🌀🗞 The FLUX Review, Ep. 96
April 20th, 2023
Episode 96 — April 20th, 2023 — Available at read.fluxcollective.org/p/96
Contributors to this issue: Neel Mehta, Ade Oshineye, Ben Mathes, Erika Rice Scherpelz, Dimitri Glazkov, Boris Smus
Additional insights from: Gordon Brander, a.r. Routh, Stefano Mazzocchi, Justin Quimby, Alex Komoroske, Robinson Eaton, Spencer Pitman, Julka Almquist, Scott Schaffter, Lisie Lillianfeld, Samuel Arbesman, Dart Lindsley, Jon Lebensold
We’re a ragtag band of systems thinkers who have been dedicating our early mornings to finding new lenses to help you make sense of the complex world we live in. This newsletter is a collection of patterns we’ve noticed in recent weeks.
“The hazard of being an autodidact is that you don’t know what you don’t know.”
😶🌫️🕴️ The disappearing problem
Imagine a product suffering ongoing latency problems. To prevent it from becoming unusable, we create a task force. We know what to do. Instrument! Gather measurements! Analyze! We are motivated to hunt down the culprit. There’s only one problem: the culprit doesn’t exist. We discover that no single component bears responsibility for the latency. The end-to-end latency is undeniable, yet in the detailed view, the latency seems to disappear.
This is an example of a disappearing problem. A disappearing problem is one that is very much present in the overall picture but which cannot be located in any particular part of the system. Disappearing problems are unintuitive. We tend to assume that the more closely we study a problem, the more easily we will be able to solve it. That is not always the case.
Disappearing problems often show up in high-complexity systems. Our example looked at a technical disappearing problem. We also see these frequently in human problems. Why can’t we follow through on a decision? Why can’t we pivot? Most of the time, no single person or team is to blame, but the system as a whole won’t respond.
Disappearing problems are a matter of perspective. The problem is undeniably there, but something about our framing of it creates the disappearing act. If we can change our perspective, then we can see the problem more clearly.
How do we do that? The first step is to recognize if we have a disappearing problem in the first place. Look for the common markers. Does the problem evade close examination when we try to break it down? Do our deep analyses reveal nothing of interest? Or, perhaps, do they become convoluted beyond our comprehension? If so, we might be dealing with a disappearing problem.
Next, we can examine our lenses. What assumptions, metaphors, and mental models are we using? Where are we taking shortcuts that might hide the problem? How are we defining the parts versus the whole? Does the way we’re looking at the problem align with the agency and incentive structure of the system? With the right perspective shift, clarity can suddenly emerge.
In the case of the high latency product, the problem disappeared behind our assumption that one particular component is the cause. If we shift our perspective to look at the system architecture, we might see that the latency adds up due to more fundamental properties of the architecture.
Oftentimes, we need to try on multiple perspectives before finding one that is effective. Then we need to accept it. Placing blame on one element is easier than accepting that the system needs to change. Our reluctance to change the system is why we have a hard time seeing the problem in the first place, especially if we’re caught in a Schelling Trap.
Finding the perspective that keeps the problem from disappearing is only the start. From there, we need to figure out how to change the system, and changing systems is hard. It will require experimentation, adaptation, and a willingness to accept failure. (Simple Habits for Complex Times is a book we’ve recommended in this space.)
🛣️🚩 Signposts
Clues that point to where our changing world might lead us.
🚏👾 Citing AI training, Reddit will start charging companies for API access
Reddit data has been highly valuable to AI companies, who need to train their Large Language Models (LLMs) on large, structured corpora of human conversations — something that’s hard to come by on the open internet. Reddit seems to have noticed this (and the potential risk of an AI company using Reddit data to build a competitor) and now plans to charge the likes of OpenAI for using the Reddit API to extract the platform’s data.
🚏🍻 Researchers put 25 AI agents in a virtual world and had them interact
In a recent paper, researchers created a virtual world called “Smallville” — rendered using old-school 16-bit graphics — and populated it with 25 virtual characters powered by LLMs. Each character had its own backstory and could interact with other characters. Surprising things started happening: characters could develop relationships, share information, and coordinate with each other. It culminated with one character planning a Valentine’s Day party and inviting friends; one of those friends then invited the first character’s “crush.”
🚏☢️ Europe’s largest nuclear reactor is now live in Finland
This week, Finland turned on the Olkiluoto 3 nuclear power plant — the largest single reactor in Europe. The plant is producing a whopping 14% of Finland's electricity and is expected to remain operational for “at least the next 60 years.” The launch came just days after Germany closed its last three nuclear power plants, the final step in the country’s 20-year-long plan to shift away from nuclear power.
