The medical textbook that kills patients
In 1980, a medical student reading a respected gastroenterology textbook would learn that stomach ulcers are caused by stress and spicy food. The treatment was antacids and dietary changes. This was not fringe medicine — it was the consensus of the field, printed in authoritative sources, taught in the best medical schools in the world.
In 1982, two Australian researchers — Barry Marshall and Robin Warren — identified Helicobacter pylori, a bacterium, as the actual cause of most peptic ulcers. Marshall, unable to get his colleagues to take the finding seriously, infected himself with the bacterium, developed gastritis within days, and cured it with antibiotics (Marshall & Warren, 1984). They would win the Nobel Prize in Physiology or Medicine in 2005 — twenty-three years after their discovery.
Here is the temporal context problem: between 1982 and the mid-1990s, when the medical establishment finally updated its guidelines, millions of patients were treated for a bacterial infection with antacids and stress management. The textbook was correct — in 1980. By 1985, it was actively harmful. The information had not changed its words. The temporal context had shifted underneath it, and the same sentence that was medical knowledge in one decade became medical malpractice in the next.
This is not an isolated case. It is the normal condition of human knowledge. What is true now was not always true. What is true now will not always be true. And most of the errors in your thinking are not wrong facts — they are right facts from the wrong time.
Knowledge has a half-life
Samuel Arbesman, a scientist who studies the science of science itself, demonstrated in The Half-Life of Facts that knowledge decays at measurable, predictable rates — and that different fields decay at different speeds (Arbesman, 2012).
The metaphor is borrowed from nuclear physics. A radioactive element has a half-life — the time it takes for half of its atoms to decay. You cannot predict which specific atom will decay next, but you can predict with great precision how long it takes for half the sample to transform. Arbesman showed that knowledge works the same way. You cannot predict which specific fact will be overturned next, but you can measure how quickly a body of knowledge in any given field turns over.
The numbers are sobering. A study of nearly five hundred medical research articles on cirrhosis and hepatitis found that half of the accepted facts in those fields were overturned or significantly revised within forty-five years. In faster-moving fields, the half-life is much shorter. In surgery, the half-life of clinical knowledge has been estimated at approximately seven years. In computing and software engineering, it is arguably measured in months.
This means that if you learned a body of professional knowledge ten years ago and have not systematically updated it, a significant fraction of what you "know" is no longer accurate. Not because you remembered it wrong. Because the world moved and your knowledge did not.
The implication for your cognitive infrastructure is direct: every piece of stored knowledge needs a temporal tag. Not just "what is this?" but "when was this established, and what is its expected shelf life?" Without that tag, you are navigating with a map that may have been redrawn since you last looked at it.
The presentism trap
The decay of knowledge is one temporal context failure — treating old information as still current. But there is an equally dangerous failure in the opposite direction: treating current standards as if they were always obvious.
Historians call this presentism — the practice of judging past events, decisions, and people by the standards and knowledge of the present (Stearns, 2002). It sounds like moral progress. It feels like clarity. It is actually a failure of temporal context that distorts your understanding of both the past and the present.
Consider how people routinely judge historical figures for holding views that were utterly standard in their era. This is not a defense of those views — it is an observation about what happens when you strip temporal context from an evaluation. When you judge a seventeenth-century physician for not using antibiotics, you have not demonstrated superior morality. You have demonstrated an inability to account for what was knowable at a given point in time. The physician did not reject antibiotics. Antibiotics did not exist in his temporal context. Judging him for their absence reveals nothing about him and everything about your failure to contextualize.
Presentism is dangerous not because the past deserves uncritical reverence, but because it makes the present invisible to itself. If you believe that present standards are obviously correct — so obvious that anyone in any era should have held them — then you cannot see that your own era has blind spots equally invisible to you. The person in 2025 who mocks a 1950s doctor for recommending cigarettes for stress is often the same person who uncritically accepts the current consensus on some other topic that a 2075 observer will find equally absurd.
