AI-Based Disinformation Will Not Be An Especially Big Deal in the 2024 Election

Plus! Diff Jobs; Discount Pods; China and Real Estate; Microcaps; Unraveling; Market Structure

In this issue:

AI-Based Disinformation Will Not Be An Especially Big Deal in the 2024 Election

A persistent concern about AI is that it enables the creation of hyper-realistic fake audio and video, which will usher in a dark age in which no one can trust anything they see online, and that elections will ultimately be determined by whether Russia or China get the most GPUs—or something like that. It's unclear, because the implication of most disinformation-discourse is that it's a warning that other people will fall for it, in harmful ways. Nobody says "My biggest fear is that I, personally, will switch the candidate I support on the basis of a fake thirty-second clip posted on Twitter by Belorussian intelligence." It's always a vague model of somebody else's bad decisions.

The main protection most people have against falling for AI-generated propaganda is choosing to tap the NYT app icon instead of the X one. The Times is not perfect, but they presumably have rules against writing stories about random social media rumors without at least going through the motions of vetting these for truth. Meanwhile, apps themselves are reducing the salience of political content generally, which would cover the fake kind as well as the new kind; Threads will not recommend political content to people who aren't following the relevant accounts, so most content will be opt-out while the high-risk kind will be opt-in. The FCC has also stepped in to ban AI-generated robocalls.[1]

In fact, one of the ways that you can't readily trust mainstream media: there's bias in which stories count as news, and that bias exists both in the heads of people who write the news and in the social networks through which stories propagate.[2] Some stories get written because something specific happened and somebody needs to sum it up—the State of the Union, the outcome of the Super Bowl, a big merger, a war. And some stories require more specific agency: a PR person makes a pitch, or a source shares something with a writer.

In that last category, there's no doubt that some stories happen entirely because a journalist gets an idea, digs into it, and produces a coherent narrative. Either there isn't a catalyst, or there's a subtle one that's easy to miss: a classic instance of this would be noticing that someone has been quietly dropped from a scheduled presentation, or noticing an explanatory footnote in some document that leaves more questions than it answers.[3]

The real genesis for those stories is less seemly: when two people really love what they can do for each other, and one has an audience and another has information, they can be combined to gestate into an exclusive story. This is the usual genesis for articles about pending legislation or executive orders, about mergers, and product launches.

Which raises the question: what is each side doing for the other? When someone tells a financial journalist that company XYZ is probably bidding for company ABC, why do they do it? It's a trial balloon (Wikipedia has some nice examples): share the story as an off-the-record musing, see what the reaction is, and decide based on that reaction whether it was a definite plan or a hypothetical musing. This effect, where media coverage collapses uncertainty and makes events either definitely happen or definitely not, is incredibly powerful; one SPAC merger in 2021 happened because there was an incorrect rumor about it, and the stock price reaction convinced management to do the deal.

But disclosure has degrees of freedom. Suppose you're involved in the legislative process, and you hear about a bill to which you're opposed. This bill has pluses and minuses, but you know, realistically, that it can probably get passed. What can you do? You can make a list of those minuses, and share them with someone who will highlight them. For example, if for obscure DC politics-related reasons, you are opposed to a proposal for incentivizing people to precisely quantify the risk of terrorist attacks, you could probably emphasize that doing this involves betting on tragedies, which strikes many people as grotesque.[4] Is there a plan that changes how some people are paid? The plan's success is highly sensitive to whether the first leak talks about who gets more or who gets less.[5]

In fact, we can even see this in AI-disinformation discourse! This Economist guest piece opens by talking about a deepfaked video of Olaf Scholz banning the Alternative für Deutschland party ($). It's not completely out of the question, but extreme enough that someone might think twice before floating it. But now it's been floated; the video came out two weeks ago, but more recent coverage ($, FT) suggests that such a ban is a live possibility.

