Hi friends, it’s been a while! I’ve been spending a bunch of time learning how to make animated maps (this is harder than I thought) and generally improve video quality (this was more time consuming than I thought). I’m happy to be back to writing. Hope you enjoy!

In 2021, Alabama’s redistricting commission released a map that had a couple peculiarities. The peculiarities were hard to pick out with the naked eye, but the map diluted the voting power of the state’s black population by packing the 7th district and drawing lines to split the remainder of the population amongst the 1st, 2nd, and 3rd districts. The Supreme Court agreed that this map had to be redrawn in a 2023 decision:

Left: Alabama 2020 Districting Map produced by Caliper by Maptitude. Right: Districts overlaid with black voting share by Census tract via Bloomberg

The legal saga prompted me to look into the complexity of algorithmically detecting a gerrymandered map. Conceptually, the problem is systemically important to ensure votes aren’t wasted and the House is responsive to voters. But difficult because of the lack of clear standards, myriad legal requirements, and vast space of possible map configurations (a fun statistics / sampling problem). Let’s go through each in turn.

📝 Standards (, Lack of)

One thing you may have missed from the image above is the tiny text at the bottom “© 2021 Caliper”. Let’s talk about this for a second. Caliper is a mapping tech company that develops a product called Maptitude. This product appears to be used to develop the blocked Alabama map (I’m sure others too). Using mapping tech on its own is pretty neutral - necessary really - but it underlines the importance of having a standard by which to judge a map’s fairness. A map can be optimized to favor one party’s voters without any visible irregularities as below:

Districts that are clearly, visibly not compact via Math for the People

“Fairness” is a vague term but most definitions I’ve seen boil down to protecting a “one person, one vote” standard - voters’ right to be represented is protected equally. Metrics to judge fairness represent this standard:

• Deviation from Partisan Symmetry (aka “Partisan Bias”): If one party won some percent of the statewide vote (say, 55%) resulting in 3 seats in a state with 5 states, we would expect the opposition party to have won 3 seats as well if they won 55% of the vote. The deviation from this symmetry is what we measure, that is how many seats above / below the vote share a party wins

• Efficiency Gap: This is a measure of “wasted” votes. Every vote above 50% + 1 for the winning candidate and under 50% for the losing candidate is considered “wasted” as the winning candidate would have won with 50% + 1. The difference between wasted votes for the controlling and opposition parties can measure if votes are packed into a district or cracked across several to dilute voting power.

The court has not tapped either of these metrics as a standard for judging the fairness of a map.

Gerrymandering cases before the court without establishing a standard

🏛 Encoding Legal Requirements

Just as a quick note, before getting into the algorithms for detecting gerrymandering, I just want to appreciate that any algorithm needs to consider state laws in addition to federal laws. While I don’t personally know what it’s like to capture these specific legal requirements into code, I can appreciate that it mustn’t be easy nor quick:

A states bin map of the different legal requirements for redistricting by state (blank = federal standards - eg VRA, Equal Protection Clause of the 14th Amendment)

🤖 Proposed Algorithms

We’ll use Alabama as an example for the following 3 algorithms. For context, Alabama has 7 representatives and 67 counties. The potential number of map configurations for AL is at least in the trillion trillions but ultimately unknown. If you conceptualize a state as a matrix (e.g. a 5 county by 5 county square), the number of configurations scales as the matrix becomes more complex - this is a combinatorial explosion. States are way more complex than a 5 × 5 grid, so these algorithms are meant to help us sample from an option space of unknown size.

• Compact Random Seed & Grow: To start, we’ll allocate a representative to each of the 67 counties. So we need to go through a merging exercise. To do that, we’ll randomly pick (or seed) 7 counties and merge adjacent counties until we have the 7 districts that we’re looking for. We pick the adjacent counties based on the distance between the geographic centers of the counties. This process doesn’t guarantee that populations will be the same across the board. To fix that, cRSG looks at adjacent districts with too much deviation and finds a county that’s furthest from the centroid of both districts and swaps it from the district with a greater population to the district with a smaller population; we repeat this swapping again and again too until we have parity. This is what has been traditionally used but has some drawbacks (e.g. the time it takes to do all the limitations and bias caused by random seeding)

Seed 7 Districts

Merge Iteratively

Merge Until You Get 7 Districts…Then Make Sure You Have Population Parity

• Markov Chain Monte Carlo (ReCom): We need to start with a previously created map for this method and explore what happens if we recombine it. So, we take adjacent precincts, we merge them into something called a spanning tree. We then just have to break one connection as long as it maintains population parity to get 2 new districts.

Our final map with 2 counties merged with a spanning tree

Breaking a connection to recombine the districts

• Sequential Monte Carlo: This time we don’t need an initial plan, we start off with 1 huge district. We create a spanning tree like in our ReCom approach, then make a split, so we have 2 districts. 1 of them will be final and the other will be split again. And we keep doing this until we get 7 districts. There’s an extra bit here that makes this approach more efficient and more complex. We won’t go into the specifics of this formula but Sequential Monte Carlo basically looks at all our constraints and weights them to create a distribution of priors (basically the distribution a plan should fall into if it follows the rules) - if it doesn’t fall into that distribution then the plan is rejected; every time we iterate on this approach, those weights update. 

Start with one huge district

Split the huge district using a spanning tree as in ReCom

Split the remainder again

These algorithms can simulate thousands of maps and help us construct a distribution of partisan bias and efficiency gap. We can then compare an enacted plan to the distributions.

Conclusions

Mathematicians across the country aren’t just trying to answer this fair maps question because of its complexity but also for its importance. There are outsized costs to getting this wrong or being slow in making a determination. Election cycles - especially for representatives - pass quickly, so politically biased maps create irreparable harm that can ultimately repeat after the next Census. 

A lot of the discourse around fair congressional maps center around independent commissions. Currently, 4 states have independent commissions that create congressional maps, 6 states have only 1 district, and the other 40 have court- or politically determined maps. And we should push for these types of commissions in the other 40 states that have more political processes. Because one necessitates the other: equations that determine the fairness of a map are nothing without independent commissions to put the insight into practice and an independent commission must have a baseline for what constitutes fairness in the first place. 