The antidote to presentism is not relativism — it is not the claim that all historical practices are equally valid. The antidote is temporal humility: the recognition that every era, including your own, operates within a knowledge horizon that is partial, shifting, and certain to be revised. Noting the "when" of any claim is the first step in seeing both its validity and its limits.
Temporal discounting: your brain underweights the future
The temporal context failures described so far involve knowledge — old facts versus new ones, past standards versus present ones. But there is a deeper temporal distortion that operates not on what you know but on what you value: temporal discounting.
Temporal discounting is the systematic tendency to treat future consequences as less real, less important, and less urgent than present ones. It is one of the most robustly documented phenomena in behavioral economics. Given a choice between one hundred dollars today and one hundred and twenty dollars in a month, most people take the money now — even though waiting yields a 20% monthly return that no investment on earth can match (Urminsky & Zauberman, 2014).
This is not rational deliberation. It is a perceptual distortion. Your brain literally perceives future events as less vivid, less concrete, and less emotionally salient than present events. Brain imaging studies show that people use different neural circuits when thinking about their present self versus their future self — the future self is processed more like a stranger than like you. You are, in a measurable neurological sense, less motivated to help your future self than to help your present self.
The consequences cascade across every domain. You do not save enough for retirement because the retired version of you does not feel real. You do not exercise because the health consequences are decades away. You do not invest in learning because the payoff is diffuse and distant. In each case, the information is available. The temporal context — "this matters later, not now" — systematically downgrades its perceived importance.
Temporal discounting is a temporal context failure of the highest order. You have the facts. You understand the logic. But the "when" of the consequences — future rather than present — changes their meaning in your perceptual system. The same fact, framed as happening today versus happening in five years, produces different emotional responses, different urgency assessments, and different decisions. The content is identical. The temporal context changes everything.
When best practices become anti-patterns
Nowhere is the half-life of knowledge more viscerally obvious than in technology. In software engineering, the gap between "best practice" and "anti-pattern" is often nothing more than a few years and a shift in the surrounding infrastructure.
In 2015, the Singleton pattern was standard object-oriented design — a way to ensure only one instance of a class existed. By 2020, it was widely considered an anti-pattern because it created hidden global state that made testing difficult. The pattern did not change. The surrounding practices — test-driven development, dependency injection, containerized deployments — changed the context in which the pattern operated. What was a sensible constraint in one temporal context became an unnecessary coupling in another.
The same arc played out with microservices. In the mid-2010s, at companies like Netflix and Amazon operating at enormous scale with hundreds of engineering teams, decomposing monolithic applications into independently deployable services was genuinely transformative. The practice was correct — for that temporal context, at that scale, with those organizational constraints. By the early 2020s, startups with three engineers were spending months building distributed systems for applications that could have been a single deployment. "Microservices" had become a temporal context error — a conclusion imported from one era and scale, applied without checking whether the conditions that made it true still held.
Thomas Kuhn described this phenomenon at the largest scale in The Structure of Scientific Revolutions (1962). Scientific fields do not progress in a smooth line. They operate within paradigms — shared frameworks of assumptions, methods, and accepted truths — that work well until anomalies accumulate and a paradigm shift occurs. What was normal science under one paradigm becomes obsolete under the next. Newtonian mechanics was not wrong — it was temporally bounded. It was the correct framework for the problems and measurement precision available in its era. When the temporal context shifted — when measurements became precise enough to reveal relativistic effects — the framework that had been true for centuries was revealed as an approximation.
The lesson for your cognitive infrastructure: every best practice, every trusted framework, every "how things are done" carries an implicit timestamp. When you adopt a practice, you are adopting a solution that was optimal for a specific set of conditions at a specific point in time. When conditions change — and they always change — the practice may persist long after its temporal context has expired.
The Lindy effect: time-tested versus time-expired
If knowledge decays, does that mean you should always prefer new information over old? No. And understanding why not is the key to calibrated temporal reasoning.
