The Internet Was for Everyone. AI Should Be Too.: Historical Evolution, Platform Enclosure, and the Democratization of Generative Systems

Full report

On this page

Historical Chronology of Web and Artificial Intelligence Standards

The structural evolution of digital publishing over the past four decades demonstrates an ongoing systemic tension between open, royalty-free architectural protocols and centralized corporate enclosures1. In the early era of distributed computing, open protocols ensured that technical friction—while non-trivial—corresponded directly with absolute authorial autonomy and un-gatekept publishing2. The emergence of proprietary platforms and modern machine learning models has systematically eliminated technical entry barriers while simultaneously centralizing computational infrastructure, data sovereignty, and algorithmic discovery3. The following timeline provides a cited historical analysis of key technological, institutional, and economic shifts from the origin of the World Wide Web to the deployment of Transformer-based generative systems.

Comparative Structural Analysis: Open Protocols vs. Enclosed Models

The Evolution of Creative Gatekeeping and Technical Barriers

The early web architecture demanded functional mastery of semantic markup syntax, remote protocol execution, and client-server interactions4. Authors in the mid-1990s constructed homepages by manually composing HyperText Markup Language (HTML)—a structural language derived from SGML principles—specifying semantic tags such as headings, paragraphs, and hypertext anchors before transferring files via File Transfer Protocol (FTP) to remote web servers4. While this structural paradigm created a high initial barrier to technical entry, it guaranteed complete authorial control2. An author authoring raw HTML owned the underlying document source, commanded its rendering behavior, and published within an open ecosystem maintained by neutral standards bodies like the World Wide Web Consortium (W3C)2. Generative artificial intelligence radically alters this operational dynamic by shifting execution from explicit, rule-based syntax to natural language prompt interactions7. Systems built on self-attention Transformer mechanisms abstract away technical implementation details entirely15. A user no longer needs to comprehend programming languages, graphic rendering pipelines, or audio synthesis frameworks; articulating desired outcomes in natural language allows neural networks to synthesize functional code, high-resolution artwork, complex prose, or video sequences7. However, this elimination of technical friction introduces an ideological trade-off. Where early web authors retained complete authority over explicit source code and structural logic2, generative creators interact with statistical approximations derived from massive training datasets7. The creative gatekeeper has transitioned from an individual's capacity to master technical syntax to the implicit boundaries, corporate alignment filters, system prompts, and latent-space distributions enforced by model developers7.

Discovery Paradigms: Hyperlinks vs. Algorithmic Feeds and Generative Syntheses

The discovery graph of the early World Wide Web was structurally decentralized and relying on explicit human curation1. Information traversal depended on direct hyperlinks, static webrings, curated blogrolls, and topical directory nodes—such as GeoCities' 41 neighborhood categories2. Within this network topology, every node was anchored by a persistent Universal Resource Identifier (URI)2. Discovery was an active process of document traversal across an un-siloed graph where no single entity possessed the structural authority to demote, alter, or monetize an independent publisher's presence1. The rise of commercial web portals, centralized search engines, and social media platforms systematically reorganized discovery around algorithmic feed curation optimized for continuous user engagement3. Platform operators shifted online interaction from pull-based document navigation to push-based content streams3. Generative AI represents the culmination of this structural consolidation. Instead of functioning as indexers that route web traffic outward to independent primary sources via URIs2, large language models ingest the open web's collective corpus to synthesize direct, conversational answers within closed interfaces7. The traditional hyperlink—the structural link of the open web—is disintermediated inside generative response windows, concentrating user attention, interaction, and discovery within a singular proprietary interface2.

Compute Economics, Capital Intensity, and Resource Concentration

The physical infrastructure supporting the 1990s open web was commodity-based and highly distributed1. Early web servers operated on standard desktop workstations—such as the NeXT computer used by Berners-Lee at CERN—connected via standard TCP/IP network links1. Hosting a personal homepage required minimal disk space and negligible bandwidth5. The marginal cost of publishing approached zero, enabling hobbyists, educational institutions, and independent enthusiasts to run autonomous infrastructure without relying on centralized venture capital or cloud hosting conglomerates1. Conversely, modern generative AI architectures are characterized by extreme capital intensity and infrastructural centralization8. The training of foundation models relies on massive GPU compute clusters designed to execute parallel linear algebra calculations7. As detailed in the groundbreaking 2017 Transformer specification, training the baseline "Transformer big" model required [Figure omitted from source export] floating-point operations (FLOPs), utilizing eight NVIDIA P100 GPUs continuously for 3.5 days8. Contemporary frontier models require tens of thousands of specialized accelerators running for months, pushing training costs into hundreds of millions of dollars8.