“Who Controlled Redistricting in Every State?” Brennan Center for Justice

It may seem like overkill to go this in-depth on what constitutes “fairness” and pull out algorithms to determine that. But these maps are literal building blocks of democracy. Fair maps make for more competitive races, a more truly representative House, and most importantly more power to voters.

Sources / Shoutout:

• ALARM Redist R Package: https://github.com/alarm-redist/redist

• ALARM 50-State Simulations: https://alarm-redist.org/fifty-states/

• MGGG Redistricting Lab Metagraph: https://mggg.org/metagraph/

• Quanta Magazine. "How Math Has Changed the Shape of Gerrymandering", 2023. quantamagazine.org/how-math-has-changed-the-shape-of-gerrymandering-20230601/

  - Sequential Monte Carlo research paper (Kosuke Imai, Cory McCartan): https://imai.fas.harvard.edu/research/files/SMCredist.pdf

• Bloomberg Demographic Overlaid Congressional Map: https://www.bloomberg.com/news/features/2022-10-01/what-supreme-court-case-on-alabama-voting-maps-may-mean-for-black-americans

• Princeton Gerrymandering Lab: https://gerrymander.princeton.edu/redistricting-report-card/?planId=recotR1k8LgVMbB2g

• Geometry and Compactness. https://web.stevenson.edu/mbranson/m4tp/version1/gerrymandering-math-topic-compactness.html

• Brennan Center for Justice. “Who Controlled Redistricting in Every State”. https://www.brennancenter.org/our-work/research-reports/who-controlled-redistricting-every-state

Thank you for reading this article! If you enjoyed it, consider subscribing and sharing Statecraft or the video with a friend. Until next time! - Arman

The 737-Max incidents that have haunted Boeing the last 5 years are jarring. Not least because they highlight fundamental design flaws from a company known for its engineering culture (see, the B-52 program). But to really understand the $20B worth of flaws in just over a decade of the 737-Max’s existence, we need to understand how Boeing has morphed over a century. How have antitrust policies, politics, and corporate culture morphed over time to lead us here?  

Constructing a Colossus

Boeing was founded in 1916 by William Boeing. At first, they started out by manufacturing smaller aircraft out of an old ship yard. They were pivotal in pioneering aircrafts into what we know today for mail, military, and civilian use. 

A 1920s map of Seattle with images of William Boeing and Boeing Plant No. 1 highlighted. Maps courtesy of https://www.davidrumsey.com/

The Boeing we know today was originally born out of antitrust action in 1934 that broke up aeronautical corporations. The government wanted to prohibit one company from basically vertically integrating the entire aviation industry — they’d be too powerful. So they made 3 companies: Boeing, United Airlines, and United Technologies (now part of Raytheon). One to design and manufacture aircraft, one to operate an airline, and a third to do research and development on aviation technology. So the company we know today as Boeing was born here.

Boeing’s acquisitions prior to 1934 (The US Postal Service didn’t like negotiating contracts with a vertically integrated corporation). Design by https://youtube.com/@theArmanMadani 

Boeing’s first president post-break up was Clairmont L. Egvedt, a chief engineer and aircraft designer. Boeing was known for heading the company with engineers and aircraft designers — basically people who knew about the intricacies of building an aircraft. 

Their follow ups to the B-52 program and entry into commercial aviation aren’t too shabby. The 707 program was the first commercial jet airliner delivered in 1958 and flew through 1983 in the US. The 737 program has run for 60 years, raking in a whopping 17,000 orders in its lifetime but was considered a massive risk that Boeing executives were “betting the company on”. The engineering culture and calculated risk in a Cold War environment fueled the company’s rise to a dominant position in the market. Critically, the success validated the approach of putting the engineers in charge.

But by the time the 1970s rolled around, attitudes had changed regarding what the airline industry would like like in the new century.

Deregulation

In 1978, Congress passed the bipartisan Airline Deregulation Act — an act that prohibits states from regulating prices or service provided by airlines and disbanded the US Civil Aeronautics Board.  

Congressional votes for the Airline Deregulation Act of 1978 *it wasn’t close

Airlines adopt a hub-and-spoke model where a few hubs like New York, Chicago, or Atlanta as hubs that connect smaller markets (the spokes). This maximizes load on any individual aircraft (because you’re carrying people going to the hub and people connecting) but requires fewer aircraft for any individual carrier. But the act also allows smaller, low fare, and regional airlines to grow, increasing the aggregate demand for large aircraft. The act also established the Essential Air Service which subsidizes flights from smaller communities (roughly $300M / year) which Boeing indirectly benefits from. For example a flight from Wichita to Boise is almost certainly not profitable to maintain while a flight from San Francisco to New York is. These subsidies and regulations are mostly aimed at airports and airlines but are boons for Boeing as they prop up aircraft demand. But there’s direct help as well.

Delta’s Hub-and-Spoke model starting in 1980. Courtesy of Smithsonian National Air And Space Museum https://airandspace.si.edu/stories/editorial/airline-deregulation-when-everything-changed 

The Clinton administration blesses and protects a merger between Boeing’s only other American competitor, McDonnell-Douglas. Leaving only the French Airbus as a competitor, but the government only buys American so the contracts Boeing wins from the government starts to balloon.

NYT Timesmachine. July 18, 1997. https://timesmachine.nytimes.com/timesmachine/1997/07/18/188727.html?pageNumber=80

According to Boeing’s 2022 10-K, 40% of Boeing’s total revenue comes directly via the Department of Defense. It began heavy lobbying to remain the incumbent aircraft manufacturer ($288M since 1998). The FAA even lets the company regulate its own planes starting in 2009 (“self-regulation”).

Graph of via SEC EDGAR. https://www.sec.gov/edgar/browse/?CIK=0000012927

Shareholder Value

Boeing’s executives changed after the McDonnell-Douglas merger. The new executives are more professional manager types as opposed to draftsmen or aircraft designers as before. The priorities seemed to shift to focusing on winning contracts and returning shareholder value. Boeing was a leader in dividend and stock buyback outlays until the 2 Max crashes of 2019: 

Dips in shares outstanding corresponding to stock buybacks. 1998– 2019. 

