Intellectual Property in the Digital Age

Introduction: What Is Intellectual Property?

Consider a thought experiment from an 1813 letter of Thomas Jefferson's. Light a candle from mine and you have fire, but my flame is no smaller. Give you half my sandwich and I have half as much. Physical goods are rivalrous — my use prevents yours. Ideas are not: share one and you can pass it on a hundred times over, and none of us has any less. If ideas cannot be "taken," what would it mean to "own" one?

He who receives an idea from me, receives instruction himself without lessening mine; as he who lights his taper at mine, receives light without darkening me.

Jefferson's puzzle is at the heart of intellectual property law — the rules granting creators exclusive rights for a set period. Unlike physical property rights, which protect what already exists, IP rights create legal scarcity where nature provides abundance. A book can be copied perfectly at near-zero cost; a drug formulation can be synthesized by any competent chemist. IP law manufactures the scarcity deliberately, on the argument that without it creators would have too little incentive to create.

Whether that argument is sound — whether IP systems serve their stated purpose or have been captured by incumbents — is a central debate of this chapter. It matters for digital technology because computers are at their core copy machines, and networks distribute those copies globally for nothing. Nearly everything a computer does collides with traditional IP law's assumptions.

The Four Types of Intellectual Property

Modern IP law recognizes four categories, each with a different basis, duration, and rules. The differences matter because debates conflate them: arguments that work for patents may not apply to copyright.

Copyright matters most in the digital age, since it governs text, music, images, video, and software. Its crucial distinction is between ideas and expression: it protects the way an author expresses an idea, not the idea. Anyone may write a novel about a young wizard at a magical school; J.K. Rowling's copyright covers the words, characters, and plot of Harry Potter, not magical education as a premise. Ideas are the building blocks of intellectual activity, and the law treats them as unownable.

Fair use is the copyright system's pressure valve, and it is deliberately vague: no list of permitted uses, no bright line, only four factors a judge weighs after the fact. That is a feature if you think copyright must bend to circumstances its drafters could not foresee, and a bug if you are a filmmaker trying to learn in advance whether you will be sued. Nearly every hard question in this chapter, from Google's use of Java to the training of language models, turns on how a court weighs those factors.

Patents differ from copyright in requiring public disclosure. An inventor must explain how the invention works, and that explanation becomes public record; in exchange comes a twenty-year monopoly on making, selling, and using it. The theory is a fair bargain: society gets the knowledge, the inventor a temporary exclusive right, and afterwards the invention is free for anyone. Whether twenty years is the right price the pharmaceutical case study below will sharpen.

A Brief History of Expanding Rights

The striking feature of IP history is the relentless expansion of rights. U.S. copyright began at 14 years and now runs to the author's life plus 70; patents now cover software, business methods, and algorithms earlier courts would never have found patentable. Each extension was justified as "encouraging creativity," which applies awkwardly to retroactive ones: extending copyright on an existing work cannot incentivize its creation. Something else is driving the expansion.

Philosophical Justifications for IP

Why should intellectual property rights exist at all? Three traditions offer competing answers, each with different policy consequences. A Lockean thinks IP should track the labor invested; a utilitarian cares only whether the system produces more creation than the alternatives; a Hegelian thinks certain authorial rights are inalienable regardless of economics.

The Labor Argument

Locke's argument, adapted from his Second Treatise of Government , tracks our everyday sense that people deserve credit for what they make. But ideas are never purely original. Newton "stood on the shoulders of giants"; Shakespeare borrowed his plots wholesale; virtually every song shares chord progressions with countless predecessors. If creations always build on prior creations, how much of any work is genuinely the creator's own? The Lockean framework struggles to draw the line.

Locke added a proviso: appropriation is justified only so long as "enough and as good" is left for others. In the physical world that runs out — there is only so much land. In the world of ideas it never does, since Jefferson's expressing an idea does not stop anyone else from expressing it in their own words. So even if the labor theory justifies some IP rights, it justifies none of the strongest, which stop others building on existing ideas.

