The Value of Content in an AI World
Navigating the new content landscape
What separates good thinkers from great thinkers is:
1) The number of mental models at their disposal;
2) The accuracy of those models; and
3) How quickly they update them when they're wrong.
― Shane Parrish
People are using generative AI to do incredible things such as greatly reduce their time to create content, or even be a data analyst. With these productivity gains, people are going to be able to create better content at higher volumes. What are the implications of that? Let's try and figure out what the value of content is in an AI driven world by playing with some mental models.
For the purposes of this discussion, let’s refer to content as any media that someone could consume whether intentionally or not (books, tv, Youtube videos, ads, blog posts, Substack articles, etc). We'll start from one basic premise - generative AI will create a step function that causes more content to be created at a greater rate than before it existed. Through this discussion, we'll go through first order, second order, and third order effects.
First Order Effects
We’ll start with the basics, supply and demand. First let's look at the supply side. Prior to AI, let's assume that the average rate of output per person was relatively fixed. While these are nascent studies, some groups are showing that worker productivity can increase as much as 59% from using ChatGPT. We can debate what the actual percentage is, but the takeaway is that AI assistants make people more productive. This means we can expect the rate of content creation to increase, making more content available. This would be on top of the current rate of data creation (see Figure 1).
Figure 1. Growth of data over time. Courtesy of firstsiteguide.com.
Next, let’s consider demand as the individuals that want to consume content, along with the portion of their day that they do so. For demand, around ~64% of the global population is online. While India is just starting to come online, their ~1B additional internet enabled individuals (along with other countries) will bring this number into the 75-80% range in the next 5 years. However, the internet population growth pales in comparison to the growth of data worldwide (see Figure 1).
Combining these two points, that demand is relatively fixed and supply is going to increase dramatically, implies that the average value of a piece of content is decreasing. That's a simple equation of average value = total demand / total content created where the total content created is increasing at a much higher rate relative to the total demand. However, this is a bit of a naive high-level view since not all content is created equal. Nonetheless, the takeaway about the directionality for average content value should generally hold. It should look something like shifting from a normal distribution to a power law distribution.
This raises an interesting observation. If the average value of content is decreasing but the area under the value curve stays the same, where does the demand shift to? It makes a select few pieces of content extremely valuable. In a world full of noise, things with signal become hypervaluable.
Second Order Effects
Messages that are broadly valuable are those that many people are receptive to (see Figure 2). Tim Urban from Wait Buy Why breaks this down in his discussion on how different messages are felt within a nation (note: I had to use an archived link because Tim Urban has apparently turned the post series into a book and removed the original posts). We can repurpose this mental model to understand the value of content.
Figure 2. The number of people reached by a message based on its receptiveness to the audience. Courtesy of Wait But Why?
The issue with creating a broad message in a world of increasingly less valuable content is that only a select few messages will be of sufficient quality to make it through and actually be valuable. People want their content to succeed. The natural next step is content personalization. Individuals find content that has been personalized to their needs much more valuable than content that is meant for a broad base. Just think of when a friend recommends something to you or how you break down a topic differently when talking to your parents, your boss, or your peer.
There are two problems with personalized content. The first is that it can be time consuming to create, especially in large quantities. The second is that it becomes much harder to find your audience for a piece of personalized content. Generative AI solves the first problem by being able to take a base piece of content and then personalize for a given audience or individual. The second problem is much harder to solve. You have to overcome both platform filters, in the form of algorithms, and mental filters by those who your content does reach.
If you want a message to get through, it needs to be personalized correctly for the right audience. Personalized content generally targets a much smaller audience. Meaning you need to generate way more content than before to reach the same number of individuals (many smaller audiences vs one large audience). This is a bad feedback loop as the act of generating much more personalized content will cause others to do the same, with the process repeating. Content creators will create a piece of content and then run it through AI personalizers that could potentially spawn thousands of individualized pieces of content, each slightly different but with the same underlying message. As people keep creating more content to try and reach the same individuals, competition becomes fiercer and the amount of content available skyrockets.
What content creators are going to find through this process is what types of personalization works, what algorithms work well, and how different platforms respond. It becomes an act of evolution. Ever seen how to make bacteria immune to antibiotics (see the video below)? It's a bit like that. Content creators are going to be creating a lot of decreasingly valued content until a few break through the walls into higher value and then the process starts all over as personalization algorithms and platforms improve.
