Introduction: A Watershed Moment for Content and Knowledge

Today marks a significant career evolution for Cloudflare as John Graham-Cumming transitions from his role as Chief Technology Officer to join the company’s board of directors. I’ve known John professionally for many years and watched his thoughtful approach to technology challenges transform Cloudflare into an essential pillar of internet infrastructure. His technical vision has protected and accelerated millions of websites worldwide, and he’ll continue to drive innovation from his new board position.

Among John’s many contributions to technology discourse, I’ve been particularly captivated by an analogy he explored in tweets and blog posts, comparing pre-ChatGPT content to low-background steel. While his specific tweet on this subject is no longer available (as he took a hiatus from Twitter for a period), the brilliance of this comparison has continued to resonate throughout tech communities. This parallel offers profound insights for those of us in media, journalism, and AI – industries where I’ve spent decades navigating technological disruption as former Chief Technology Officer of The New York Times, Chief Product Officer & CTO of The Wall Street Journal, and as a technology leader at numerous other media organizations.

Having witnessed multiple paradigm shifts in publishing and content – from print to web, web to mobile, and now the AI revolution – I believe this analogy offers a uniquely powerful lens through which to understand our current moment. It deserves serious examination, especially by those responsible for creating, curating, and preserving our collective knowledge.

The Origins of an Analogy

Before diving into implications, it’s important to acknowledge that while John Graham-Cumming has helped popularize this analogy through his work and advocacy, the comparison appears to have emerged organically across the tech community. Programmer and artist Kyle McDonald made a similar observation in an X (formerly Twitter) post on December 5, 2022, noting that just as particle detectors require special steel manufactured before nuclear contamination, future AI might need training data collected before 2022. Others independently reached similar conclusions around the same time, showing how this parallel resonated naturally with many observers watching AI’s rapid evolution.

The convergence of multiple thinkers on this analogy speaks to its explanatory power. As we in journalism know well, when multiple independent sources arrive at the same metaphor, it often signals a truth that deserves attention.

The Science of Low-Background Steel

Low-background steel, also known as pre-nuclear, pre-atomic, or pre-war steel, refers to steel produced before widespread nuclear weapons testing and use, resulting in lower levels of radioactive contamination compared to modern steel. It’s called “low-background” because of its naturally lower levels of background radiation, making it desirable for applications needing minimal radioactive interference, such as particle detectors and some medical equipment.

To understand this analogy, we must first grasp what makes low-background steel special. Before July 16, 1945 – the date of the first nuclear test codenamed “Trinity” – all steel was produced in an atmosphere free of artificial radioactive isotopes. After nuclear testing began, the atmosphere worldwide became contaminated with radionuclides that were incorporated into steel during production, as the manufacturing process involves passing air through molten metal.

This contamination, while too minimal to affect everyday use, created a significant problem for certain scientific instruments. Devices like Geiger counters, radiation detectors, and components for particle accelerators at institutions like CERN require materials with extremely low background radiation to function accurately. For these sensitive applications, steel produced before the nuclear age became uniquely valuable.

The only source of this “pure” pre-nuclear steel today is salvage from old structures, shipwrecks, and decommissioned vessels manufactured before 1945. Scientists and engineers have gone to extraordinary lengths to obtain this material – salvaging steel from the German fleet scuttled at Scapa Flow after World War I, retrieving metal from 100-year-old bridges, and even recovering components from shipwrecks on the ocean floor. This scarcity has made pre-war steel a precious resource for specific scientific purposes.

The Digital Parallel: Pre-ChatGPT Content

The release of ChatGPT in late 2022 represents a similar watershed moment for digital content. Before widely accessible AI text generators, online content was exclusively human-authored. Now, an increasing proportion of internet text is AI-generated or AI-influenced, creating a new kind of “background radiation” in our information ecosystem.

Just as nuclear testing permanently altered the composition of the atmosphere and consequently all steel produced thereafter, the widespread adoption of generative AI is fundamentally changing the nature of content creation. This transformation raises important questions about the future value of demonstrably human-authored content.

As a media technology executive who led digital transformation at The New York Times, The Wall Street Journal, and other major publications, I’ve observed how content ecosystems evolve with new technologies. Each transition – from print to web, from desktop to mobile, from text to multimedia – changed not only how we produced content but also how readers consumed and valued it. The shift to AI-influenced content represents perhaps the most profound transformation yet.

Consider what’s happening across newsrooms, publishing houses, and content farms worldwide: AI is being deployed to generate first drafts, summarize information, translate content, and even produce entire articles on topics like sports scores, financial reports, and weather updates. The flood of synthetic content is already altering the digital landscape in ways that parallel how nuclear fallout changed our physical atmosphere – invisibly but irrevocably.

Tech blogger Matt Rickard captured this concern in his June 2023 essay “The Low-Background Steel of AI,” noting that modern datasets are increasingly “contaminated” with AI-generated text, making purely human data harder to find. The scientific publication community has begun discussing whether papers need to be labeled as pre-ChatGPT or post-ChatGPT, recognizing that research methodologies and writing styles may be fundamentally different after the widespread adoption of these tools.