The Incentive Argument

The incentive argument is how IP is most commonly defended in courts, legislatures, and economics textbooks. It is also the most testable, and the evidence is messier than the theory. In pharmaceuticals, where development cost vastly exceeds copying cost, patents do appear necessary to sustain R&D. Elsewhere, industries have thrived for centuries without meaningful IP. Consider four of the most creatively dynamic fields in modern life, all operating with minimal protection:

• Fashion design. Garment designs generally cannot be copyrighted in the United States, yet Paris houses have driven global trends for a century on prestige and lead time.
• Restaurant cuisine. Recipes cannot be copyrighted, yet fine dining innovates continuously. Ferran Adrià's techniques at El Bulli were freely copied worldwide, and his restaurant remained the most coveted reservation on earth.
• Stand-up comedy. Once a joke is told publicly anyone can repeat it, and lawsuits are unheard of. Comedians keep producing new material, policed by a norm against joke theft.
• Sports strategy. A coach who invents a new play cannot patent it; opponents adopt it the following week. Coaching innovation continues unabated.

In none of these cases has the absence of IP stifled creation. It has forced them to compete on ongoing innovation rather than past monopoly — on the utilitarian's own terms, not obviously worse. IP incentives are neither universally necessary nor universally effective.

Term length sharpens the question: how much protection is enough? Economic analysis puts the optimal term far below current law. A songwriter is not more likely to compose because copyright will run seventy years after their death rather than fifty; those extra decades pay heirs and rights-holders, not the creator. Current terms exceed what incentive theory justifies, which raises the question of whose interests they serve.

The Personality Argument

The personality theory, associated with Hegel's Elements of the Philosophy of Right and developed most fully in continental European law, gives rise to moral rights — rights independent of economic interests that in many jurisdictions cannot be signed away. A French author who sells the copyright in a novel still keeps the right to be named as its author and to object to alterations damaging their reputation. These appear across Europe and in Article 6bis of the Berne Convention, but are far weaker in the United States, where copyright is predominantly commercial.

A revealing example: in 1975 ABC broadcast Monty Python's Flying Circus with roughly a quarter of the running time cut for commercials and material thought offensive to Americans. The troupe sued and, in Gilliam v. American Broadcasting Companies (2d Cir. 1976) , won a preliminary injunction — but through copyright, since the editing exceeded ABC's license, and through the Lanham Act's ban on false designation of origin, the court noting that American copyright does not recognize the droit moral. Their complaint — this is no longer our work, yet it bears our name — is what moral rights address, and US law had to reach it sideways. A French author would have had a direct claim.

The Social Contract of IP

All three justifications share a structure: IP rights are a bargain between creators and the public. Creators receive a temporary exclusive right; in exchange, the public eventually receives full, free access. The bargain is explicit in patent law, which requires disclosure for the monopoly, and implicit in copyright, which expires. Working properly it balances incentive against access — neither so weak that creators cannot recoup their investment, nor so strong as to be a permanent monopoly on knowledge.

Critics like James Boyle argue that the bargain has broken down . Terms have been extended so often that works effectively never enter the public domain within the lifetime of anyone alive at their creation. The result, Boyle argues, is a "second enclosure movement" — the fencing off of the intangible commons of the mind, analogous to the enclosure of English common lands.

Digital Disruption

IP law was built on assumptions about the physical world: that copying is expensive and imperfect, that distribution requires infrastructure, and that enforcement is practical at the level of commercial reproduction. Digital technology invalidated all three at once. A perfect copy costs nothing; the internet distributes it globally in seconds; the volume of copying makes suing copyists impractical. The result was one of the severest disruptions any industry has experienced.

Thomas A. Edison, c. 1877–1880. Public domain.