So there’s a duo of incentive problems: (1) you lacking any domestic competition in commercial aviation and (2) incentivized to think on a quarterly basis.

The company also began looking to “risk sharing partners” (as far as I can tell, this just means outsourcing) and cut costs. For example, aircraft design was outsourced to a design center in Moscow until the 2022 invasion of Ukraine and software engineering for the Max was outsourced to Indian firms where engineers were reportedly paid $9 an hour . This practice has somewhat presciently been maligned by internal Boeing engineers — not least because wages for American labor have stagnated: 

Essay circulated on September 8th, 2003. via https://www.airliners.net/forum/viewtopic.php?t=213075

Conclusions

It’s a pattern of complacency, profit seeking, and corruption that kicks off the catastrophes for Boeing. The Max is emblematic of a colossus turning into a corporation, which in turn becomes corrupted by government guarantees and greed.

So the Boeing story isn’t about one company’s momentary mishap but rather a series of cautionary tales spanning decades. A cautionary tale of how even a one-time bastion of American engineering can spiral in the face of greed. A cautionary tale of how governments can — perhaps willfully — fly blind for decades. A cautionary tale of what happens when a private entity is devoid of accountability. And a cautionary tale of how catastrophes are constructed bolt-by-bolt.

Sources: 

• Library of Congress. https://tile.loc.gov/storage-services/service/ll/uscode/uscode1934-00103/uscode1934-001039013/uscode1934-001039013.pdf 

• SEC EDGAR ($BA). https://www.sec.gov/edgar/browse/?CIK=0000012927

• GovTrack.US. https://govtrack.us/congress/votes/95-1978/h1532

• Smithsonian National Air & Space Museum. https://airandspace.si.edu/stories/editorial/airline-deregulation-when-everything-changed 

• Simple Flying https://simpleflying.com/airline-deregulation-united-states/

• The BIG Newsletter by Matt Stoller. https://www.thebignewsletter.com/p/its-time-to-nationalize-and-then 

• Boeing — McDonnell Merger Vintage Article via New York Times https://timesmachine.nytimes.com/timesmachine/1997/07/18/188727.html?pageNumber=80 

• “Delegating Safety” by Joanna Schacter (Cornell). https://community.lawschool.cornell.edu/wp-content/uploads/2021/11/Schacter-final-1.pdf 

• Bloomberg. https://www.bloomberg.com/news/articles/2019-06-28/boeing-s-737-max-software-outsourced-to-9-an-hour-engineers

• Boeing Employee Essay https://www.airliners.net/forum/viewtopic.php?t=213075 

This video and post were inspired - in part - by the following:

• “Recoding America” by Jennifer Pahlka

• “The Righteous Mind” by Jonathan Haidt

• Amanda Su’s (NYT) “From Substance to Shouting: The Demise of Political Debate in America”

Spent a fair amount of time editing this one. Hope you enjoy!

On November 7th, 2023, the New York Times published an opinion piece titled: “From Substance to Shouting: The Demise of Political Debate in America”. The video by Amanda Su is very good and touches on the real devolution of presidential debates. I don’t want this article to be viewed as a critique but more as an addendum or supplement to that video because one rule change is consequential to the tone of televised debates, but it’s part of a much larger shift in partisanship, organizational structures, and the American media landscape. Let’s first take a look at the 2008 rule change.

📝 Context: The Rule Change

In 2008, the Commission on Presidential Debates (CPD) - the nonprofit responsible for organizing presidential debates - changed debate rules to allow candidates to directly respond to one another as opposed to responding directly to a moderator. This loosens the structure of a debate and allows a substantive topic to spiral into a considerably less substantive one rapidly.

A contentious moment during the final 2016 DNC primary debate (The original question was about raising the taxable income limit for Social Security)

What does a moderator even aim at here? (The original question was about a ceasefire in Syria)

The structural vacuum is taken advantage of by individual candidates, but it’s ultimately the organizers’ responsibility to create rules that prevent such chaos.

Instances of “crosstalk” in debate transcripts (see footnote for more detail)

With that said, we need to talk about the Commission on Presidential Debates and how they came to run the debates in the first place. We need to go back to 1988.

🤔 Bipartisanship = Good? LWV vs. CPD

The League of Women Voters (LWV) - a nonpartisan nonprofit group - was responsible for organizing debates in the 1976, 1980, and 1984 general election cycles. From 1988 onward, the Commission on Presidential Debates - a bipartisan nonprofit sponsored by the Democratic and Republican National Conventions - took over. I want to draw attention to nonpartisan vs. bipartisan, a subtle but important distinction. “Bipartisanship” carries with it a value judgement and is thought of broadly as good (or, at least, value neutral). It gives a sense of compromise and consensus. But the major political parties can agree on courses of action that are not in our interest. Making debates another campaign stop is one example of consensus resulting in a worse product.

By the October 1988, LWV controlled debates in consultation with the candidates’ parties. 2 weeks before the first debate in the 1988 general election, the parties together approached LWV demanding complete discretion over: the questioners/moderator, audience composition, hall access for the press, and control over “other issues”.

The League politely rejected the demands saying:

After LWV refused to participate in debate organization, debates no longer sat outside the apparatus of the two-party campaign cycle. The CPD - and the parties by extension - were free to control every aspect of debates including barring third party candidates from participation and rule changes.

We can take a broader view of the debates and the media landscapes in which they exist if we zoom out a bit further though. Let’s look back to debates before either LWV or CPD. Let’s go back to 1858.

🎥 Media Landscape

Debates have changed substantially since the Lincoln-Douglas debates of 1858 in terms of their structure:

Blocks of speech from sample of debates from 1858 - 2020. Patterned segments represent periods of heavy interruption / crosstalk

And obviously in terms of the media ecosystems surrounding the debates:

• In 1858, Lincoln and Douglas debated at a time when information transmission had rapidly accelerated because of the inventions/construction of railroads, phonographs, and telegraphs. Making the debates a media event that boosted the profile of both men

• In 1948, the first Republican Primary debate was held over the radio so an estimated (roughly estimated) 40-80M Americans tuned in to listen to the men argue over one topic: outlawing the Communist Party in America

• In 1960, John F. Kennedy and Richard Nixon debated in the first televised presidential debate

• In June 2007, the Democratic and Republican primary debates featured social or content networks (namely, YouTube) for the first time

Evolving debate structures are, in part, a function of the news mediums of the time. Today, debates take place in the context of mature, short-form social/content networks where engagement metrics are the most important things. The post-2008 rules in the context of this media ecosystem create the conditions for candidates to simultaneously: (1) project power during the debate by taking up more space and time while giving themselves ample opportunity for a grandstand viral moment and (2) deny their opponents that same space, time, and virality. Interrupting in this paradigm is just good strategy.