Why the Analogy Matters for Media and AI

The comparison between pre-ChatGPT content and low-background steel is compelling for several reasons that should concern everyone in journalism, publishing, and AI development:

1. Historical Dividing Line: Both scenarios feature a clear “before and after” moment that fundamentally changed their respective landscapes. In media, we’ve experienced several such moments – the launch of the web, the introduction of social media, the shift to mobile. None, however, has altered the fundamental nature of content creation itself the way generative AI has.
2. Contamination Concept: In both cases, a new technological development introduced a form of “contamination” that became impossible to avoid in new production. For publishers, this means even carefully human-written articles now exist in an ecosystem where they compete with and may be influenced by AI-generated content.
3. Value for Special Applications: Pre-nuclear steel and pre-AI content both hold special value for specific applications requiring purity or authenticity. In publishing, these applications include training future AI models, establishing historical records, conducting linguistic research, and maintaining authoritative reference materials.
4. Feedback Loops: When AI systems train on content that includes AI-generated material, they risk a form of “model collapse” through amplification of their own patterns and quirks – a digital parallel to how radiation detection equipment built from contaminated steel would give false readings. Researchers from institutions like Stanford University have demonstrated that when language models train on outputs from other language models, the quality deteriorates dramatically over generations, eventually producing incoherent or highly repetitive text.
5. Preservation Challenges: Just as scientists had to develop methods to identify and preserve low-background steel, media organizations now face the challenge of preserving and authenticating pre-AI content as a valuable resource. This raises questions about archiving, metadata standards, and content provenance that publishers have only begun to address.

At ScalePost.AI, where I serve as a founding technology advisor, we’re keenly aware of these challenges as we work to strengthen collaboration between media companies and AI developers. The distinction between human-authored and AI-generated content is crucial for responsible AI development and trustworthy information ecosystems.

A 2023 article in Scientific American described how AI researchers have observed model degradation when training on synthetic data, noting that “some insiders believe a similar cycle [to low-background steel] is set to repeat in generative AI – with training data instead of steel.” This concern has led to calls within the AI research community for preserving pre-2022 datasets specifically for training and benchmarking purposes.

Where the Analogy Breaks Down

Like all analogies, this one has limitations that deserve careful consideration, particularly by those of us who have spent careers at the intersection of media and technology:

1. Nature of Contamination: Steel’s radioactive contamination is physical and measurable, while content “contamination” by AI is conceptual and contextual. The presence of radioactive isotopes in steel is an objective fact; the influence of AI on content creation exists along a spectrum and depends heavily on how the technology is deployed.
2. Detectability: Radioactivity can be precisely measured with a Geiger counter, while distinguishing AI-generated content from human writing remains challenging and imperfect. This creates significant challenges for publishers, researchers, and platforms trying to maintain content provenance.
3. Continued Production: Unlike steel, which cannot be produced without exposure to today’s atmosphere, humans continue to create original content alongside AI. The question isn’t whether human-authored content will disappear, but how it will be authenticated, valued, and distinguished in an environment increasingly populated by synthetic text.
4. Intentionality: The radioactive contamination of steel was an unintended consequence of nuclear testing; the integration of AI into content creation is often deliberate and can be controlled. Publishers, journalists, and content creators have agency in how they adopt these tools.
5. Value Judgment: While radiation in steel is objectively detrimental to certain scientific applications, the presence of AI influence in content isn’t inherently negative. Some AI contributions to content – fact-checking, translation, accessibility features – may actually enhance quality and reach.

These distinctions matter for media organizations planning their content strategies. During my tenure as CTO at The New York Times and The Wall Street Journal, I learned that new technologies rarely fit cleanly into existing frameworks. They require nuanced approaches that acknowledge both their transformative potential and their limitations.

Despite these nuances, the core insight remains powerful: a technological breakthrough has permanently altered our information landscape, potentially creating special value for content from the “pre-AI” era. The tech blog The Luddite argued that the comparison to pre-atomic steel “is a very valid analogy” for understanding how synthetic content might influence our knowledge ecosystem, even while acknowledging the parallel isn’t perfect.

The Media Industry Response: Credibility in the Age of Synthetic Content

For media organizations, the low-background steel analogy prompts urgent questions about content provenance, archiving practices, and the future value of journalistic work. Having led technology strategy at major news organizations through multiple digital transitions, I see several imperatives emerging from this analogy:

1. Authentication and Provenance: Just as scientists needed methods to verify the age and purity of steel, publishers need robust systems to authenticate human-authored content. Blockchain-based verification, cryptographic signatures, and metadata standards will become crucial infrastructure for trustworthy publishing.
2. Strategic Archiving: Media organizations should consider creating secure, authenticated archives of pre-2022 content with clear provenance information. These archives may become invaluable assets for future research, AI training, and establishing historical records.
3. Content Labeling and Transparency: Clear disclosure of AI involvement in content creation – whether it’s generation, editing, fact-checking, or enhancement – will become an essential practice for maintaining reader trust.
4. New Economic Models: The scarcity dynamic that makes low-background steel valuable may create similar value for verifiably human-authored content in specific contexts. Publishers should explore how this might influence subscription models, syndication rights, and content licensing.
5. Specialized Training: Just as scientists had to learn new techniques for working with low-background materials, journalists and content creators need training to effectively collaborate with AI while preserving their distinct human contributions.