Thomas Edison (1847–1931)

American inventor and businessman, Edison held 1,093 U.S. patents — a record that stood for decades. At Menlo Park, New Jersey, he pioneered the industrial research laboratory, his "invention factory": researchers whose job was to produce patentable inventions on a schedule. His best-known work includes the phonograph, the incandescent light bulb, and early motion picture technology.

His career raises a persistent question about patents and innovation. The laboratory's output was remarkable. But his aggressive patent strategies — especially in the "War of Currents" against Nikola Tesla and George Westinghouse — are widely seen as having retarded adoption of the superior AC system. Edison used IP as a competitive weapon, not merely a reward for invention: a combination of creativity and strategic litigation characteristic of large actors today.

How Digital Changed Everything

Before digitization, the economics of creative industries rested on the cost of physical reproduction: a record required pressing, a book printing and binding, a film celluloid and a projector. Record labels, publishers, and studios were therefore not primarily in the business of creating content but of manufacturing objects, and their power came from controlling that infrastructure.

Digital technology severed content from carrier. A music file is not an object but a pattern of numbers, copyable perfectly at negligible cost. When Napster launched in 1999, students were sharing millions of songs with no product, no money, and no license, and industry revenue collapsed. The response — litigation, legislation, and DRM — shaped the internet for decades.

From Ownership to Access: The Streaming Model

The dominant response to digital disruption has been a shift from ownership to access. Consumers once paid to own a physical copy; now they pay a monthly fee for a catalog they do not own and that can be modified or withdrawn at any time. The IP implications are easily overlooked in celebrations of streaming's convenience.

Streaming reveals a gap between formal IP rights and economic power. An independent musician owns the copyright to their recordings but, to reach listeners, must accept terms set by a near-monopolist. The right is present; the bargaining power to extract value from it is not. That pattern — rights in creators, power in platforms — drives the critiques in the next section.

Critiques and Abuses of IP

A significant body of scholarship argues that IP law, whatever its original justifications, now concentrates power in incumbent corporations rather than rewarding genuine creation. Three scholars have been especially influential.

Case Studies in IP Abuse

The clearest illustration is Mickey Mouse, who first appeared in "Steamboat Willie" in 1928 — a work that under the law of the time would have entered the public domain in the 1980s. Disney lobbied successfully for extensions in 1976 and 1998, the latter sardonically called the "Mickey Mouse Protection Act," each time just before Mickey would have gone free. Steamboat Willie finally entered the public domain on January 1, 2024 — though Disney still holds trademark on the modern Mickey, so even that version cannot be used in ways that confuse consumers.

A second abuse is the "patent troll" (the industry's term: "non-practicing entity"): a company that owns patents but makes no products. It acquires vaguely worded patents covering broad concepts and threatens companies that do make things, extracting settlements cheaper than defense. A patent suit costs two to five million dollars to defend through trial, so settling for $100,000–$500,000 is rational even for a defendant certain to win. Personal Audio LLC famously sued podcasters over a patent on "episodic content distribution"; the Electronic Frontier Foundation invalidated it, after years. Trolls extract billions annually, most of it going to lawyers rather than inventors.

Big Tech and IP as a Competitive Weapon

Large technology companies use intellectual property not to recover investment costs but as an instrument of competitive advantage — IP as corporate strategy rather than reward for creators.

One pattern is "mutually assured IP destruction." Google, Apple, Microsoft, and Amazon each hold tens of thousands of patents, so any one that sued another could expect an immediate counterclaim. The giants cross-license implicitly, by threat alone. The détente does not extend to smaller competitors, who lack the portfolio to retaliate; for them, litigation from a well-resourced incumbent is a serious barrier to entry — which is, of course, the point.

Terms of Service as Private Law

Terms of Service agreements function as private IP law. A typical ToS takes a perpetual, worldwide, royalty-free license over anything the user uploads, lets the platform remove content and change the agreement at will, and requires arbitration rather than court — IP-like control without copyright's limitations: no fair use, no term limit, no public domain, in a contract of adhesion signed by users with no alternative.

