The End of Attention Machine
The End of the Attention/Narrative Machine
In my years in the media industry, I’ve witnessed a few seismic shifts, moving from print to digital and from centralized publishers to individual influencers.
The death of “traditional” media isn’t just another industry disruption - it’s a fundamental shift in how humans consume and process information. I’ve watched this transformation from the inside: from managing a 120-person print editorial team to leading a lean three-person digital outfit at Taiwan’s largest new media company just three years later. In 2017, I witnessed perhaps the most visceral example of this change when I helped Condé Nast China restructure, resulting in a 75%+ reduction in staff across print and digital, even though they had record-high revenue growth.
But what we’re seeing now goes beyond the familiar print-to-digital narrative. We’re entering an era where the very nature of storytelling is being reimagined by AI.
In the post-print era, the traditional media companies built what I call “attention/narrative machines” - systems designed to maximize engagement at any cost. These machines work by exploiting our cognitive biases, serving us increasingly extreme content to keep us clicking and scrolling. The result? A media landscape that prioritizes sensation over substance, outrage over understanding.
This model was powerful but fundamentally flawed. Like many first-generation technologies, it optimized for what was easily measurable (clicks, views, time on page) rather than what actually mattered (understanding, insight, truth). The attention/narrative machine gave us what we wanted, not what we needed.
The Rise of the Proactive Information Machine?
AI presents an opportunity to build something fundamentally different: proactive information machines.
These systems don’t just aggregate or amplify content - they reshape how information is constructed and understood.
What makes the proactive information machines different? Three key principles:
- Synthesize Sources: Instead of amplifying a single perspective to maximize engagement, the proactive information machines can synthesize multiple viewpoints into coherent, nuanced stories. They’re designed not to reinforce what we already believe, but to help us understand what we don’t.
- Context Over Clicks: Traditional media optimizes for emotional response. The proactive information machines optimize for understanding. They can automatically incorporate relevant background, data, and alternative perspectives, giving readers the context they need to form informed opinions.
- Interactive Depth: Rather than serving pre-packaged narratives, these systems enable readers to actively explore different angles of a story. The goal isn’t to tell people what to think, but to give them the tools to think better.
The goal for AI companies shouldn’t be to retrofit AI into existing old media structures, but to fundamentally reimagine how information flows through society. Companies like Perplexity seem to understand this, using current conflicts over data access not as battles to be won, but as experiments to understand the true value and velocity of information loops in their products.
In my opinion (IMO), the current attempts at OpenAI-publisher collaboration are doomed to fail for several reasons:
- Data: Publishers need usage data to understand their audience; AI companies need training data to improve their models. Neither wants to share 100% of their secrets.
- Quality Control: Maintaining editorial standards while scaling AI-driven content creation requires new frameworks for accuracy and accountability.
- Economic Alignment: Traditional revenue-sharing models don’t capture the complex value exchange between publishers and AI platforms.
A New Data Loop?
Companies like Perplexity and OpenAI should focus on developing a world where people are empowered to make their own choices based on the information they receive. It’s just a matter of time before Google matches their feature-set, just like OpenAI has done with ChatGPT Search. OpenAI has the model advantage while Google has the search data advantage. Perplexity has neither.
The winners in this new landscape won’t be those who build the best information machines. They’ll be the ones who fundamentally reshape how humans interact with information itself - to create new kinds of data loops involving user input.
This is where the new proactive information machines will truly make a difference, not just in how we consume content but in how we interact, create and understand it.
A voice HAI demo by Ben South:
What if you could interrupt a podcast episode to ask a question? Or tell the host to get to the point? Or even bring in another guest? An experiment with @C_H_Wood