It is popular among journalists these days to warn that AI might have catastrophic effects on humanity. These concerns are overblown with regards to humanity as a whole. But they are actually quite prescient with regards to journalists themselves.
To understand why, let’s take a closer look at the sub-disciplines that we collectively call AI. AI is the widest umbrella term, but we can generally break it down into rule-based systems and machine-learning systems. Machine-learning systems can be broken down by their application (video, images, natural language, etc). Among these, we’ve seen the greatest recent strides made in natural language processing. Specifically, we’ve seen the invention of the transformer model in 2017, followed by rapid growth in the size of transformers. Once the model exceeds 7 billion parameters, it is generally referred to as a large language model (LLM).
The core “skill” (if you might call it that) of an LLM is its ability to predict the most likely next word in an incomplete block of text. We can use this predictive mechanism to generate large blocks of text from scratch, by asking the LLM to predict one word at a time.
If you train the LLM on large datasets with variable quality, this predictive mechanism will often produce bad writing. This is the case with ChatGPT today. This is why, whenever I broach the topic with journalists, I encounter skepticism – journalists see how badly ChatGPT writes, and they assume AI poses no threat to them because it’s inept.
But ChatGPT is not the only LLM out there. If an LLM is trained on a carefully-selected dataset of text written by the best journalists – and no one else – then it will develop the ability to write like the best journalists.
Unlike journalists, however, this LLM will require no salary.
Writing vs. Knowing What to Write
Before we proceed, we need to distinguish between the mechanics of writing and the creativity required to know what is worth writing about. AI can’t interview whistleblowers or to badger a politician long-enough for the politician to accidentally tell the truth.
AI cannot gather information. But it can describe information gathered by humans in an eloquent way. This is a skill that journalists and writers used to have a monopoly over. They no longer do.
Given the current rate of progress, within a year, AI could write better than 99% of journalists and professional writers. It will do so for free, on demand, and with infinite throughput.
The Economics of Zero-Cost Writing
Anyone who has a list of facts to convey will be able to turn these facts into a well-written article. Anyone who finds an article on any subject will be able to produce another article, covering the same subject. This derivative article will be just as good as the first one, and won’t plagiarize it or violate its copyrights..
The marginal cost of written content will become zero.
Currently, the economics of written media are based on human labor. Well-written content is scarce, so it has value. Entire industries were built to capture this value.
When AI can produce high-quality content for free, the financial foundation of these industries will collapse.
The Abolition of Publications
Consider traditional publications. For decades, companies like The New York Times have employed skilled writers to produce a limited number of articles each day (typically around 300). This model is inherently constrained by the number of writers and the costs involved.
In a world where AI can generate an unlimited number of articles at no cost, why limit production to a fixed number? Why not create personalized content for every reader, tailored to their interests and generated on demand?
In this new paradigm, the traditional model of periodic issues and fixed article counts becomes obsolete. Publications can shift to a model where content is continuously created and personalized, catering to the specific needs of individual readers. One reader might need a single article each day. Another might need 5000.
Publications whose primary product is packing 300 articles into a single daily issue will go extinct.
Search Engines Becoming Answer Engines
Search engines act as distributors, connecting users to pre-existing content. To achieve this, they perform four steps.
First, they index vast amounts of pre-written content. Second, they receive a query from the user. Third, they search the pre-written content to find items that are relevant to the user’s query. And fourth, they rank the retrieved content and present a sorted list of results to the user.
So far so good. But if content can be created on demand, for free, then why would search engines return pre-existing content to the user? They could simply generate the answer instead. The user would certainly be happier with a single answer to her query, instead of a long list of results whose quality may vary.
Now let’s consider the logical next step. If search engines no longer lead users to any content written by others, what would happen to the “content” economy?
Most content on the internet was written to be monetized. People write articles, rank on Google, receive traffic, and turn it into income (using ads, affiliate links, or direct sales of products or services).
What will happen to this ecosystem when the traffic disappears?
Social Media: The Next Domino
Social media platforms were initially designed to facilitate interaction between users. I am old enough to remember the days when people logged into Facebook to write on a friend’s wall, poke, or throw a virtual sheep at someone.
Today’s social media is different. The most common number of followers users have on Instagram is zero. The second most common number of followers is one. The vast majority of views, shares, comments and followers is amassed by a small number of professional creators. Most users post nothing and are followed by no one.
Simply put – most users visit social media to find content they might enjoy. Social media companies act as distributors, just like search engines. The main difference between Facebook and Google is that Google uses a query to select content, whereas Facebook selects content without one.
If this is the case, then the next step becomes obvious. Why would social media promote user-generated content, when they can generate AI-based content on demand? Text-only at first, perhaps, but eventually images and videos too.
And once social media no longer leads users to content made by creators, what will happen to the “creator economy”?
The Star Trek Replicator Analogy
We are entering a new paradigm where AI functions as a Star Trek replicator for content.
In Star Trek, there is no need for farmers who grow food, stores who sell food, chefs who cook food or waiters who serve food. The replicator can create any food you like, on demand, by directly transforming raw materials into the final product.
