It’s not journalism’s job to make AI more accurate

Why public policy will be necessary to support "a learning society" under AI.

Note: Since it no longer goes without saying, the newsletter is 100% human-written, because the author believes that writing is an act of learning by doing.

As reality sets in, and the most egregious propaganda of recent years gets exposed by increments to have been so much bullshit, it’s clear that artificial intelligence as of 2026 is just a normal technology.

AI causes disruptions much like any other innovation. Its impacts, while real, are not fantastic or mysterious. The jobpocalypse — mass human unemployment caused by AI — is obviously not coming soon, nor is robot superintelligence. Like all other machines fashioned by human genius, AI is not conscious. The familiar drives of human nature and capitalism have not been fundamentally changed by AI’s arrival. People, companies and societies will use AI for logical and illogical purposes; these uses will have benefits and harms.

This is because AI is a normal technology.

It’s important to sweep away the conceptual clutter around artificial intelligence so we can talk more plainly about the appropriate role of AI in our communities and societies.

AI does not have a destiny. AI is not even inevitable. It is software. AI is simply here, much like trains, cars and computers are here. The effects of trains, cars and computers are good and bad, both for the people who use them and those who don’t. So too with AI: because AI is a normal technology.

But the unique hype around AI gives us a chance to talk about the kind of society we would want to live in, and what more might be necessary to create it.

AI’s complicated role in a “learning society”

This decluttering of AI rhetoric — the normalization of AI debate — is especially critical for those of us for whom journalism, technological innovation, libraries, universities, an open web, etc., are means to a particular end: creating a more enlightened public.

The Nobel-winning economist Joseph E. Stiglitz and Bruce C. Greenwald called this the work of “creating a learning society” and wrote a 2015 book of economic theory of the same name.

Creating a “learning society” is a fantastic goal. When people learn more stuff, they become more engaged citizens, better workers, cleverer innovators, more insightful artists. In a learning society, everybody benefits. Standards of living get higher, we cure more diseases, we get more magnificent art.

But a learning society doesn’t happen by accident. You need public policies (like taxes and subsidies), institutions (like schools and news organizations) and norms (like openness and intellectual integrity) that proactively support lifelong public learning.

The intent to create a learning society is very important, because we do not live in the most knowledgeable of all possible worlds.

Markets for information are inherently “inefficient,” meaning we usually don’t produce or spread as much information as might be ideal for society. There is a fundamental conflict between how easily information spreads and the ability for information providers to recoup the costs of producing or obtaining that information.¹

This is called the Grossman-Stiglitz Paradox, and it’s why information producers so frequently lock up, patent or paywall socially valuable information, and why too few people pay for information they can otherwise get freely.

Solving this paradox is a real problem, because the most important economic function of a learning society is creating more knowledge “spillovers” where good information travels far and wide in a society.

Think about impactful investigations from news organizations, new management best practices from field-leading entrepreneurs, and new technologies. Knowledge spillovers ripple through other companies, fields and communities and help eliminate waste and corruption and spur growth.

AI — much like libraries, universities and Google Search before it — aspires to be an ultimate “spillover” technology, by collecting all that has been thought and said and delivering the most useful knowledge more efficiently than before.

But AI in 2026 presents some serious problems for this work of creating a learning society:

• There is a persistent misalignment between AI’s actual capabilities and users’ beliefs in its perceived capabilities.

• AI is causing cognitive offloading in schools and workplaces and increasing short-run productivity at the cost of longer-term learning, which is the true source of long-term economic growth.

• AI as an information source is harvesting not just the content but the audiences that other knowledge producers rely upon to fund or inspire further knowledge creation (which AI itself relies upon).

• AI as a source of content pollution has nullified presumptions of integrity and harmed the openness of academic, creative and intellectual communities: Teachers increasingly fear their students to be cheaters, freelancers are feared to be AI fakes, and slush piles are feared to be slop.

• To the extent AI causes unemployment in certain jobs or careers — one of the most socially and economically destructive things that can happen to a worker — AI hurts longer-term economic dynamism by eliminating the ability of workers to learn by doing.

In short, the evidence suggests AI is drawing from the pool of public knowledge without doing enough restock the pond.