As a note and observation: the parties likely had good reason to fear losing America’s collective attention. In the 1992 and 1996 general elections, Ross Perot won 18.9% and 8.4% of the popular vote - a ridiculously good performance by today’s standards and undoubtedly a threat to the major parties’ duopoly over American politics.

The results of the 1992 general election’s popular vote

Meanwhile, the debates in 1996 and 2000 saw a noticeable wane in viewership. And to connect this back to the media landscape, leading cable news(?) programs like “The O’Reilly Factor” had hosts with a more combative stance towards dissenting views; they pulled in massive audiences in 2004. For the CPD, all signs point to turning debates into a spectacle.

Debate viewership (in millions)

Now, we have to look at party evolution. Let’s look at the parties since 2012.

📊 Party Evolution

This is the part where I risk a controversial take, but here goes: The major political parties from 1988 through 2012 were more similar than they were different. The neoliberals and neoconservatives that dominated both parties ultimately occupy the same center-right space on the political spectrum. Because there’s ultimately little choice in their policy platforms, the parties are measured against one another rather than in reference to a broader political spectrum.

But the absorption of policy positions from the 2012 Tea Party movement into mainstream conservative thought and Trump’s introduction into the national political spotlight in 2015, pushed the party in a more paleoconservative direction. That direction change is far more dramatic than the more modest shifts to the left by the Democrats.

The shift away from one another creates a less focused and more hostile feel overall:

Polarity and Subjectivity for all debates from 2000 - 2020 split into pre- and post-2012

Conclusions

The end result of all of these factors - rule changes, organizational adjustments, evolving news mediums, and drifting political parties - is predictable chaos.

I do believe there is appetite amongst Americans to have real substantive debates in an environment that’s made for a 21st century media ecosystem. Debates that:

• Strictly and fairly enforce rules

• Confront candidates on factual inaccuracies in real time

• Live outside the campaign circuit

• Most importantly, foster substantive conversation on issues that effect a majority of Americans (there are many to choose from)

But until then, if you’re looking for genuine policy discussion, maybe these debates are better left unwatched.

This post was inspired - in part - by:

• “Recoding America” by Jennifer Pahlka

• “The Righteous Mind” by Jonathan Haidt

• Amanda Su’s (NYT) “From Substance to Shouting: The Demise of Political Debate in America”

Footnotes and Sources:

• The League of Women Voters organized the first debate in 1976 to 1988 (there were no debates between 1960 and 1976). Transcripts for primary debates are available from 1999 onward

• “Interruptions” include an instance of the debate’s transcription service writing “[crosstalk]” or some variant in the transcript or a candidate’s speaking block ending with a dash followed by the opposing candidate speaking.

• Subjectivity [0, 1] is a measure of how subjective, that is how influenced a statement is by emotion, opinion, or judgement. A higher number closer to 1 is more influenced by emotion, opinion, judgement

• Polarity [-1, 1] is a measure of how negative or positive a statement is. A statement that is more negative, that is contains more negative words, has a score closer to -1. In the context of presidential debates, it’s rare to get a debate that is truly “negative” or “positive” in aggregate using traditional models so we judge relative to other debates

• Paleoconservatism

• LWV Full Press Release

• Debate transcripts provided by UCSB’s “The American Presidency Project”

• All views = my own

Thank you for reading this article! If you enjoyed it, consider subscribing and sharing Statecraft or the video with a friend.

Hope everyone had a Merry Christmas and Happy Holidays! I’m so grateful that you’ve joined me here at Statecraft in its inaugural year - your support makes this newsletter possible and it truly means the world to me. I hope you’ve enjoyed my articles so far and I’m excited to bring you even more in the new year. Hope you enjoy this subscriber-exclusive post + video! 

Scale almost always finds a way of amplifying preexisting issues and laying bare some unforeseen challenges with new technology. 2023 was the year that “ChatGPT”, “LLMs”, and even “AI” truly entered the mainstream vernacular. And as such, I want to outline some questions and musings about AI going into the new year.

Graph showing Google Search Interest for AI topics via Google Trends. 12/27/2019 - 12/27/2023

Please feel free to comment your thoughts if you feel so inclined. Maybe as the new year goes by, we’ll revisit some of these questions in isolation. Now, let’s get into it:

🍪 Will AI-Generated Content Eat Itself? And How/If Platforms Adapt?

I know this is 2 questions but they’re intertwined so I’m taking my liberties. I came across NotByAI.fyi a few months back and that really kickstarted this train of thought. The site poses an interesting question that reminds me of that game “Snake” on Nokia phones:

Snake on a Nokia phone - rendering. Courtesy: Nokia Phones

Basically, the thought process goes:

1. AI has lowered the bar for creating tremendously detailed or realistic content that achieve “viral” status (think of the Pope’s Balenciaga jacket or Trump’s less-than-peaceful arrest). This lower bar allows for a critical mass of AI generated images, text, and video to flood the Internet at a much higher rate than new human-generated content.

AI Generated images of the Pope in a Balenciaga puffer jacket

2. But this AI-generated content (“AIGC”) is imperfect despite its realism. See the Pope’s right hand in the left image above and his left hand in the right image above

3. Because models are trained from content and its associated metadata (e.g. alt text, captions) scraped from around the internet, models will incorporate AIGC into newer training data. Assuming there is no reliable, systematized way to tell if a piece of content was generated by Dall-E, Midjourney, Stablity, Gemini, ChatGPT, LLaVa, BARD, etc.