🚏🇮🇳 India’s infrastructure push is making it lose forest cover faster than all but one country
Between 2015 and 2020, India lost almost 700,000 hectares (2,700 square miles, or about the size of Delaware) of forest cover per year. India thus had the second-highest rate of deforestation of any country, trailing only Brazil. A major factor is the Indian government’s current infrastructure push, which is clearing vast swathes of the jungle to build roads, railways, mines, power lines, and irrigation projects. (Interestingly, Brazil’s deforestation rate has fallen more than any other country since 1990, while India’s rate has risen more than any other country.)
🚏💸 Demand for cash fell to its lowest level in 20 years
According to financial projections from a major manufacturer of banknotes, demand for cash has dropped to its lowest point in two decades. COVID-19 was a key factor, since it accelerated the shift to digital payments, but technologies like QR codes have helped hasten the transition as well. Another report showed that cash accounted for 27% of point-of-sale transaction volume in 2018, but just 16% in 2022 — and that figure is projected to fall below 10% in 2026.
📖⏳ Worth your time
Some especially insightful pieces we’ve read, watched, and listened to recently.
Why Are There No Empires in Age of Empires? (Unmitigated Pedantry) — Critiques the historicity of strategy games like AoE and Civilization, which put you at the reins of an empire conquering other empires by means of total annihilation and population replacement. This is misleading, because the point of real empires throughout history has been to access the resources and labor of a subordinate population.
How Corporate Consolidation is Killing Ski Towns (Wendover Productions) — Examines how the rampant horizontal and vertical consolidation of the ski resort industry has harmed the towns that host these resorts. Locals are priced out, profits are siphoned off to distant headquarters, and well-paying management jobs are extracted from the local economy. It’s a surprising parallel to the problems caused by mining and other extractive industries.
Startups Are an Act of Desperation (Elad Gil) — Argues that, since startups are an ultra-high-risk, high-reward endeavor, most people will only do startups if they have little to lose and a lot to gain (whether that’s money, status, or career progression). There’s a reason that you don’t see that many big tech VPs going on to found companies: “they have already ‘made it’ career wise and financially, and there is less of a need to take risk.”
What Really Happened in 1971… (Michael W. Green) — Argues against the popular notion that productivity and wage growth suddenly diverged in 1971 (and that this was supposedly caused by the abandonment of the gold standard). In truth, the trend had been happening for a while before 1971, and the continuance of the trend has been more due to trade policy and conscious actions by the Fed.
How To Be a -10x Engineer (Taylor Troesh) — A humorous look at the many ways that you can waste 400 hours of engineering time a week: adding unnecessary overhead, creating “pointless rituals… that resemble work,” building bad technical architecture, indulging in the sunk-cost fallacy, and more.
🔍🪤 Lens of the week
Introducing new ways to see the world and new tools to add to your mental arsenal.
This week’s lens: Schelling Traps.
Schelling points are the default solutions that people arrive at if they can’t communicate; the classic example is meeting at noon at Grand Central Terminal in New York City. Schelling traps are situations where individuals or groups fall into suboptimal patterns of behavior because they can’t coordinate or communicate at the right level of abstraction. The prisoner’s dilemma is an example of a Schelling trap, where the lack of communication means that individuals are incentivized to betray each other even though they’d be better off if they both stayed silent.
It can be difficult to break out of these patterns, even when we believe we can do better. Breaking free requires a mutual and simultaneous deviation from the status quo. If we break out but others do not, then we’ll be less effective than they are. The status quo may be broken, but at least it’s some form of coordination.
Schelling traps are common within and between organizations, especially large ones. As the size of an organization grows and its structure becomes more complicated, the complexity of communication, coordination, and incentives increases. This leads to the emergence of whatever default is expedient — “I’ll use the same planning spreadsheet as last time” — rather than what is most effective.
An example of an organizational Schelling trap is a company that maintains redundant, uncommunicative, and siloed departments working on similar projects. Each department thinks their work is unique and vital. They are unaware of the duplicative or misaligned work from the other departments. Or perhaps they are aware, but they keep working independently. These departments lack the information or incentives to collaborate. Breaking free from this Schelling trap requires a coordinated effort from higher-level management. However, higher-level management may itself be caught in its own Schelling traps, leaving the organization unresponsive to the inefficiency.
It’s tempting to believe that at a sufficiently senior level of management these problems can be easily solved. This is the “if only the czar knew” trap. However what’s really needed is a higher level of awareness. With a sufficiently abstract view of the situation, the players can see the trap and work together to fix their broken coordination game.
© 2023 The FLUX Collective. All rights reserved. Questions? Contact flux-collective@googlegroups.com.