So, in this case a deepfake arguably had an effect: people who thought the video was real responded in a way that made it more likely to happen. So AI is influencing politics, but the specific way it is is that it democratizes the trial balloon. If there's an idea that's in the range of things that could be proposed, but also in the range of proposals that would be opposed, a trial balloon actually creates information by determining where those distributions overlap and where they don't. This "what if" game used to be restricted to the well-connected among either leakers or writers, but now anyone who can make a video plausible enough to be taken seriously but interesting enough to go viral can have the same sort of influence.

So if there is a deepfake effect, it's one that makes the market more efficient. It's easier than ever to get a large group of people to temporarily believe a politician said something they didn't, and for a politician to see whether they'd like it or not. And since a politician is someone who has opted into a job that's pure popularity contest, or at least one where the more serious judgment doesn't happen until after they're dead, it will be pretty easy for such people to end up endorsing the surprisingly popular views that are ascribed to them, and rejecting the unpopular ones. The audience that's swayed by deepfakes is some combination of low-information voters and the hyper-partisan (often both!); they're more likely to reinforce existing beliefs than to change them. Even though the deepfakes in question are fake, they're actually contributing to a more efficient process of converting vague notions into concrete policy proposals. And even though increased market efficiency is uncomfortable to the people who prosper from the inefficiencies, in the long run it's a good thing to have and impossible to stop.

Disclosure: Long META.

  1. They say that this is to reduce scams, but it's already illegal to scam people; the interesting question is a) whether this will be easier to enforce, and b) what otherwise-legal activities would be banned. ↩︎

  2. To be clear, I'm using "bias" in the morally-neutral sense that people all have different priorities and interests, and that institutions naturally select for some personalities over others. Media is an industry where the pay is low compared to the credentials required, so it's going to select more for people who care about status over money. And caring about status means being sensitive to the opinions of status-conscious people. All of us have this kind of bias; I'm much less likely to know about or write about sports than the average person, so one of The Diff's biases is that it underrates both the importance of sports and sports as an intuition pump for understanding other parts of human behavior. Acknowledging biases doesn't mean reversing them; the time investment to get something of value out of football is more than I'm willing to make, so this bias will be with us for a while yet. ↩︎

  3. The usual way to engineer this is to do the reading. This is easier in business/finance reporting than anywhere else, because so much of the material you'd use to tell the story already comes in the form of either publicly-available legal filings or fairly widely-available transcripts. Don't want to be stuck continuously rewriting Apple press releases until you get replaced by an AI? Start reading earnings transcripts from Apple's suppliers, and their suppliers' suppliers, until you start to find the stories that Apple doesn't want to put in a press release. ↩︎

  4. It's worth meditating, however briefly, on how much better your life is because there are people who profitably bet on horrible outcomes. When civilization fails to collapse after a natural disaster, it's because someone prepared for it and then put their plan into action when it happened. That person had a motivation; perhaps they're at a nonprofit and want to feel good about their contributions to the world, or perhaps they're a supply chain manager who knows that they'll still get fired if their company isn't back in operation by the time their competitors are. Financial markets are a less direct way that people who get paid to bet on disaster make your life better, but this still holds in an indirect way: when someone responds to geopolitical risk by buying futures in some at-risk commodity, they're signaling to the consumers of that commodity that it's a good idea to look for substitutes, build up a stockpile, or both. ↩︎

  5. I don't think anyone who has enough information to know this for sure can be trusted, but: this does present a fun Leaks Hypothesis of Polarization. Every bill is a negotiation, and the more complicated ones will involve more give-and-take from both sides. But that increases leak risk on two dimensions: first, a longer negotiation means there's more time for someone to backchannel parts of it to the media. And second, a more complex set of tradeoffs gives a leaker more negatives to highlight. So the temptation to leak gets stronger, and when it happens, both sides trust each other less. To really test this, though, you'd need some way to track how official and systematized leaks are, and that's hard to do. A hundred years ago, an off-the-record comment from an official familiar with the President's thinking could have come directly from the President himself, at a secret but regularly-scheduled press conference. ↩︎

Diff Jobs

Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:

Even if you don't see an exact match for your skills and interests right now, we're happy to talk early so we can let you know if a good opportunity comes up.

If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority.