The same logic runs through the right-to-repair fight. Buy an iPhone, a John Deere tractor, or a medical device and you get the object without full rights over it: manufacturers combine DMCA anti-circumvention provisions, parts pairing (software that rejects components they have not authenticated), and proprietary diagnostics to make independent repair impossible or illegal. A farmer locked out of a tractor's diagnostics may wait days for a technician rather than fix it in hours. If you cannot repair, modify, or resell a product, in what sense do you own it? Several states have passed right-to-repair laws since 2020, but the tension remains.

AI and Intellectual Property

Artificial intelligence creates two clusters of IP problems existing law is ill-equipped to handle: whether training on copyrighted works infringes, and who — if anyone — owns the copyright in what an AI generates. Public discussion runs them together, but they are separate questions with separate answers: as of 2026 one is settled and the other very much not.

Training Data and the Copyright Question

Large language models are trained on massive collections of text, including enormous quantities of copyrighted books, articles, and academic papers; image models on billions of images, including the work of living artists. The legal question is whether copying protected works to extract statistical patterns infringes — and it contains several separate questions, which courts have answered differently.

What the Courts Have Actually Held

It is commonly said that "the courts ruled AI training is fair use." That summary is too coarse, and in one respect simply wrong. The most consequential decision so far, Bartz v. Anthropic (N.D. Cal., 23 June 2025) , delivered no single verdict on "AI training" at all. Judge William Alsup pulled the conduct apart into three distinct acts and evaluated each separately.

Anthropic settled the third claim rather than take it to trial. The settlement — $1.5 billion plus interest, roughly 482,000 titles at about $3,000 per work, four times the $750 statutory minimum — was approved on 20 July 2026 and is the largest copyright class settlement in United States history. Notice what it was for. Not for training a model, already held fair use. For downloading and keeping the books in the first place.

This is the most useful thing to carry away: the fight is turning less on what models learn than on how the training corpus was acquired. A company that buys books, scans them, and trains on the results is on reasonably firm ground. One that torrents a shadow library to avoid the purchase price is not — and Anthropic's shortcut cost far more than the books would have. Note the irony: because the case settled, Alsup's reasoning was never tested on appeal. It is a district court opinion that binds nobody.

Market dilution is the theory to watch, and the case introducing it is routinely reported backwards. In Kadrey v. Meta (2025) , Judge Vince Chhabria granted Meta summary judgment on thirteen authors' training claim, and coverage read this as a second blessing for AI training. What Chhabria wrote was narrower: these plaintiffs lost on this record, having shown no evidence that Meta's models diluted the market for their books. He went out of his way to call market dilution a legitimate way of arguing the fourth fair-use factor — effect on the market — on which a better-prepared plaintiff could win. A separate claim about Meta's use of BitTorrent remains live. Again acquisition outlives training.

The one case where a fair-use defense has actually failed at trial level is Thomson Reuters v. Ross Intelligence (D. Del., 11 February 2025) , where Judge Stephanos Bibas held that Ross infringed 2,243 Westlaw headnotes in building a legal research tool. Two features limit its reach: Ross's system was not generative — it retrieved judicial language rather than producing new text — and it competed directly with the product it learned from, making the fourth factor easy for the plaintiff. Ross appealed, and that decision will be the first appellate word on any of this.

Outside the United States, the first judgment on generative-AI training resolved remarkably little. In Getty Images v. Stability AI ([2025] EWHC 2863 (Ch), 4 November 2025) , the English High Court took up Getty's claim that Stability trained Stable Diffusion on its photo library. Getty largely lost: it abandoned the training and copyright claims mid-trial, because the training happened outside the UK, beyond the court's reach, and its secondary-infringement claim failed on a consequential point — a model's weights are not an "infringing copy," since they do not store the images. Getty was left with narrow trade mark findings, settling nothing about training.