This compute bottleneck produces three profound economic consequences:

• Training Oligopolies: The astronomical capital required for state-of-the-art training runs restricts frontier model development to hyper-capitalized technology corporations8.
• Inference Dependency: Unlike static HTML files rendered client-side by open browsers at near-zero incremental cost2, generative inference requires server-side GPU processing for every output token, subjecting users to ongoing API fees and subscription metering8.
• Commons Enclosure: Foundation models systematically ingest public expression harvested from the open web4, converting shared human knowledge into proprietary latent weights monetized behind corporate paywalls7.

Evaluation of the Historical Analogy

Evaluating the historical parallel between the early open web and the generative AI ecosystem reveals significant strengths alongside critical structural divergences. The core strength of the analogy lies in democratization of production capability. Just as early web protocols and personal homepage hosts like GeoCities allowed non-journalists and amateur writers to globally publish media for the first time3, generative AI allows non-programmers, non-illustrators, and non-composers to synthesize functional software, artwork, and music7. Both eras mark a profound compression of the execution gap between human creative intent and final media distribution6. However, the analogy breaks down across three fundamental structural dimensions:

• Protocol Ownership vs. Model Black Boxes: The early web was constructed on royalty-free, transparent specifications2. The "View Source" capability of early browsers allowed any user to inspect, learn from, and replicate web pages2. Modern generative models are closed, opaque black boxes whose internal attention weights, training provenance, and safety alignments are obscured from public scrutiny7.
• Marginal Cost Dynamics: A static HTML document can be hosted indefinitely on distributed servers at virtually no ongoing cost1. Generative AI outputs require energy-intensive, real-time GPU inference for every interaction, making long-term access vulnerable to platform pricing changes, API throttling, or corporate decisions8.
• Archival Resiliency: When GeoCities was shuttered by Yahoo in 20093, volunteer initiatives like Archive Team were capable of scraping and preserving tens of millions of raw HTML files because the underlying format was non-proprietary3. If a commercial generative AI provider terminates an API or alters a base model's behavior, the latent generation capabilities and historical prompt dynamics vanish permanently, leaving no inspectable source files behind3.

Complete Documentary Script: "The Internet Was for Everyone. AI Should Be Too."

Target Runtime: 10:00 – 12:00 Minutes Narrative Perspective: First-person voice of a software developer who began building web pages in the mid-1990s. Production Tone: Highly technical, historically grounded, reflective, and critically analytical.

Prologue: The Sound of a 28.8k Handshake (00:00 – 01:45)

Visual Cues

• Pitch black screen. High-fidelity audio of a 28.8k dial-up modem handshaking, featuring squeals, static, and dual-tone chimes.
• A vintage cathode-ray tube (CRT) monitor flickers to life in a dark room. An amber command cursor blinks inside an MS-DOS prompt.
• Macro lens shot of hands typing raw markup into Windows 95 Notepad: \<html\>, \<head\>, \<title\>My First Homepage\</title\>, \<body\>.
• Fast-paced archival visual montage: Early university computer labs, physical copies of 1990s computer magazines, NeXT computer workstations at CERN1, television broadcast clips explaining "The Information Superhighway."
• Clean monospace graphic text overlay across a dark grid background: THE INTERNET WAS FOR EVERYONE. AI SHOULD BE TOO.

Sound Design

• Diegetic dial-up audio blending into a warm, analogue synth pad. Keystrokes synchronized with retro mechanical keyboard click audio.

First-Person Narration

If you weren’t writing code in 1995, it is difficult to convey what the digital world felt like2. It wasn’t a utility dominated by four corporate conglomerates holding the keys to public discourse1. It was an open continent—and someone had just handed us the blueprints1. I built my first web page in the summer of 1995 using Windows Notepad and a shell account4. There were no build tools, no JavaScript frameworks, and no cloud providers issuing monthly infrastructure bills. There was only raw markup, a dial-up modem, and a web browser that let you right-click any page on earth and select 'View Source'4. That simple browser function—'View Source'—was a revolutionary architectural philosophy2. It declared that if someone built something remarkable, you had the right to inspect the mechanics, learn how it functioned, copy the markup, and build something better2. When Sir Tim Berners-Lee created the World Wide Web at CERN, he explicitly refused to patent the technology1. CERN released the client and server source code into the public domain on a royalty-free basis4. The foundational premise of that era was unambiguous: This is for everyone1.