The need for public policy to support a learning society under AI

The benefit of thinking of AI as a normal technology is that we’re not obligated to think of its disruptions as necessary or inevitable. AI is another means, not an end unto itself.

The estimable Gina Chua recently said of AI agents that “our energy should be spent on making them more accurate, not just complaining that they aren’t.”²

I disagree, and I’ll talk about this problem from the perspective of the news industry (since it’s what I know best) to illustrate why everybody’s incentives are out of whack from what might be ideal for a learning society. Public policy is going to have to intervene.

For example: A marginal analysis might strongly suggest that news providers’ limited time and energy is more productively spent not making AI more accurate.

The world in which AI is really good at providing current-events information is a world in which AI becomes a better destination than the news providers who collected that information — creating what economists call a “substitute good” to replace the original. This is a bona fide social problem when it’s news providers, and not AI companies, funding journalists to keep stocking the pool of public knowledge.

Even in a scenario where AI companies were required to pay fair licensing or similar fees to news companies for scraping, this is why there can be bona fide competition reason for some news outlets not to participate. This is not even factoring in the unknown math of whether the unknown revenues of AI licensing nets out to a profit once you factor in the unknown transaction costs of setting yourself up for those kinds of deals. These transaction costs are especially daunting with regards to smaller news outlets.

Well, from the macro perspective of the welfare of a learning society, who cares about the existence of news companies? If AI companies voluntarily funded their own armies of journalists to feed AI agents with accurate original reporting, then maybe society might be better off from an information-maximizing perspective. (Though the history of Silicon Valley suggests this kind of strategy is highly unlikely.)

But a world without publishers and journalists has problems even beyond information production. News providers don’t just produce knowledge but also community, serving as collectors of what Alexis de Tocqueville called the “wandering spirits” of democracy.³ AI, by comparison, is practically inherently anti-community technology.⁴ Much how freedom and equality are the binding agents for liberal democracy, so too do knowledge and community work together to create the learning society.

Furthermore, 20 years of Facebook and Twitter strongly suggests that the arc of tech development does not bend toward accuracy — if anything, tech’s history strongly suggests a teleology away from accuracy and toward slop for the sake of pursuing maximum market share.

The history of social media and the journalism industry, if anything, is a graveyard of sunk costs. The tolerance for declining returns is shockingly alive and well, as evidenced by the number of journalists still fucking around on Elon Musk’s X, nursing legacy follower counts at the expense of anything else they could be doing more productively with their preciously limited time and energy.

One of the benefits of Joe Stiglitz’s work and of behavioral economics in general is that we can say plainly that consumers are not utility-maximizing robots who gravitate toward the most accurate or useful information. Sometimes people just get stuck because of status-quo bias or annoying “switching costs,” even when they don’t even like the services they’re using.

Polling on the increasingly negative public opinions about AI suggest there might be a lot of this sort of sentiment already running around, which suggests that standing apart from AI might be a viable strategy for news publishers.

Last week I was talking to high school students at a journalism camp at Mizzou, and interestingly, they all resented how much time they had to spend on social media and knew that what they saw on the services was not inherently trustworthy.

But one of the students said she’d become an avid reader of the New York Times due to a subscription program run through her high school: essentially a de facto public subsidy to cancel out the paywall that NYT erected to protect and sustain its own journalism. While modest, this is an example of a valid intervention to correct the Grossman-Stiglitz Paradox and resolve the inherent conflict between the funding and sharing of accurate information.

A future where AI is an increasingly accurate source of information, just like a future where there are still news organizations providing news and creating communities for democracy’s “wandering spirits,” are not necessarily a given. It’s not even clear if we can have both those futures at the same time. But if we do, public policy will probably have played a part. The learning society is something we have to want to build.

Note: These are my own views and not necessarily those of my employer, my loved ones or anybody else, really.

1. Also known as the Grossman-Stiglitz Paradox, which I wrote about here:

The simple economic paradox wrecking the news

2. What's It All About?

3. https://wisc.pb.unizin.org/ps160/chapter/7-3-tocqueville-on-democracy/

4. When "To be or not to be?" is a prompt and not the question