Some publishing organizations have already begun responding to these challenges. The Authors Guild has launched a human-authored certification program to help readers identify work created without generative AI. Major news organizations are developing explicit policies around AI use in their newsrooms. Academic publishers are updating their submission guidelines to address AI assistance in research papers.

Industry Leadership in the AI Era

Companies like Flatiron Software and Snapshot AI, where I recently joined as President, are addressing these emerging challenges. Flatiron Software builds AI-powered engineering teams that deliver software solutions while maintaining human expertise at the core of their operations. Snapshot AI is developing tools to understand engineering performance through AI-driven analytics that go beyond conventional metrics.

Both organizations recognize that as AI becomes integrated into content creation and software development, distinguishing human expertise from algorithmic patterns will become increasingly valuable. Just as scientists developed methods to identify and source low-background steel for critical applications, we need new frameworks for authenticating, preserving, and valuing human-created digital content.

At ScalePost.AI, we’re specifically focused on strengthening the relationship between media publishers and AI developers. We recognize that both sides of this equation need each other – publishers need AI tools that respect their content value and attribution needs, while AI developers need high-quality, authentic content sources. Building these collaborative models will be essential as we navigate the post-ChatGPT landscape.

Three Strategic Imperatives for Media Leaders

Based on my experience leading technology strategy at major media organizations and the insights from the low-background steel analogy, I believe three strategic imperatives emerge for news and publishing organizations:

1. Preserve and Authenticate: Media organizations should immediately implement systematic approaches to preserving and authenticating their pre-AI content archives. This isn’t merely about maintaining historical records—it’s about protecting a potentially valuable asset whose worth may only increase with time.
2. Develop Hybrid Models: The future won’t be purely human or purely AI. The most successful media organizations will develop sophisticated hybrid models that leverage AI capabilities while preserving and highlighting distinctive human contributions. Having led newsroom technology transitions from print to digital at The New York Times and The Wall Street Journal, I’ve learned that embracing new tools while maintaining core journalistic values is essential for successful transformation.
3. Invest in Content Provenance Infrastructure: The ability to verify the origin, authenticity, and modification history of content will become a critical capability for publishers. Investment in robust provenance infrastructure now will pay dividends as the information ecosystem becomes increasingly complex.

Programmer Glenn Keighley suggested in a recent LinkedIn post that we might need “protected archives of verified human-created content, similar to how pre-nuclear steel has been preserved.” This approach recognizes that while we can’t prevent the proliferation of AI-generated content, we can take steps to preserve what came before and establish systems for authenticating what comes next.

Conclusion: A New Era for Content Creation and Consumption

As John Graham-Cumming begins his new chapter on Cloudflare’s board, his insightful perspectives on technology’s impact continue to illuminate our understanding of the digital world. The low-background steel analogy that he and others in the tech community have explored reminds us that technological progress often creates unexpected dividing lines in history, with implications that reach far beyond their intended applications.

The release of ChatGPT and similar technologies has created a genuine inflection point for media and publishing—comparable to the introduction of the printing press, the development of broadcast media, or the rise of the internet. Each of these shifts fundamentally changed how information was created, distributed, and consumed. The AI revolution may prove equally transformative.

Whether pre-ChatGPT content will become a treasured resource like low-background steel remains to be seen. What’s certain is that we’re living through a pivotal moment in information history—one that will likely be studied and referenced for generations to come. As we navigate this new landscape, those of us who have spent careers at the intersection of media and technology have both an opportunity and a responsibility to shape how these powerful tools are integrated into our information ecosystem.

The parallel to low-background steel offers a compelling framework for understanding what’s at stake. Just as that pre-nuclear material enables certain scientific measurements that would otherwise be impossible, pre-AI content may preserve patterns of human thought, creativity, and expression that could otherwise be diluted or lost in a sea of synthetic content. Recognizing this value doesn’t mean rejecting AI’s potential—rather, it means approaching these tools with a clear understanding of what we might gain and what we might lose as we forge ahead.

In my roles at Flatiron Software, Snapshot AI, and as an advisor to ScalePost.AI, I’m committed to developing approaches that harness AI’s capabilities while preserving the distinctive value of human expertise and creativity. This balanced approach will define the next era of media, publishing, and information technology—one where artificial and human intelligence complement rather than replace each other.

Rajiv Pant is President at Flatiron Software and Snapshot AI , where he leads organizational growth and AI innovation while serving as a trusted advisor to enterprise clients on their AI transformation journeys. He is also the founding technology advisor to ScalePost.AI , a platform elevating collaboration between media companies and AI developers. With a background spanning CTO roles at The Wall Street Journal, The New York Times, and other major media organizations, he brings deep expertise in language AI technology leadership and digital transformation. He writes about artificial intelligence, leadership, and the intersection of technology and humanity.