Act I: The Wild West of Personal Homepages (01:45 – 04:15)

Visual Cues

• CRT screen displaying Netscape Navigator 2.0 rendering personal homepages hosted on GeoCities. Animated GIF construction banners, hit counters, tiled background textures, neon text elements.
• Dynamic visual map detailing GeoCities' 41 neighborhood categories: SiliconValley, Area51, Athens, EnchantedForest, Hollywood6.
• Archival footage of GeoCities founder David Bohnett discussing virtual communities6. Smooth scrolling footage through preserved homepages preserved by the Internet Archive3.
• Split-screen visual layout: Left panel shows raw, hand-authored HTML code; Right panel shows the corresponding rendered web page rendering in real time.

Sound Design

• Upbeat 16-bit electronic chiptune melody transitioning into rhythmic analog synth pulses. Soft mouse clicks audible beneath narration.

First-Person Narration

Before corporate platforms dictated how your personal identity had to look inside standardized profile cards, we had GeoCities6. Founded in the mid-1990s, GeoCities didn't organize human expression by algorithmic engagement loops or demographic ad profiles6. It organized the web into neighborhoods6. If you were passionate about computer hardware, you established a homepage in SiliconValley6. If you were obsessed with science fiction, you claimed a parcel in Area516. Millions of everyday people—students, tradespeople, hobbyists, artists—learned structural HTML syntax for the sole purpose of claiming a stake in global communications4. The visual outcomes were chaotic—tiled starfield backgrounds, autoplaying MIDI audio files, spinning hit counters5. But those homepages belonged entirely to their creators6. We authored the markup4. We maintained the directory structures. We didn't seek algorithmic approval to be discovered; we traded direct hyperlinks through webrings and blogrolls2. The technical requirements were real—you had to understand tags like \<p\>, \<h1\>, and \<a href\>—but once you mastered syntax, no corporate entity could alter your text or suppress your reach behind an engagement filter2.

Act II: The Great Enclosure and the Death of the Open Web (04:15 – 07:00)

Visual Cues

• Dramatic lighting change to cold blue tones. Historical newspaper headline scan: Yahoo\! To Buy GeoCities for \$4.58 Billion in Stock (January 1999\)13.
• Graphic animation: A network graph showing independent personal sites collapsing into three massive monolithic centralized hubs labeled Search, Social Feed, and Video Portal3.
• Archival web page capture from October 2009: GeoCities has officially shut down. Screen glitch effect across thousands of homepages transforming into broken image icons and 404 error displays3.
• Text graphic overlay: 38,000,000 Homepages Deleted3. Visual showcase of Archive Team servers frantically downloading web pages during the final shutdown countdown3.

Sound Design

• Music drops to an ominous, low-frequency bass drone. Industrial server hum and tape-stop audio effects.

First-Person Narration

The open web was too decentralized for corporate monetization6. In January 1999, at the height of the dot-com speculative boom, Yahoo\! acquired GeoCities for \$4.58 billion13. The corporate strategy was clear: consolidate millions of independent web pages into a singular monetized web portal3. That acquisition marked the beginning of the Great Enclosure3. Over the subsequent decade, technology platforms recognized that open protocols didn't maximize advertising yields3. They systematically migrated users into walled gardens3. Instead of teaching citizens how to write HTML, social networks replaced personal homepages with standardized form fields4. Instead of hyperlinks connecting independent servers, platforms introduced algorithmic feeds designed to maximize session duration3. Then came October 26, 20093. Yahoo permanently terminated GeoCities US3. In a single day, thirty-eight million personal homepages—fifteen years of grassroots human cultural history—were erased3. Hyperlinks broke across the global network3. It was an unmistakable lesson in digital landlordism: when a commercial platform controls your hosting infrastructure, your cultural output can be deleted with a single corporate operational decision3.

Act III: The Generative Spark (07:00 – 09:30)

Visual Cues

• Visual transition to high-contrast, clean modern aesthetic. PDF preview of arXiv:1706.03762: Attention Is All You Need15.
• 3D motion graphic illustrating the Transformer architecture: Query, Key, and Value vector matrices, multi-head self-attention paths linking contextual relationships across word sequences15.
• Screen recordings demonstrating generative tool workflows:

1. A user types a single natural language prompt into a command line interface.
2. An AI model generates full, functional web application code in real time.
3. A diffusion model synthesizes a intricate digital painting7.
4. A text-to-video model generates high-definition visual footage from a prompt7.

