Lattice of Coincidence
Alexander Cox | Razib Khan | Note to Readers | Urban Doom Loop, Revisited | Cognitive Debt: The Downside of LLM Tools
A lot of people don't realize what's really going on. They view life as a bunch of unconnected incidents and things. They don't realize that there's this, like, lattice of coincidence that lays on top of everything.
| Alexander Cox, Miller in the film Repo Man
…
Razib Khan
I’m constantly being astonished by new findings in scientific inquiry, like the ‘cognitive debt’ research I explore later in this issue. But a lot of my curiosity is directed to other fields of inquiry, like the burgeoning research in paleogenetics.
I recently read A Whole New World by Razib Khan, which summarizes the current thinking about prehistoric human emigration from Asia to the Americas, and in particular overturns the former premise that the earliest emigrants arrived from Beringia (the now-undersea lands between Asia and North America in what is now the Bering Strait) no earlier than 20,000 years ago, and that all pre-European Americans arose from those migrations.
Paleogenetics adds new tools to archeology and paleontology, and has revealed a starling new understanding of ancient humans in the Americas, Khan tells us [emphasis mine]:
The earliest Americans were related to a now-extinct group in coastal Siberia, distantly related to today’s Australo-Papuan populations. They were ocean-oriented and likely had small boats that they used to move along the shore. Many researchers believe that they exploited North-Pacific kelp forests, which can support maritime foragers. As they migrated eastward, they found isolated refugia surrounded by glaciated territory in western North America. Eventually, these earliest Americans arrived in America’s Pacific Northwest, south of the ice sheets, and settled in places like Oregon’s Paisley Caves.
That migration was 30,000-40,000 years ago, more than 10,000 years before Beringerians arrived and largely displaced — or absorbed — them.
The paucity of material remains, and the minimal impact on megafauna from these putative earliest arrivals leads us to suspect that these early humans existed at low population densities. The tools in Chiquihuite, Mexico are quite primitive and exhibit little change over time. Eventually, these earliest Americans moved southward and were the first occupants of the Monte Verde site in southern Chile. Here their population was larger, and the arrival of Beringians 13,000 years ago did not totally overwhelm them in the southern continent as it did in North America. Nevertheless, these unknown people who occupied the New World for nearly 10,000 years before the arrival of the ancestors of modern indigenous people remain like ghosts to us, leaving faint traces in the genetics of local populations, and only the most tenuous of material legacies. Their ecological footprint was light enough not only to allow for the flourishing of Pleistocene megafauna until the arrival of the Beringians, but it was also minimal enough that archaeologists denied their existence for decades, and geneticists failed to pick up their trace scent. But now? Both archaeologically and genetically, it’s a whole New World.
In recent decades, hidden mysteries like these first peoples in the Americas — distantly related to indigenous peoples of Australia and Papua-New Guinea, and whose genes live on in people living in the Amazon and nearby regions — have been revealed by paleogenetics rolling back the lattice of coincidence that lies upon deep time.
Khan’s writing is a grand exploration of the farthest reaches of human time, and I strongly recommend taking a look. In recent months he’s laid out the history of the Basque people, resisting all invaders in their mountain fastness, and their language, the only remaining non-Indo-European tongue in Europe. He’s also told the tale of the long, long trail of Australia’s indigenous peoples. His writing makes what could otherwise be dense and overly scientific captivating and mind-expanding.
Note to readers
In June, I was swamped by a cascade of family emergencies — in particular, my eldest son had emergency surgery, and I headed to Chicago to help out — which led to weeks of inactivity here at workfutures.io. Things are more stable — my son is recuperating well, and able to deal with the day-to-day on his own — and I am more or less back to normal. I will be trying to get back to my normal cadence.
Strangely, the number of subscribers and followers have been steadily rising, even though I haven’t been writing much at all. Hmm.
If you dislike subscriptions — too many, can’t be bothered, please don’t ask — you can instead just make a one-time contribution at Ko-Fi.com/workfutures.
Urban Doom Loop, Revisited
Via Adam Tooze, I was reminded that there’s a great deal of commercial real estate debt coming due this year. Note the dent in the ‘b. Successful Payoff’ chart next to ‘Office’, and the 2025 column in ‘a. Institutional CRE Maturity Schedule which shows 957 billion coming due. That’s almost a trillion, dear readers.
The widespread adoption of distributed work — particularly out-of-office work, or working from home — has severely depressed the value of commercial office real estate.
I’ve written about this a bunch, as in An Upside-Down Story about an Upside-Down Reality from March [emphasis mine]:
Joe Gose, in Signs of an Office Market Bottom: ‘The Worst Is Probably Over’, seems to be trying to convince himself that the huge drop in commercial real estate has come to an end, or is at least beginning to come to an end. Or maybe, as Churchill put it, coming to the end of the beginning. But closely parsing the piece, it seems Gose has let his hopefulness write an article that should be read from the bottom to the top.
Gose writes that office building sales jumped nearly 21 percent last year, and leasing activity is up. But later in the article, we learn that many sales are driven by speculators buying up distressed and marked-down properties. The ‘good news’ seems to be limited to 'well-located, high-end properties in major U.S. markets', not everywhere.