Likewise, I see no place in our future for any company who creates written content, distributes written content, mixes written content in some special way, or serves pre-existing written content to the user. The only valuable functions will be obtaining raw materials and transforming them into the final product on demand.
We still need ways to create information that did not exist before and gather information that was not publicly available before. Everything else will be achieved by AI engines that convert the available information into personalized content.
Implications for Content Creators and Distributors
Traders often talk about “positive exposure” and “negative exposure”. The easiest way to understand these concepts is to ask yourself – if this thing goes up, will I benefit or suffer?
AI is going up. And it is going up especially fast in areas like natural language and other human-generated content. The question every professional needs to ask themselves is – do I have positive or negative exposure to AI right now?
If you are a content creator – let’s say a news publication – and your cost structure is non-zero, then you are likely in trouble. You will soon be competing with content creators whose cost is zero, and that is not a competition you can win. In all likelihood, you have exactly 3 choices: exit the market; reduce your costs to zero (by becoming an AI company); or go bankrupt.
If you are on the distribution side of things, you probably have more time before the full effects reach your bottom line. Network effects will help you stave off the disruption for a few years. But eventually, things that must happen, do happen. Search engines replaced web directories. Feeds replaced a large part of the function search engines served before. And soon, on-demand content creation will replace both.
The Role of Government and Regulation
As someone who was born in the Soviet Union, I am not a big fan of government regulating speech. The moral hazards are usually higher than any temporary benefit such regulation might bring.
Nevertheless, I think that governments might have an important role to play in determining how this unfolds.
We have good and bad examples of government regulations and the effects they’ve had on industry. The “26 words that created the internet” grew a nascent industry to trillions of dollars in value. The regulation of ISPs in the 90s, however, brought down the number of ISPs in the US from over 3000 to 6, and resulted in a situation where US consumers have the worst bandwidth access in the developed world.
When asked for my recommendations, I usually point out three ways in which government regulation can help, rather than hinder, the development of this new ecosystem:
1. Mandate interoperability, and make it easier for consumers to switch providers.
Capitalism works like natural selection – companies that do things better or more efficiently will grow faster than companies who don’t. “Lock in” mechanisms that make it harder to switch, like the inability to export one’s data out of a service and port it to a competitor, slow down this evolution and result in lower growth.
If governments can mandate interoperability throughout the tech industry, we will see more good features and good behaviors rewarded. We will create incentive for companies to innovate in things people want, rather than innovating in ways to squeeze more out of a captive audience.
2. Enforce antitrust by focusing on monopoly abuses, rather than monopoly risks.
We all know that when two companies merge, the resulting entity might become large and have outsized power relative to its customers. But the existence of outsized power does not always lead to bad service or predatory pricing.
Meanwhile, companies who already have outsized power are often engaging in anti-competitive behaviors right before our eyes. And yet the FTC focuses on blocking mergers and acquisitions.
If governments focus on banning and strict enforcement of anti-competitive practices like dumping and bundling, especially with regards to tech products that are used by the majority of the population, the entire system will become unclogged.
Some specific examples might help illustrate this point.
Providing a browser, which is a very complex piece of software that costs billions to develop, for free – is a clear case of dumping. New browser companies like Cliq or Brave find it hard to innovate in this space because their much larger competitors give this expensive product away for free. The result is that all browsers look the same these days, and there’s been no significant innovation in this space since 2016.
Providing a corporate messaging app as a part of a document editing suite that every business must buy – is a clear case of bundling. Even a very successful startup like Slack was essentially forced to sell itself to a larger company, just to be able to compete as a paid product in a space where their main competitor is bundled with something their customer must have anyway.
As AI develops into a new ecosystem that becomes larger than the internet, we’re bound to see even greater abuses in this nascent space – unless governments step in and ensure that dumping and bundling do not pay.
3. Consider ways to subsidize or protect original content creation.
Government funds basic research and science through grants and other subsidies. It also protects new ideas that people discover in their research through patents. The reason these two mechanisms are necessary is that copying an idea that works is much cheaper than coming up with a new idea that works. Without intervention, this might lead to a tragedy of the commons where everyone copies from their neighbor and no one creates anything new.
In journalism, and content creation in general, these mechanisms were unnecessary because copying without violating copyrights was a difficult process. But with the advent of AI, this is no longer true. As the price of paraphrasing others’ writing approaches zero, we will need mechanisms to incentivize something other than paraphrasing – and the best answers might look a lot like the ones we have in basic research today.
Making the Best of this Challenge
The transformation brought about by AI is one of the greatest challenges facing humanity today. Journalists and other content creators will be affected first. Distributors of content will follow soon thereafter. We will eventually enter a completely new paradigm, which I referred to as the “Star Trek Replicator” model for content creation and distribution.
We have an opportunity here to build something much better than what exists today. Just as the invention of the printing press led to the Enlightenment, the invention of AI could lead to a second Enlightenment. But unfortunately, not all the possible futures are benign.
It’s up to us to nudge this evolution in the right direction.
Credit: Source link