4. Those new training data will incorporate imperfections or echoes from the uncanny valley more and more as the concentration of AIGC on the internet grows

An empirical study of the uncanny valley

Tying back to Snake: this is a feedback loop where the quality of model output degrades as the concentration of AI generated content grows.

Theoretical depiction of Human vs. AI generated content

I’ll be happier if this process doesn’t play out or if training data becomes more curated to preempt this.

AIGC’s explosion in 2023 has given rise to changing aesthetics and policies on content platforms. Here are a few platforms I selected:

• DeviantArt: Worked with Stability AI, then effectively forced opt-in by artists to have their images used in AI training data for DeviantArt’s in-house image generator (DreamUp), then reversed that policy. All while the artist community was being divided by both the initial cooperation with Stability and AI propagation on the platform

• Medium: Made changes to its incentive structure to incentivize “high quality human writing”, explicitly requiring Medium membership for monetization and adjusting monetization metrics to more heavily weight engagement

• TikTok: Added a “Label as AI-Generated Content” button (anecdotally this isn’t used very often and seems like it’s just there to preempt EU / US inquiry into AIGC on the platform)

As a content creator, I’ll be listening to platforms on what even constitutes “AIGC” in their view and how - or if - they prioritize human-generated content.

🔐 Will Privacy and Copyright Haunt AI?

Convoluted privacy settings, opaque training data, personal info leaks, and lawsuits breed trust crises. All of these have been discussed in the context of AI to some degree. Privacy researchers have pointed out the labyrinthian process of requesting to remove personal data from AI training datasets as well as how prodding AI for personal info may actually yield PII. Although I mostly see these concerns coming from the AI / privacy practitioners of the world, I’ll be curious to see if the concerns will spill over to the general public (e.g. will “OpenAI sees all your data on the internet no matter what” become the new “Facebook listens to your private conversations to serve ads”).

I’ve been acutely aware of legislation restricting AI use working its way through statehouses and cities. They generally seem to have privacy and data protection (for both consumers - “Private Sector Use” - and governments - “Government Use”) at their core:

Chart of AI Legislation categories across the US in 2023

Additionally, the New York Times has sued OpenAI for copyright infringement, trademark dilution, and unfair competition. The Times specifically calls out its articles’ prominence in the Common Crawl dataset as well as ChatGPT’s ability to seemingly bypass NYT’s paywalls. It’s a case that will be highly consequential for the viability of training sets in commercial AI products and the future of journalism:

The court filing is filled with examples like this, fun read.

Questions of ownership - of both copyright and personal data - will continue into the new year and probably well beyond.

💡 How Will Ideological Splits Effect AI?

We probably could use a whole newsletter / video just for this one. One of the reasons I get particularly hyped when talking about AI is it asks a lot of questions - fundamental questions - about us, humans. “What is art? What is intelligence really? Who owns what? Is Artificial General Intelligence just around the corner or 50 years away?” And because of that, the vision for where AI will be in the next decade plus varies wildly depending on who you’re talking to.

This split is partially capture by the ousting and swift reinstatement of Sam Altman of OpenAI.

Perhaps the most intriguing schism I’ll be following is the Effective Altruist (EA) community and academic researchers. There seems to be fundamental disagreements here surrounding what issues to even focus on. Whereas academic researchers seem to focus on issues like: bias, disinformation / hate speech, privacy, and child abuse material (CSAM) that have an urgent need to be safeguarded against in the now, the time horizon for EAs is perceptibly longer-term. EAs focus on the benefits of Artificial General Intelligence as an end that justifies moving quickly without hinderances (e.g. safety guardrails) and point to human extinction level events as primary safety concerns to focus on.

🏅Honorable Mentions

Those are my top three going into the new year but that’s just me! They’re far from the only storylines. Here are a couple more topic-specific stories that I’ll be monitoring as well, we’ll call these “Honorable Mentions”:

• Labor movements and labor commodification

• Urbanism, self-driving, and the fight for the future of cities

If you enjoyed this article or video, please consider subscribing below. It’s a free way to support the content, we’ll never send you spam, and you can always change your mind later. Thanks for reading/watching!

Arman Madani

I like to overanalyze things. Mostly: big ideas at the intersection of tech, policy, and media

www.youtube.com/@thearmanmadani

📝 Notes/Sources:

• “The Pope Francis puffer coat was fake – here’s a history of real papal fashion” - TheConversation

• “Fake Trump arrest photos: How to spot an AI-generated image” - BBC

• On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?

• “DeviantArt’s Decision to Label AI Images Creates a Vicious Debate Among Artists and Users” - ARTNews

• “New Partner Program incentives focus on high-quality human writing” - Medium

• The AI Trust Crisis - Simon Willison’s Weblog

• “See The Websites that Make AI like ChatGPT Sound Smart” - Washington Post

• Artificial Intelligence 2023 Legislation 

• NYT Sues OpenAI

• “Can I remove my personal data from GenAI training datasets?” - Knowing Machines

• OpenAI’s Board Pushes Out Sam Altman, Its High-Profile C.E.O.

And we’re back. Thank you to all the new members who’ve joined us from Medium over the last couple weeks and existing members for sticking with us! Your support makes this newsletter possible.

This is our first entry in the new Antitrust Series, in which we study concentrated corporate power, industry calcification, political influence, and - of course - policy solutions. We’ll revisit this series from time-to-time.

Since 2020, Americans have experienced rising food prices while farm closures have ticked steadily upward. Inflation and supply chain issues stemming from the pandemic have been explicitly or implicitly blamed in the news. However, the inflation narrative overlooks a more endemic, structural problem with the industry at large.

Through a string of unchecked acquisitions over 30 years, Tyson Foods, Cargill, JBS USA Holdings, and National Beef have gained control of roughly 85% of the total hog, cattle, and poultry processing market. For brevity, we’ll call these four meat processing corporations “BigAg”.

Designed by @theArmanMadani. Analytics powered by Dubo.gg

As we conducted our research, we found example after example of clearly negative economic outcomes for consumers, workers, and farmers tied directly to BigAg’s dominance. The upshot? There are policy solutions and hope for a better food system as a whole…if our elected officials would kindly act on them. We’ll go over 3 effected groups and discuss policy solutions at the end of each section.