Leasing activity appears to be up because companies are taking advantage of realtors willing to up the level of build-outs, in effect lowering the mortgage prices considerably. But this is summarized at the top of the story, a hypothetical reader would have to cut-and-paste the facts into a coherent story to get there.
And, pulled from the bottom third, this fact, which should be the headline: office vacancies
continued to climb last year in plenty of the largest U.S. office markets with vacancy rates of more than 20 percent, including Atlanta, Chicago, San Francisco and Los Angeles.
And then this out-of-the-blue statement, totally unsupported by data:
Companies are looking for more office space as work-from-home policies peter out, office landlords and developers said, with many seeking premium buildings with state-of-the-art amenities.
The nice folks at WFH Research have the actual data on working-from-home which shows a fairly consistent line of about 26%.
Now, we will see what the default rate looks like this year, and what are the implications for urban centers and the financial markets. But it could be stark and bleak.
Cognitive Debt: The Downside of LLM Tools
A large group of researchers — including Pattie Maes — created an experiment and reported on it in Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. They wanted to understand the differences in brain activity of people writing essays in three distinct, controlled groups: using AI (the LLM tool ChatGPT), using search engines (Google), and ‘brain only’, meaning no external tools.
The researchers relied on brain scans to see what parts of the brain were engaged by the subjects, and judged the essays with human teachers and an AI judge.
The ‘brain only’ group showed a broader engagement across a rich neural network across the brain, significantly greater cognition than the other groups, and were scored more highly by human judges. They write [emphasis mine]:
External support tools restructure not only task performance but also the underlying cognitive architecture. The Brain-only group leveraged broad, distributed neural networks for internally generated content; the Search Engine group relied on hybrid strategies of visual information management and regulatory control; and the LLM group optimized for procedural integration of AI-generated suggestions. These distinctions carry significant implications for cognitive load theory, the extended mind hypothesis, and educational practice. As reliance on AI tools increases, careful attention must be paid to how such systems affect neurocognitive development, especially the potential trade-offs between external support and internal synthesis.
The researchers tentatively offer these observations, stipulating that further research is required:
Perhaps one of the more concerning findings is that participants in the LLM-to-Brain group repeatedly focused on a narrower set of ideas […]. This repetition suggests that many participants may not have engaged deeply with the topics or critically examined the material provided by the LLM. When individuals fail to critically engage with a subject, their writing might become biased and superficial. This pattern reflects the accumulation of cognitive debt, a condition in which repeated reliance on external systems like LLMs replaces the effortful cognitive processes required for independent thinking. Cognitive debt defers mental effort in the short term but results in long-term costs, such as diminished critical inquiry, increased vulnerability to manipulation, decreased creativity. When participants reproduce suggestions without evaluating their accuracy or relevance, they not only forfeit ownership of the ideas but also risk internalizing shallow or biased perspectives.
Cognitive debt accumulates as participants rely on tools and fixed processes to escape the effort of thinking. The term ‘debt’ suggests that it must be paid back, but the reality may be that the true cost is the loss of capacity for critical thought and creativity, or never developing those abilities, at all.
Here are the researchers’ conclusions, with a few of my observations interspersed:
As we stand at this technological crossroads, it becomes crucial to understand the full spectrum of cognitive consequences associated with LLM integration in educational and informational contexts. While these tools offer unprecedented opportunities for enhancing learning and information access, their potential impact on cognitive development, critical thinking, and intellectual independence demands a very careful consideration and continued research.
The LLM undeniably reduced the friction involved in answering participants' questions compared to the Search Engine. However, this convenience came at a cognitive cost, diminishing users' inclination to critically evaluate the LLM's output or ”opinions” (probabilistic answers based on the training datasets). This highlights a concerning evolution of the 'echo chamber' effect: rather than disappearing, it has adapted to shape user exposure through algorithmically curated content. What is ranked as “top” is ultimately influenced by the priorities of the LLM's shareholders.
A very chilling insight: the LLM-using participants’ cognitive architecture is being shaped by the priorities of the companies behind the LLMs. A dystopian variant of ‘we make our tools, and they shape us’; ‘they make our tools, and their tools shape us’.
Only a few participants in the interviews mentioned that they did not follow the “thinking” aspect of the LLMs and pursued their line of ideation and thinking.
The undertow of LLMs: it’s hard to avoid their seduction.
Regarding ethical considerations, participants who were in the Brain-only group reported higher satisfaction and demonstrated higher brain connectivity, compared to other groups. Essays written with the help of LLM carried a lesser significance or value to the participants, as they spent less time on writing, and mostly failed to provide a quote from theis essays.
Since the LLM group didn’t do ‘the work’ they don’t own the results, and can’t remember much about it.
Human teachers “closed the loop” by detecting the LLM-generated essays, as they recognized the conventional structure and homogeneity of the delivered points for each essay within the topic and group.
Humans are likely better teachers than the ‘AI judge’.
We believe that the longitudinal studies are needed in order to understand the long-term impact of the LLMs on the human brain, before LLMs are recognized as something that is net positive for the humans.
I would go farther and say that LLM-based tools should be restricted until we understand the implications of long-term use on human cognition.
Keep reading with a 7-day free trial
Subscribe to workfutures.io to keep reading this post and get 7 days of free access to the full post archives.