elements (the words, sentences, images, sounds) are stored within the models.
    3. Unlike Sega and Sony, the GenAI companies provide none of the expressiveness their models exhibit

12

13 of 28

About those works…

  • ChatGPT and Gemini seem to know something OpenAI and Google don’t…

(screenshots from 5/2/2025)

13

The Question

ChatGPT’s Response

Gemini’s Response

14 of 28

14

267 terabytes = 267,000 gigabytes

It's like the old saying: "If it's too hard to do legally, it's ok to do it illegally."

​

This is why it's ok to rob a bank if you aren't making money quickly enough by working a full-time job.

15 of 28

15

What Does it Mean If GenAI Companies Are Right?

You never have to pay for any content ever again!

16 of 28

The Four Factors

  • EVERY argument that works in favor of GenAI works at least as well for humans

​

  • Factor 1:
    • Transformative?
      • Humans are far more likely to transform information learned from copyrighted works.
      • Ever hear of calculus? Cubism? Evolution? General relativity? Honky tonk?
      • How many new scientific theories has GenAI developed? New fields of study? New genres of music or movies?
      • How many by humans? ALL OF THEM!

16

17 of 28

The Four Factors

  • Factor 2:
    • It’s indisputable that GenAI companies train on highly creative works.

​

    • Both models and brains are interested in the underlying facts and ideas of copyrighted works and the expressive elements.

​

    • Brains brains aim to understand the underlying relationships between words and sentences, colors and shapes, sounds and rhythms, etc. and they store some representation of them.

17

18 of 28

The Four Factors

  • Factor 3:
    • Humans do not require mass copyright infringement to be useful, and that should count in favor of humans just as this factor has always favored less unauthorized reproduction of copyrighted works, not more.

​

    • More infringement doesn’t equal less infringement!

18

19 of 28

The Four Factors

  • Factor 4:
    • Humans rarely unfairly compete with copyright owners and do so no more frequently than GenAI models.

​

    • Competitive?
      • Know anyone who can recite a chapter from Harry Potter from memory?
      • How many times have you sold an album of songs that sounded substantially similar to Taylor Swift?
      • Write any books substantially similar to Stephen King’s recently?

19

20 of 28

The 8 Most Common Arguments

  1. Licensing Materials: GenAI should not be limited to only authorized use of copyrighted works.
  2. Expressive v. Underlying Ideas: GenAI merely extracts the underlying meta information from the source content.
  3. GenAI As Humans: For some purposes, society should afford GenAI the same legal rights as humans for fair use.
  4. GenAI as Non-Human: For other purposes, society should not treat GenAI like a human at all.
  5. Only Output Matters: The collection and use of copyrighted work as inputs should not matter for copyright analyses.
  6. Quantity of Infringing Output: GenAI's generation of infringing content is significantly less important than its frequency of generation.
  7. Enabling Greater Creativity: GenAI provides substantial benefits to society and deserves more permissive treatment under copyright law.
  8. Stifling Progress: If fair use is not broadly interpreted and applied to GenAI, it will stifle innovation.

20

21 of 28

GenAI as Humans

  • “The right to learn” is silly
    • Humans cannot download every book, song, and movie we want for free and claim it’s so we can learn.
  • GenAI is trained on billions of documents, making up trillions of tokens
    • At most, a human could read about 5,000 books in a lifetime. (1 book a week for 60 years)

21

A trillion is bigger than you think!

22 of 28

Only Outputs Matter

  • How often do you recreate an almost identical version of the novel you read?

22

23 of 28

Stifling Progress

  • First, innovation doesn’t equal progress
    • Asbestos, Roundup, cigarettes, lead paint, etc.

​

  • GenAI causes tremendous environmental harm, and it’s not clear it’s a good tradeoff for humanity

23

24 of 28

Stifling Progress

24

  • Humans:
    • Physics
    • Calculus
    • Evolution
    • Quantum mechanics
    • microRNA
    • Antibiotics
    • DNA’s structure
    • Thermodynamics
    • Etc. etc.

25 of 28

Stifling Progress

25

  • Humans:
    • Physics
    • Calculus
    • Evolution
    • Quantum mechanics
    • microRNA
    • Antibiotics
    • DNA’s structure
    • Etc. etc.
  • GenAI:
    • *crickets chirping*

26 of 28

In Summary

  • Prior case law is fundamentally different from the facts and context of GenAI

​

  • To build the most capable GenAI systems, companies must download millions of books and articles from pirate sites.

26

27 of 28

In Summary

27

  • Every argument a GenAI company could make applies at least as powerfully to humans.
    • Humans cannot legally download anything they want to learn from without authorization or payment.
    • Just license the content!

​

Why should the GenAI companies at issue, which are all for-profit, receive special legal privileges?

28 of 28

The GenAI Should Be Allowed to Learn Like Humans Argument

  • Humans pay for stuff we use either directly or indirectly

​

  • Also…
    • GenAI trains on many orders of magnitude more data

​

    • GenAI processes data many orders of magnitude faster

​

    • GenAI is more likely to memorize and regurgitate training data

​

    • GenAI companies have far more market power than individual humans

​

    • GenAI has the potential to obliterate markets, unlike individual humans

28