AI-Generated Content: The Authorship Question

The second cluster concerns what AI systems generate. If a language model writes a story, who — if anyone — holds the copyright? Here, unlike the training question, American law has a settled answer. The Copyright Office has long held that copyright requires a human author, and in Thaler v. Perlmutter the D.C. Circuit agreed, holding in March 2025 that the Act's "author" must be a human being . The Supreme Court denied review on 2 March 2026. Human authorship is no longer an open question.

What that means in practice was spelled out in Part 2 of the Copyright Office's report on copyright and artificial intelligence, published 29 January 2025 . Purely AI-generated output is not copyrightable. Where a human exercises genuine creative control — selecting, arranging, modifying — those contributions may be protected, case by case. The line that surprises students most: prompts alone do not confer authorship, however long and detailed. A prompt describes a desired result but does not determine the expressive choices the system makes in reaching it.

The companion report on training, Part 3, is a different matter: released 9 May 2025 as a "pre-publication" version, it still has not been finalized . Its central claim counterweights confident assertions in either direction — fair use for AI training cannot simply be presumed. Commercial use of vast troves of copyrighted work to generate competing content goes beyond established fair use, the Office argued, "especially where this is accomplished through illegal access." That is its view, not a court's holding — but it is the distinction Bartz drew, reached independently.

The rule has practical consequences. Purely AI-generated content enters the public domain at once, so a publisher has reason to obscure the AI's role in order to claim protection, leaving courts to decide how much human input is "enough." The Hegelian tradition suggests a deeper rationale: if moral rights are grounded in the expression of a creator's self, a work a human selects, directs, and refines has a stronger claim than one produced with a single prompt — roughly where the Copyright Office landed, by way of case law rather than German idealism.

Open Source and Alternative Models

The most important challenge to the incentive justification is not an argument but a fact: enormous quantities of high-quality work are produced outside the IP system entirely, through voluntary collaboration. Open source software is the most striking example, though not the only one.

The Open Source Movement

Linux — which runs most of the world's web servers, Android on billions of smartphones, and much of the internet's infrastructure — was not built by a corporation seeking IP profits. Linus Torvalds began it as a personal project in 1991, and thousands of volunteers have developed it since with no IP-derived revenue. On a typical website nearly every layer between you and the content is open source: the server, the database, the code generating the page, the browser rendering it. That stack was given away — by volunteers, and by employees of firms that found sharing cheaper than building privately. Either way, exclusivity was not the motivator.

The open source model uses IP law itself to ensure openness. "Copyleft" licenses like the GNU General Public License let anyone use, modify, and redistribute the software — but require any modified version to carry the same license. This "viral" property means no one can take a copyleft project and redistribute it as proprietary software: the moment you distribute, you must share your changes. "Permissive" licenses like MIT and Apache allow code into proprietary products. Both use copyright to achieve goals opposite to its traditional function.

If IP incentives are necessary for high-quality creation, how do we explain Linux, Wikipedia, and the thousands of other projects that form the de facto infrastructure of the internet? Researchers studying these communities find a mix of motives: professional reputation, community belonging, ideological commitment to open knowledge, and the benefit of improving tools one uses oneself. None require exclusivity.

Creative Commons

Founded in 2001 by Lawrence Lessig and others, Creative Commons provides standardized licenses for non-software content. Where traditional copyright is all-or-nothing, CC licenses let creators grant specific permissions in advance, spanning nearly public domain to only slightly looser than full copyright.

Wikipedia is the most prominent example: its content is CC BY-SA, so anyone can use, translate, or adapt its articles provided they attribute it and license derivatives on the same terms. OpenStreetMap, Khan Academy, and most open educational resources use CC licenses too — an intellectual commons the traditional IP system does not easily accommodate.

The AI era has complicated this. Many image generators and language models were trained on CC-licensed content, including images from Flickr and text from Wikipedia. Whether that matches the spirit, if not the letter, of licenses chosen to enable sharing is a live debate: creators who picked CC licenses for human adaptation may feel that training a commercial model on their work without compensation is a use they never authorized, even where the license text permits it.

Key Points

References