"The AI Copyright Fight: A Guide to Product Market Fit"

"The AI Copyright Fight: A Guide to Product Market Fit"

In early 2023, the AI copyright fight came to the forefront as artists and Getty Images separately sued AI image generators for copyright infringement. This battle over copyright has its roots in the royal courts of Europe around 500 years ago. Back then, monarchs granted privileges and licenses to their favored artists, giving them exclusive rights to publish and distribute works of art. This exclusivity led to the term "copyright."

Initially, only royally appointed publishers could afford printing presses and adhered to strict censorship standards. However, with the rise of Gutenberg's more affordable printing press in the mid-1500s, the power dynamics shifted. Censor-publishers struggled to keep up with the reproduction of unlicensed work, and public opinion began to favor the "liberty of unlicensed printing" advocated by English poet John Milton.

The Copyright Act of 1710 in England was the first law to grant copyrights to authors. Ironically, it was a last-ditch effort by the royal publishers to retain control over printing power. Copyright continued to evolve over time, with the introduction of the concept of "fair use" in the Copyright Act of 1976. This allowed certain entities, initially nonprofits and government entities, to use copyrighted materials under specific conditions.

In recent years, courts have broadened the definition of fair use to include commercial use if other compelling factors are present. This was seen in cases such as Perfect 10 suing Google for indexing magazine covers and Authors Guild suing Google for scanning copyrighted library books. In both cases, the courts ruled that Google's use of the images and snippets was "highly transformative" and served a public benefit by augmenting public knowledge.

The concept of transformative use has been a focal point in many copyright lawsuits, although its determination is subjective and decided on a case-by-case basis. Judges consider not only how much a reproduction has changed but also the content of the copying. For example, a work like "50 Shades of Gray" is considered transformative of "Twilight" to the point where copyright infringement would be difficult to prove.

When creators adapt copyrighted material from one medium to another, such as in film, books, or music, it is often considered a derivative work. Licensing fees must be paid if enough aspects of the original story remain unchanged. Fan fiction and parodies fall into a gray area, with courts ruling differently depending on the case. Transformative works, protected under fair use and the First Amendment, are prevalent in our culture, but creators often lack a definitive way to determine if their use qualifies without going to court.

The recent lawsuits surrounding AI image generators revolve around the use of copyrighted images without permission and the production of derivative versions. The question of whether reproducing copyrighted work in AI training constitutes fair use is being debated in court. One plaintiff goes further, arguing that AI generators producing work "in the style" of an artist should require commission or licensing. If such precedents are set, it could infringe on copyright whenever inspiration is drawn from others' work.

It is unclear how judges will rule on these cases, but a recent high-profile music case upheld a similar copyright infringement claim. This could potentially lead to a regression of copyright, with a few chosen artists monopolizing ideas and enforcing their preferred forms of censorship. The remaining creators would have no legal means to publish and distribute their work, resulting in a dystopian world.

To navigate the complexities of copyright and fair use, AI companies should seek permissions from copyright holders before using data to train AI models. While scraping publicly available data may not be illegal, using it for AI training is a legal and ethical gray area. Attributions to artists should also be given by AI-generated images, as is commonly required by creative commons permissions.

Furthermore, AI-generated images that have not been modified by a person should remain in the public domain without copyright. Machines and algorithms are not humans and should not be granted the same creative protection rights. The concern is that AI could invade the territories of art, potentially supplanting or corrupting it if not properly regulated.

Switching gears to the topic of product market fit (PMF), it is essential to focus on the market first when seeking PMF. PMF means being in a good market with a product that can satisfy that market. It is a misconception that PMF is a discrete, big bang event or that it is always obvious when it is achieved. PMF can be lost, and competition should still be taken into account even after achieving it.

Hiring before achieving PMF can slow down a startup, while hiring after achieving PMF can accelerate growth. Cohort retention rate serves as a fair metric for determining PMF. A flattened retention curve indicates PMF for a specific market or audience. Additionally, if 40% of users would be "very disappointed" without a product, it is considered a leading indicator of PMF.

When it comes to product/market fit, it is better to have a small number of people who want the product intensely rather than a large number who want it only mildly. Consumer products generally aim for a 25% floor, while B2B SaaS products aim for a 70% floor in terms of user satisfaction.

The market is constantly moving and changing, requiring startups to keep their thumb on the pulse of product/market fit. Eric Ries and Keith Rabois offer different models, with one emphasizing market orientation and the other vision orientation. Ultimately, the fundamentals of a product should remain consistent while adapting to market changes.

In conclusion, both the AI copyright fight and product market fit require careful consideration and understanding. AI companies must navigate the complexities of copyright law and fair use, seeking permissions and providing attributions to artists. Meanwhile, startups should prioritize the market when seeking product/market fit, constantly monitoring the pulse of their market and adapting their product accordingly. By staying informed and proactive, creators and entrepreneurs can navigate these challenges and thrive in the ever-evolving landscape.

Three actionable advice:

1. Seek permissions from copyright holders before using data to train AI models.
2. Provide attributions to artists for AI-generated images, even in the public domain.
3. Constantly monitor the market and adapt the product to maintain product/market fit.