Discount Pods

Pod shops have been a frequent topic of discussion in The Diff (you can start with the pod primer here). Pods are well-known for producing decent-to-excellent risk-adjusted returns (depending on which pods you ask, and on whether or not you ask someone who knows that they are not targeting the same volatility as the S&P 500). But they're even better-known for creating large paydays for managers. That combination creates a natural drive for a stripped-down model that keeps the analysts but automates portfolio construction ($, Business Insider).

Getting rid of the best-paid employees means a cheaper cost structure, but also raises the obvious question of why other firms haven't thought of this. It's easier to add analyst headcount than portfolio manager headcount, so if analysts were higher ROI you'd expect the big funds to have fewer teams with more headcount over time, instead of adding new teams. But it's also possible that some of that portfolio manager compensation is a premium for depending on a skillset that's more automatable. Converting a set of fundamental theses into a portfolio is partly a problem of math and partly a problem of taste, but the usual bet to make is that when mass production gets cheap enough and good enough, there's little room for craft.

China and Real Estate

One way financial bubbles can come about happens when politically-favored groups get better treatment: savings and loans were a sympathetic category in the early 80s, when they were bailed out of bad interest rate bets by getting permission to make bad credit bets instead. In the early 2000s, investment banks lobbied for lighter regulations partly on the grounds that there had not been a massive financial crisis any time recently. (Whoops!) In China today, one important feature of their regulatory constraints is that 80% of household wealth is in the form of housing, compared to 30% in the US ($, Economist). China's real estate privatization was, according to China's Economy, one of the largest wealth transfers in history. Not only does real estate appreciate nonlinearly as a country develops, but the initial privatization took place at artificially low prices as a handout. This skewed balance sheet means that if economic growth slows, the government's next-best option is to transfer wealth away from other sectors and towards housing.


The penny stock boom, highlighted in this Diff post about the surprisingly strong correlation between a low share price at IPO and poor subsequent performance, continues apace. The most-traded stock in the US, by share count, is a money-losing mobile app company whose $132m market cap values it at ~9x rapidly-declining revenues, and Bit Brother, a former tea company that pivoted into crypto but that is not Long Blockchain ($, FT). Retail speculation in penny stocks is not an important macroeconomic phenomenon; a good-sized secondary offering from an S&P 500 company probably has more economic impact than all penny stock trading in the aggregate. But it matters a lot to people who get sucked into this and end up losing all their money. It's hard to ignore the fact that this business is booming at the same time that literal gambling is more legal and, for the moment, more socially-acceptable. Usually penny stocks would compete with, not complement, sports betting, so it's a sign of a secular increase in demand or tolerance for both that they're rising at the same time.


The WSJ has a look at how investment bank B. Riley ran into increasing trouble financing a buyout organized by a hedge fund manager who has since been accused of fraud ($). There's a long history of banks growing through an entrepreneurial, eat-what-you-kill model where the bank acts as a platform for dealmakers who source and execute transactions mostly independently. This decentralized approach works well, but eventually means that the bank is leasing its reputation too cheaply. The growth is great for a while, but eventually problems accumulate. It's a good reminder that adverse selection is everywhere: any deal you get is a deal someone else didn't do, and they might have had a reason.

(The Diff covered B. Riley, and the risks of its increasing reliance on trading income rather than fees, last year ($).)

Market Structure

Italy's economy is good at producing new companies and bad at scaling existing ones ($, Economist), resulting in economic underperformance. It's useful to think about this in factor terms: one way to describe Italy's underperformance is that its economy is overweight small value and underweight large growth, so in an environment in which large growth companies have done well, Italy does relatively worse. Factors usually describe price changes, but they can be sensitive to macroeconomic variables. In other words, they partly describe an economic environment by using asset price changes as a convenient yardstick. As in markets, over-indexing to something can happen when certain assets are selected out of a portfolio; the linked article mentions that many of Italy's highest-profile brands are owned by foreign companies, and that process continues; over the weekend fashion brand Tod's agreed to sell to a PE firm backed by LVMH ($, FT).