🛒 Consumers

Let’s start at the checkout aisle. There have been high inflationary periods throughout history, but none have been driven as much by corporate profits as the inflation spike of 2021-2022. In theory, if everything including raw materials and labor get more expensive, prices will increase but profit margins should not. But when we overlay BigAg’s input costs and prices consumers pay, we see the following:

Designed by @theArmanMadani. Analytics powered by Dubo.gg

In this chart, price changes (“store price change”) may seem consistent with various inflation-related price increases. But importantly, the producers’ input costs - a leading indicator of what consumers will pay - diverges from the the price consumers end up paying in grocery stores. Looking forward, more consolidation in the market also opens the door to further price increases as meat products are sourced from BigAg as opposed to in-state, regional, or local sources (higher transportation costs).

The chart is an example of profiteering/price gouging at a time when Americans are finding it harder to get by financially. But taxpayers are actually paying BigAg twice: once at the checkout aisle and again via taxes. Namely, through the Farm Bill.

The Farm Bill sets the food policy for the country. Most of the provisions in it - 81% - go to SNAP programs (colloquially called “Food Stamps”). Meanwhile, the other 19% increasingly benefits BigAg. Congress renews the bill every 5 years; it will be renewed again this year. The non-SNAP provisions cost taxpayers $165 billion in 2018 and are slated to cost $287 billion in 2023 with the passage of a $1.51 trillion dollar renewal (the largest in US history).

Designed by @theArmanMadani. Analytics powered by Dubo.gg

Tyson’s operation covers roughly twice the land area of New Jersey at about 10 million acres or 22,500 times the size of an average farm. That’s a lot of insurance for taxpayers to cover.

Proposed Policy Solutions: Break up the 4 meat processors to promote competition and honest price discovery. Expand the Packers & Stockyards Act to monitor margins on meat products, prevent opportunistic price gouging/fixing, and prevent further acquisitions.

🛠 Labor (featuring Regulatory Capture)

We estimate that about 350,000 people work for Tyson, JBS, Cargill, and National Beef in the US. 350,000 employees who are largely powerless against behemoths which abuse three fundamentally broken systems. (1) Labor Laws: that are at best outdated and, at worst, actively anti-labor. (2) The Immigration System: that simultaneously enables BigAg to procure cheap migrant labor and can be weaponized to prevent that same migrant labor from organizing for or demanding humane working conditions. (3) The Criminal Justice System: namely, that involuntary servitude can be used as a punishment for a crime (13th Amendment) and BigAg corporations are active, willing participants in this system.

We really want to emphasize this point: labor’s powerlessness in this industry is enshrined in laws that were crafted by BigAg corporations. Here are just a few examples:

• Anti-whistleblower laws (“Ag-Gag Laws”) ostensibly crafted to prevent whistleblowing related to animal abuse, also outlaw free speech any whistleblowing by laborers in states across the country, so public outcry over working conditions

• States in which BigAg companies largely operate have started loosening or repealing their child labor laws as in Arkansas - Tyson’s home state (bleakly opening a new pool of cheap, vulnerable labor)

• And in 2020, a lobbyist was able to secure an executive order keeping meatpacking facilities open without protections for workers which resulted in COVID spreading rampantly throughout workers’ communities

Proposed Policy Solutions: I mean…there’s modern slavery, free speech suppression, and bribery lobbying here so there’s a lot of systemic change needed. Start with repealing Ag-Gag laws, implement strong national child labor laws, abolish involuntary servitude, strengthen the National Labor Relations Board to crack down on union-busting, and end Fair Labor Standards Act exemptions for farm workers (to start).

🚜 Farmers

Tyson, JBS, Cargill, and National Beef sit at a chokepoint in the market: between grocery stores and the actual farms. While BigAg increases prices for consumers, farmers lose revenue. After a huge overhaul of US farming in the 1960s and ‘70s spearheaded by more non-farm employment opportunities and increased productivity, the number of US farms has decreased approximately 20% over the last 30 years:

Designed by @theArmanMadani. Analytics powered by Dubo.gg

BigAg’s consolidated power throttles the price discovery process. In this case, price discovery is a literal auction for livestock and poultry. At auctions, the more buyers there are, the higher the potential sale price as buyers bid against one another. But meat processors are now the primary buyer from feedlots who in turn get their cattle from background ranches (who get their cattle from cow-calf ranches). Cheaper sales from feedlots to meat processors, reduce prices downstream for background ranches and cow-calf ranches.

Over time the share of feedlot-to-meatpacker sales done via contract has increased to 72%, so BigAg sets the price outright with no price discovery process at all. There have been Senate hearings aimed at curtailing BigAg’s power in this vein specifically; namely, to reduce contract purchases to no more than 50% of total feedlot-to-meatpacker sales. Again, this is a good start but it’s ultimately a band-aid for an industry that needs more acute intervention to encourage real competition. Collusion and price fixing would just become a cost of doing business (or at least more so than they seem to already be).

Proposed Policy Solutions: Worth noting again, break up the 4 meat processors. Limit the percentage of contractual sales to 40% for each individual meat processor. Implement a fairer Farm Act that (1) targets subsidies to incentive food production rather than feed production and (2) invests in local and regional food systems.

💭 Closing Thoughts

What gives us hope is the proposed policy solutions are not out of reach and the benefits of implementing the changes help an overwhelming majority of Americans while also promoting food security. The post-pandemic economic shocks co-opted by BigAg as air cover to gouge consumers, squeeze margins out of farmers, and exploit workers, may backfire and spur political will to create real reform.

Thank you for reading this edition of Statecraft! If you found it informative, consider sharing Statecraft with a friend. You can also connect with us on YouTube, Twitter, or Instagram. Links below:

🎥 YouTube

🐦️ Twitter

📸 Instagram

I couldn’t fit everything I wanted to talk about in this article, including the fact that there’s a whole climate/environmental dimension to BigAg’s operation and the complex minutia of the actual auctions, so I’ve included articles and videos that I watched to prep this article below:

1. “Are declining auction prices a result of bidder behavior or auctioneer strategy? Evidence from cattle auctions”. Journal of the Agricultural and Applied Economics Association. Jeffrey H. Dorfman, Wenying Li, Jingfang Zhang.