The natural evolution is going to occur in AI generated content based on what people respond to. This gets around platform filters/algorithms/recommendation systems and peoples brains. The rate of how language changes is going to increase dramatically as content personalizers understand what people respond to.
Third Order Effects
We’ve established that the average value of content will decrease and that content will become more personalized. What does this mean? It means that for a given individual, there is likely a lot of information that would be valuable to them but that it will be difficult to find. This is either because they are unable to process the overwhelming amount of content coming at them or because it is difficult to find, as other people do not find it as valuable.
With so much content coming at you, can you personally determine what's valuable before spending your time on it? I don't think one can ever be perfectly efficient at this. Historically, large media companies played the role in determining what content was valuable to people. Then social media fragmented this to a more personalized level. Now distributed media (such as Substack and personalized advertising) is taking this a step further. What changes now is that once people figure out the right messaging, generative AI can scale the volume of content for a resonating message very quickly, across mediums, and across channels. So, in a world where you are being bombarded by content you perceive to be highly valuable, how do you make choices?
This is where AI agents and editors come in. An AI agent is an automated program that performs actions on your behalf. An editor is a person or program that approves or distills the content you do see. Both of these entities can work to find content that is important and valuable to you as a consumer.
Then we have agent to agent communication. People can use agents to expand simple thoughts or compress expanded thoughts back into the main points. Imagine a content creator called ‘Dog Lover’ who thinks of a new article about ways to spoil your dog. His agent could then take this core idea, ‘spoil you dog’, and develop a few of blog posts, tweets, instagram and then personalize each of those posts to hundreds of different audiences. Very easily, the agent has developed a large amount of content from one core idea.
A consumer’s agent could go scour the internet with the mission of looking for good ways to take care of pets. The agent might come across a personalized agent generated blog post written by ‘Dog Lover’ which speaks in a way the consumer loves. The consumer’s agent knows its owner prefers bullet point summaries and therefore condenses the key points from Dog Lover’s article. Ultimately, a lot of computational work was done for a weird type of match making. It comically looks something like this:
Figure 3. Computational waste from expanding key points and then summarizing them back down.
This implies that people will start increasing their reliance on services to help find, filter, and provide information to them. In fact, we already kind of do that with Google and social media. Except now, we can create programs that can find the content that is valuable to us as individuals instead of what a platform’s algorithm determines. There are still many questions I don’t readily have answers to, and I would welcome your thoughts on them. What happens to recommendation algorithms in a world of hyperpersonalized content? What about for content that has broad messaging that appeals to many but then has been broken up into hyperpersonalized chunks? Do recommendation algorithms look at the “family” of content? There's a lot of disruption that is going to happen to algorithmic editors on platforms because the search space of content may become too great for these algorithms to handle. What will this look like and what happens then?
Implications
As we’ve explore these mental models, we’ve covered a lot on what happens to content in a world driven by AI. Let’s summarize the implications on what will happen to content as generative AI becomes more widespread.
Decreased average value of content: With the increase in content supply, the average value of a piece of content is likely to decrease, making it harder for creators to stand out and monetize their work.
Increased need for content personalization: To maintain relevance and value, content creators will need to focus more on personalization, targeting their content to specific audiences or individuals.
Intensified competition: The explosion of content will lead to fiercer competition among creators, as they struggle to find their target audience and make their content stand out.
AI-enhanced content discovery: The rise of AI-generated content may lead to the development of new AI-driven content discovery tools that can identify and recommend high-quality, personalized content to users, helping them navigate the overwhelming volume of available information.
Shift in power from platform algorithms to AI agents: As hyperpersonalized content becomes more common, platform recommendation algorithms may struggle to handle the vast search space, leading to a shift in power towards AI agents that can find and deliver valuable content to individuals.
New business opportunities: As content creation, personalization, and distribution become more AI-driven, new business opportunities will arise for companies that can develop and provide these AI-driven services effectively.
AI-driven content regulation: With the increasing amount of AI-generated content, there may be a need for new regulatory frameworks and standards to ensure quality, authenticity, and ethical content creation. This could involve the development of AI-driven content certification systems or the establishment of industry-specific guidelines.