• Montage of creators—writers, musicians, non-technical artists—using generative systems to execute creative projects.

Sound Design

• Uplifting electronic instrumentation building in complexity. Precise digital click sounds and audio pulses corresponding to visual data rendering.

First-Person Narration

In June 2017, eight artificial intelligence researchers published a paper titled Attention Is All You Need15. They introduced the Transformer—an architecture that abandoned recurrent neural networks in favor of parallel self-attention mechanisms15. The Transformer allowed software to analyze complex contextual relationships across massive amounts of data simultaneously15. When generative AI tools entered public adoption, it felt like 1995 all over again1. The technical syntax barriers that had prevented millions of people from writing software, composing music, or producing digital animation were eliminated almost instantly7. Suddenly, an individual without a computer science degree could type a natural language prompt and generate full-stack code, synthesize visual art, or construct complex data visualizations7. You no longer needed mastery over compiler syntax or expensive software production suites7. The execution gap collapsed7. Creative intent was uncoupled from technical gatekeeping7.

Act IV: Compute Monopolies and the Enclosure of Intelligence (09:30 – 11:30)

Visual Cues

• Dark, cinematic visuals inside a hyperscale cloud data center. Infinite aisles of high-density server racks featuring flashing green and blue LEDs. Industrial cooling fans rotating at high speeds.
• Data visualizer graphic showing computational scale growth: 2017 Transformer Big Training: 2.3 x 10^19 FLOPs8 \-\> Contemporary Frontier Models: Tens of Thousands of GPUs, Hundreds of Millions in Capital.
• Side-by-side graphical comparison:
• Left Side (1995): Desktop server running static HTML files locally for minimal cost1.
• Right Side (Present): Closed commercial API gateway charging per token, enforcing proprietary content filters, metering access behind monthly subscriptions.

Sound Design

• Low industrial drone, deep mechanical ventilation hum, accompanied by tense, atmospheric cello and synth arrangements.

First-Person Narration

Herein lies the structural paradox of the generative AI revolution7. While generative tools dramatically lower skill barriers for individual creators, they radically consolidate power at the physical infrastructure level8. Writing an HTML file in 1995 required negligible computational power1. You could host a web server on a standard desktop computer in your bedroom1. HTML was an open protocol governed by a non-profit international consortium2. Nobody owned the underlying markup specification2. Modern artificial intelligence operates under completely different economic realities7. Training a frontier foundation model requires billions of parameters, exaflops of computational energy, and tens of thousands of specialized liquid-cooled graphics processors running inside multi-billion-dollar datacenters8. The foundation of generative AI is not an open, royalty-free protocol; it is extreme capital concentration8. A small group of hyper-capitalized technology corporations scraped thirty years of human writing, art, and source code from the open web—work built by open-web creators4—and converted that shared public commons into proprietary model weights7. They placed those weights behind gated APIs, monthly subscription tiers, and opaque content policies. We transformed the public open web into private computational property3.

Epilogue: Who Controls the Conversation? (11:30 – 12:00)

Visual Cues

• Slow zoom on a modern laptop screen in a dimmed workspace. On the left monitor panel, an archived 1990s GeoCities HTML page runs inside a web browser; on the right monitor panel, a modern conversational AI prompt terminal blinks.
• Montage of open-source artificial intelligence communities, decentralized compute networks, local model inference terminals, and open-weight model repositories.
• Final visual text graphic: DEMOCRATIZE THE MODEL. DECENTRALIZE THE COMPUTE. RECLAIM THE COMMONS. Screen fades out to black, leaving only a glowing white terminal cursor blinking in the center.

Sound Design

• Music swells to an emotional, reflective climax before resolving into a quiet piano chord. Audio of a single final keystroke.

First-Person Narration

We stand at the exact same architectural crossroads we faced thirty years ago1. Generative AI has demonstrated something undeniable: human creative ambition was never the bottleneck6. Given accessible tools, human beings possess an extraordinary desire to write, construct, design, and express6. Which brings us back to the ultimate question: Did we build the internet so everyone could express themselves, or only so professionals could control the conversation? \[cite: 1, 6\] If artificial intelligence is to fulfill the historic promise of the early web, we cannot allow it to become another locked corporate portal1. We require open-weight foundation models, public computational infrastructure, and decentralized protocols that grant creators true data sovereignty1. The internet was built for everyone1. Artificial intelligence must be too. Right-click. View source. Reclaim the open network.