2. “Bidding Strategies and Price Discovery”. Community Economics by Forte (Medium)

3. “USDA ERS Climate Change Data”

4. “How 4 Companies Control the Beef Industry”. Vox (YouTube)

5. Various food/farm competition policy articles. Farm Action US

Quick updates before we begin:

New Perks for Subscribers! Starting today, we’ve added the following perks:

• Enabled comment section (leave feedback or suggest a topic!)

  • We’ve already received suggestions for AI and climate topics, and those are in the works!

• Sub-only posts

• Access to our full archive

• Early access to interactive web and video projects

Lastly, Statecraft is now available on Google News! 🎉 

The outstanding student loan debt in the United States is 1.78 trillion dollars. For reference, the GDPs of developed countries like South Korea, Australia, and oil-rich Saudi Arabia are less than $1.7T respectively. The explosion in tuition costs and associated debt accumulation is undeniably a looming economic crisis. It’s a crisis that requires a basket of policy solutions to transform higher education from the bottom up - including, yes, cancellation of some federal student loans. But while the fight to cancel student loan debt continues, we think it would be especially worthwhile to explore some data and highlight potential solutions for this debt burden.

📈 How We Got Here

Tuition cost and fee inflation have outpaced CPI Inflation 4-to-1 over the last 40 years:

Animated versions of these charts on Threads (or Twitter): @theArmanMadani

The resulting $1.78T of debt includes $1.65T that is federally owned (93%). 45M Americans have federal student loan debt. Roughly half of all the debt belongs to individuals with graduate degrees while the other half belongs to individuals with undergraduate degrees; though it’s worth noting that there are fewer graduate students overall so the debt held per student is higher for graduates:

Animated versions of these charts on Threads (or Twitter): @theArmanMadani

Student enrollment in higher ed institutions has risen consistently since 2000. With higher demand for college degrees - including out-of-state demand for state universities - tuition has risen precipitously. Rising enrollment is also associated with a changing US labor profile; for example, manufacturing jobs were eclipsed by “business and professional services” jobs as well as healthcare, education, and retail jobs. One troubling manifestation of the demand for higher degrees can be seen during the 2008-09 Recession as newly unemployed individuals scrambled for retraining. From 2008 through 2009, enrollment in private for-profit colleges rose disproportionately compared to private non-profit and public colleges. Private for-profit colleges have extraordinarily high rates of default, near 16% of borrowers default within 3 years.

During this period of heightened demand, a more obvious inflationary effect took place: state funds were cut for public colleges on a per-student basis in the run-up and fallout of the Recession.

💸 What Now

[WARNING: I include my opinion in this section, I look forward to a lively discussion in the comment section]

There are short and long-term solutions needed to address the debt problem.

Short-term, students need recourse. Defaulting on loans creates a cascade of consequences that necessarily harm borrowers earning potential for years (e.g. colleges can withholding proof of attendance, decline in credit score, wage garnishments, etc.) There are loans that are highly likely to default regardless after the COVID moratorium on payments ends. These loans should be forgiven/cancelled. The fallout of default en masse - especially amongst low and middle-income degree holders - would present a much greater cost to the economy than the cost of cancellation. Additionally, controlling what information can be reported to credit bureaus (like medical debt) and regulating what information colleges can withhold would also reduce negative income effects caused by default (e.g. a private for-profit institution cannot withhold proof of attendance in the event of default).

Long-term, controlling the underlying cost of higher ed itself is an imperative. Federal support for and expansion of programs like The California Promise - which provides free tuition for California residents who attend CA community colleges - would eliminate the burden for some new graduates. Apportioning more existing or raising new tax revenue to invest in underfunded private colleges (e.g. HBCUs) and public colleges can reduce tuition costs by supplanting/exceeding the aforementioned state funding cuts as well.

This is a complex issue which requires comprehensive policy solutions. But one of the reasons student debt has received the attention it has is because of the opportunity it presents. An opportunity to extend a stimulus to low/middle income earners in the short-term and transform the higher education system in the long-term.

Thank you for reading this article! If you enjoyed it, consider sharing Statecraft with a friend. You can also connect with us on Threads, Twitter or Instagram. We’re working on our first long-form video content as well, so if you want to see what we produce, head over to YouTube or TikTok. Links below:

🧵 Threads

🐦 Twitter

📷 Instagram

🎥 YouTube / TikTok

First and foremost, welcome to all our new subscribers! We’re so happy to have you here. Our mission is to bring you honest in-depth analysis of American policy - for free. If you enjoy this article, consider sharing it with a friend too. We so appreciate your support!

While the stock market, unemployment rates, and GDP have recovered and even exceeded pre-pandemic levels, the same can’t be said about public transit ridership in cities across the US. In fact, the long road to ridership recovery (pun intended) complicates the economic picture for cities the longer the recovery drags on.

The Problem

Amongst the top 5 transit agencies by ridership, none have returned to pre-pandemic levels. If we look at all transit rail systems in the country, American public transit systems operate at 70% of pre-pandemic levels as of June 2023:

Notably, private vehicle miles traveled has recovered to its pre-pandemic level. The pandemic spurred a confluence of changes in behavior amongst consumers, businesses, and local transit agencies. These changes - listed below - are proving difficult to overcome:

• Public safety and health concerns pushed more people to use private cars or micromobility options (e.g. bike share)

• Public transit operators retired or shifted careers, causing services to degrade (e.g. delays)

• Work from home measures eliminated the need to commute from exurbs and suburbs to city centers.

The Bay Area’s BART system is especially illustrative here. BART has struggled more acutely than other major transit systems with daily ridership hovering around 30% of it’s pre-pandemic average. The system - as it was built - primarily served suburban commuters traveling to and from work in the Oakland and San Francisco city centers (where property values are also dramatically higher). In a post-pandemic, remote work environment which tech companies initially embraced, this source of ridership is severely reduced.

