A Primer on Paper: The New Media Machines

For all of you new subscribers out there, welcome to our old-new-school web site blog! We’re still working on our rotating .gif art, so please be patient. 

Here at Machines on Paper, we’re trying to build some much-needed connective tissue between the worlds of media and machine learning. 

Anyone who has worked with me knows about a phrase I always come back to: To move the world, you need a lever and a place to stand. Or, put differently: A good idea will not change anything without a strong point of view.

So, in this post, we’ll climb up a hill and survey the overall landscape in media and AI. I’ll share my thoughts on what the hell has happened in news media, and why it finds itself in such a vulnerable state to this era of LLM ascendancy. 

And no, I don’t think news or entertainment composition are especially vulnerable to replacement from AI. Instead, these industries are vulnerable to markets, individuals and companies with so little knowledge of practical creative processes that the “solutions” they impose might spell the hibernation of things like national-scale high quality reporting for a generation. [Note: This is where I stand. Quick, someone toss me a lever!]

The biggest questions right now for the media industry, regardless of whether LLMs ever match the economic impact of the Fax machine, are:

1. To what extent will media orgs be involved with their own modes of distribution?
2. What will happen to the information economy if people stop visiting news outlets and get their info directly from LLM powered apps?

What Happened to The News? 

The News (caps intended) as we know it is optimized for broadcast styles of communication: One central report being distributed to the masses. 

Then came erosion in the social media era, where important bits of news were simply diffused in the air. And finally, the slow (apparent) collapse of The News we’re witnessing in this recommendation era, where you will encounter enough diffused bits of news so long as you’re the kind of person who keeps clicking on profitable content no matter what’s in the news. 

In this recommendation era, the ability of media orgs to build an actionable understanding of their audience is falling apart. Meanwhile, the ability of individuals to understand the specific context of a single piece of The News decreases as time goes on. (After all, even intrepid reporters experience the internet via black box recommendation cluster.)

This is the crisis in communications: The collapse of observers and subjects in media. We’ll have to dive into that idea more specifically as this site lives on.

Where Are We?

It occurs to me that the above header feels a bit sanitized. If you work in news media, your questions are probably more like: Oh God, I’m so tired, what’s going on? Did I miss anything? Am I already screwed?

I am here to tell you right now: You aren’t screwed… yet. And none of us are, so long as we mind our core values and stay ready to subsume our egos to build a viable future. The art and craft of creation go on, but golden-age media formalisms like inverted pyramid news article or its kin in filmmaking may not for much longer.

I hear from my contacts in news and advertising that there’s lots of experimentation amongst individuals, but few systematic process changes that are actually working passably at scale. (Andrew Jardine from Huggingface has a wonderful post up on LinkedIn making a similar point.)

You may also hear a lot of idle talk amongst more analytically-minded ML practitioners about replacing functions like “journalist” with ML models or drastically reducing the amount of people involved in production of written content. (Who, where, which people? Are they in the room with you now? … If that was your reaction, the previous sentence was not for you. Sadly, I cannot yet deliver a single article by affinity cluster.) 

I call this "idle talk" not because no one is trying to move in these directions. I say 'idle' because these plans don’t make any sense, and I’m trying to be kind. (We’ll talk about that in more detail in a future post.)

For news in particular, two potentially killer apps seem to be developing.

1. AI Search and Aggregation

The question for media is likely: Which large media companies will sign onto a deal with the genAI giants, and what is the nature of the scraps everyone else will get?

2. Subject-specific AI publications

News update sites in the style of Political Wire are likely spoof-able using LLM systems in a “not ideal, but good enough” kind of way. Online creators will be pushed to come up with new forms of language and expression, as a mass proliferation of AI-generated websites would first degrade 'legacy' creative expressions, and then destroy themselves as they consume an ever-increasing share of their own content.

Will every unit of success in the AI info delivery industry further dismantle what remains of journalistic-quality information on the open web, thereby damaging the long term prospects of mass-scale AI informational business models? What will happen if the creation economy collapses and the cost of procuring new, high-quality training data increases towards infinity?

Think of it this way if you’re astronomy-inclined: Will LLMs lead to a big bang, or a big crunch in which new expressions of information are created only to be eventually subsumed by AI, upon which the cycle starts again?

How Should We Judge the Impact of AI?

I am rather fond of the aforementioned Krugman test: Did it have the GDP impact of the Fax machine? But country-level financial metrics aren’t exactly meant to track cultural revolutions.

I’ll leave it to my esteemed MoP co-editor Erica to talk about the impacts on the tech industry, which are immense. When it comes to media, though, I'm inclined to judge the impact of AI based on the invention of new content expressions.

• Are media orgs micro-targeting in ways that put human curators that much closer to core audiences?
• Has news reporting become micro-distributed based on knowledge, interest and personal context?
• Is there an evolution of prime social expression away from short multi-shot video?
• Do AI tools develop a cultural language, much in the way that many TikTok videos are funny simply because they subvert something else seen in the previous micro-generation of TikToks? (E.g. Is it a memetic form of content or simply a dull [and potentially dim] imposter? You know like Reel... nevermind)

History teaches us that capturing the hearts and minds of the creative and intellectual classes in a way that spreads memetically through a culture requires just this: A new means of expression that is more than just “new,” but can speak to a particular moment in time more directly, and more sincerely, than what came before.

That is precisely where we are. And I can’t wait to see what you’re building.

Please feel free to reach out, especially with AI news you'd like to see my takes on. Our operators are on the line.

questions@machinesonpaper.com