Archival Production Visual Guide

To ensure historical accuracy, the documentary production should pair the script with authentic archival assets, visual framing parameters, and motion design assets outlined below.

Strategic Governance and Policy Implications

The structural parallels between the historical enclosure of the early web and the current consolidation of generative artificial intelligence provide critical governance insights for technology policy. The migration from open standards (HTML, HTTP) to proprietary platforms resulted in systemic link rot, corporate gatekeeping, and the loss of grass-roots digital culture—exemplified by Yahoo's deletion of 38 million homepages3. To prevent generative systems from repeating this path toward platform enclosure, four strategic interventions are required:

1. Mandating Open Model Artifacts and Data Transparency: Just as royalty-free specifications maintained by the W3C preserved the open web2, public policy must establish regulatory incentives for open-weight model releases, data provenance documentation, and standardized licensing frameworks1.

2. Public Compute Infrastructure and Research Commons: To address the severe capital bottleneck where training runs require tens of millions of dollars in hardware8, democratic governments must fund public compute clusters. Providing academic institutions, non-profit entities, and independent creators with subsidized GPU access prevents frontier AI development from remaining an exclusive corporate oligopoly8.

3. Interoperable Standards and Local Inference Optimization: Developing universal specifications for model interfaces, prompt formats, and consumer-grade hardware quantization ensures that individuals can execute generative models locally, freeing creators from ongoing API metering and subscription gatekeeping7.

4. Data Sovereignty and Fair Attribution Frameworks: Aligning with early web architecture that prioritized explicit attribution via persistent URIs2, generative architectures must incorporate technical mechanisms that trace training provenance, compensate source creators, and respect user data sovereignty1.

Works cited

1. Tim Berners-Lee \- W3C, https\://www\.w3.org/People/Berners-Lee/

2. Berners-Lee \- A Short History of the WWW (1) \- W3C, https\://www\.w3.org/2007/Talks/0920-webhistory-tbl/

3. Yahoo closes GeoCities: how the web lost millions of home pages, https\://vnunet.com/history/yahoo-closes-geocities

4. History of HTML — From GML to the Living Standard, https\://www\.htmlwasher.com/history-of-html/

5. 2 \- A history of HTML \- W3C, https\://www\.w3.org/People/Raggett/book4/ch02.html

6. A Tale of Deleted Cities \- CHM \- Computer History Museum, https\://computerhistory.org/blog/a-tale-of-deleted-cities/

7. Attention Is All You Need \- Wikipedia, https\://en.wikipedia.org/wiki/Attention\_Is\_All\_You\_Need

8. Attention Is All You Need | alphaXiv, https\://www\.alphaxiv.org/abs/1706.03762

9. History | About us \- W3C, https\://www\.w3.org/about/history/

10. Information Management: A Proposal, https\://www\.w3.org/History/1989/proposal.html

11. The World Wide Web: Past, Present and Future \- W3C, https\://www\.w3.org/People/Berners-Lee/1996/ppf.html

12. Making the Web For Everyone \- The History of the Web, https\://thehistoryoftheweb.com/opening-up-the-web/

13. Yahoo\! To Buy GeoCities \- CBS News, https\://www\.cbsnews.com/news/yahoo-to-buy-geocities/

14. Yahoo\! Inc. \- Company Histories, https\://www\.company-histories.com/Yahoo-Inc-Company-History.html

15. Attention Is All You Need \- arXiv, https\://arxiv.org/html/1706.03762v7

16. \[1706.03762\] Attention Is All You Need \- arXiv, https\://arxiv.org/abs/1706.03762

17. Web History Primer \- W3C, https\://www\.w3.org/2012/08/history-of-the-web/origins.htm

18. ATTENTION IS ALL YOU NEED. Original Paper by Ashish Vaswani, https\://medium.com/@visshu78/all-you-need-is-attention-75bb845d1827

19. Story of a Brand: Yahoo\! \- Logotech, https\://www\.logotech.com/blog/post/story-of-the-yahoo-brand

20. Yahoo completes GeoCities acquisition \- CNET, https\://www\.cnet.com/tech/services-and-software/yahoo-completes-geocities-acquisition/