The Unreasonably High Return on Public Transit Investment

Public transit investments generally yield incredible returns on investment (ROI). As of 2020, the return was $5 for every $1 invested; that is, for every $1 spent on public transportation, $5 is added to GDP. A variety of industries are stimulated by the investment in physical infrastructure, jobs are created, and savings from vehicle ownership / maintenance are passed to consumers. Harder to quantify on a dollar basis but still important are the environmental benefits and access to opportunity gained from an expanded public transit network.

The Doom Loop

Fares from ridership typically pay for the daily operations of public transit systems but ridership also helps systems qualify for state and federal grants. Further, ridership is a key metric to prove to lawmakers with tight budget constraints that public transit is worthy of funding through state general funds. This makes low ridership in and of itself a problem, but the problem is compounded still by dwindling COVID relief funds which are projected to run out entirely within 2 years. The knock on effects continue from there:

• If transit systems can’t afford to operate and don’t get any additional funding, they cut back service or reduce the number of service lines.

• Wages for public transit employees are pared back

• More consumers rely on TNCs, taxis, or private vehicles and bear the respective costs

• Access to downtown areas where commerce occurs is dampened (resulting in lower revenues from sales tax)

This is a “doom loop”. We simulate this doom loop using the same $1B investment example as above:

What Transit Agencies Can (And Can’t) Do

Transit agencies can’t unilaterally implement congestion pricing to encourage public transit use nor can they change zoning laws. New York, for example, needed to get permission from the Federal Highway Administration to implement congestion pricing in 2024.

So, though additional state and federal funding are a required short-term solution, this moment calls for innovation. After all, there are only so many cuts to be made. Transit agencies across the country have begun experimenting with ideas like:

• Partnering with TNCs - Uber - to offer free rides to/from stations (Dallas Area Rapid Transit)

• Shifting services to accommodate late-night ridership and events rather than commuting for work (Bay Area Rapid Transit)

• Partnering with private real estate developers to build “transit-oriented developments” (simultaneously addressing affordable housing shortages and the last-mile problem public transit networks consistently face)

This is only the start; transit agencies can - and should - experiment with transforming commerce and spaces in/around stations as well. The next couple years will shape what public transit looks like for decades to come.

We highly recommend checking out our interactive web experience for this post! Built in collaboration with yarn.tech. Link below:

The non-profits PEN America and the American Library Association keep a catalog of banned books in the United States up to the 2021-22 academic year. In this 1 academic year alone, 1600+ books were irregularly banned from school districts across America.

Irregular bans tend to be ideologically driven. They are supplemental to and fall outside of the standard book exclusion process (e.g. standard exclusions prevent books that promote violent, hateful, or “mature/17+” topics from being placed on school bookshelves). Below, we’ve provided a sample of irregularly banned books in 2021-22:

From our initial cursory observations, we’ve identified some patterns in these irregular bans. However, we’ve collected additional information from government-provided data sources (e.g. Census data on school districts) and APIs (e.g. Google Books/Goodreads API) to speak to these bans more substantively. To our knowledge, this is the most comprehensive dataset on banned books in America. Our goal in providing this data publicly is 3 fold:

• Ongoing monitoring of banned books

• Promoting transparency into which books are being banned (and why)

• Holding public officials complicit or active in the bans accountable

Insights

First, we aim to measure the reach of the bans. By PEN America’s estimates, 2 million students attend school in a district that has banned books. When we supplement their data with information from the Census bureau, we believe this is an underestimate and, in reality, 3.8 million students are impacted (that’s roughly 7% of the school age population).

We then aim to map the states in which bans have occurred. Texas (578) and Pennsylvania (424) lead the nation, while Florida (409) and Tennessee (341) lag not-too-far behind. With Invisible Library, it is possible to map banned topics onto states as well, more on that below.

No. of Titles Banned by State. Data: PEN America. Visual: Statecraft

2 of the most valuable attributes we’ve added to the data include:

• A long natural-language description of each book pulled from Google’s APIs (this is what you might read on the back of a physical book).

• Tags or topics for each book. A book can have several tags such as “LGBTQ+”, “activism”, and “African American and Black Studies”. We wrote a script to parse this ourselves from categories provided by publishers via the Google Books API.

We wanted a better sense of the topics being banned. Before creating the chart below, we removed generic tags (e.g. “young adult fiction”) and consolidated similar topics (e.g. “Asian American” and “Asian American/Pacific Islander” were consolidated under “Asian American & Pacific Islander”).

Topic Concentration in Banned Titles. Data and Visual: Statecraft

The Topic Concentration chart above lends the clearest picture into the implied rationale behind the bans. Namely, the bans are not and have not been about the physical removal of a book from a shelf. The bans instead are meant to:

1. Virtue signal by people in positions of institutional power to voting-age parents interested in school choice, parental rights, and wedge social issues to the detriment of non-voting age students

2. Reject and exclude topics that challenge a perceived status quo from the public discourse (e.g. non-heteronormativity, non-cis identity, non-traditional gender roles, and non-Judeo-Christian books are targeted)

We did entertain the viewpoint from ban advocates that the books being banned are largely driven by age-inappropriate content. This viewpoint does not align with publisher provided maturity ratings as below:

Further, the particular age-inappropriate topics being referred to by ban advocates like violent or sexual content are specific to social violence and non-heteronormative sexual topics. Books that heavily feature heterosexual relationships and/or military violence are not being referred to by ban advocates nor are they disproportionally being targeted. The same applies to the argument that books regarding activism or examinations of race in America are too political for school age people; there is a specific activism and racial critique being referred to.

Final Thoughts

Fortunately, book bans are widely unpopular amongst parents across the ideological spectrum. The bans today look strikingly similar to bans in the 1980s. At that time, ban advocates argued books like Kurt Vonnegut’s Slaughterhouse Five were “anti-American”, “anti-Christian”, and “filthy”. Appeals against the bans made it to the Supreme Court in 1982, where the court split on a decision (Island Trees School District v. Pico); this left the First Amendment question looming.

Today, PEN America and Penguin Random House are suing Escambia School District of Florida for its bans. Perhaps this time, 40+ years later, we can get closure on the bans for good.

Twitter tweet

Thank you for reading Statecraft! We’re dedicated to providing detailed analysis on issues that matter. If you enjoyed this article, consider subscribing and maybe sharing with a friend as well!