Nassim Nicholas Taleb formalized a principle he calls the Lindy effect: for non-perishable things — ideas, technologies, cultural practices — the longer something has survived, the longer its expected remaining lifespan. A book that has been in print for fifty years is likely to remain in print for another fifty. A technology that has been in use for a century is more likely to still be in use in another century than a technology released last year (Taleb, 2012).
The Lindy effect is not nostalgia. It is a survival filter. Things that persist through time do so because they have been tested against changing conditions and continued to work. The ideas in Euclid's Elements have survived for over two thousand years not because people are sentimental about geometry, but because the geometric relationships Euclid described are temporally robust — they work regardless of era, culture, or technology.
This creates a crucial distinction for temporal reasoning. There are two fundamentally different types of knowledge, and they require opposite temporal strategies:
Domain-specific facts — treatment protocols, software frameworks, market conditions, regulatory requirements — have short half-lives. These are the facts that Arbesman's research shows decaying at measurable rates. For these, you need aggressive temporal updating. The default assumption should be that anything over a few years old requires rechecking.
Structural principles — logic, mathematics, fundamental physics, core psychological mechanisms, time-tested heuristics — have long or indefinite half-lives. These have survived the Lindy filter. For these, age is a feature, not a bug. The older a structural principle is and the more temporal contexts it has survived, the more confidence it deserves.
The failure mode is confusing these two types. Treating a domain-specific fact (like a recommended technology stack) as a structural principle (timeless truth) leads you to build on expired knowledge. Treating a structural principle (like compound interest or feedback loops) as a domain-specific fact (probably outdated) leads you to discard wisdom that has earned its authority through temporal survival.
AI and the Third Brain: temporal awareness as a system property
Large language models present a vivid case study in temporal context failure — and in the emerging solutions to it.
Every LLM has a knowledge cutoff: a date beyond which its training data does not extend. GPT-4's original training data ended in September 2021. Claude's training data has its own cutoff. This means that when you ask an AI system a question about events, technologies, or research after its cutoff date, the system does not know what it does not know. It will generate confident, fluent, plausible-sounding text based on the temporal context of its training data — even when that context has been superseded. The phenomenon researchers call "hallucination" is often, at its root, a temporal context error: the model produces an answer that was correct in its training window but is no longer correct now (Kasai et al., 2024).
This is not a flaw unique to AI. It is your flaw, made visible. You also have a "training data cutoff" — the last time you deeply engaged with a topic. You also generate confident responses based on knowledge from your training window. You also do not automatically flag when your stored knowledge has been superseded. The difference is that AI systems are beginning to solve this problem in ways you can learn from.
Retrieval-Augmented Generation (RAG) systems address temporal staleness by connecting the language model to live, external knowledge bases. Instead of relying solely on training data, the system retrieves current information at query time and incorporates it into the response. This is the architectural equivalent of the temporal audit: before generating an answer, check whether the knowledge source is still current.
You can implement the same architecture in your own thinking. When you are about to act on stored knowledge, add a retrieval step: Is this still current? When was the last time I verified this? Has the field moved since I learned this? Use AI as a temporal checking layer — ask it to identify what has changed in a domain since a given date, then verify its response against current sources. The human-AI partnership here is not about AI having better temporal awareness (it does not, inherently). It is about building a system where temporal context checking is a structural feature of how you process information, not an afterthought.
The temporal context protocol
Here is the deliberate practice that builds temporal awareness into your perceptual system.
Step 1: Timestamp every claim. When you encounter a piece of information — in a book, article, conversation, or your own memory — identify when it was produced. Not just the publication date, but the temporal context of the underlying knowledge. A 2025 article citing 2015 research carries the temporal context of 2015, not 2025.
Step 2: Classify the knowledge type. Is this a domain-specific fact with a short half-life, or a structural principle with a long one? Software framework recommendations are domain-specific. The principle that tight coupling creates fragility is structural. Treatment protocols are domain-specific. The principle that correlation does not imply causation is structural. This classification determines how aggressively you need to check for temporal decay.
Step 3: Estimate the half-life. For domain-specific knowledge, how quickly does this field move? AI and machine learning: months. Software engineering practices: two to five years. Medical treatment protocols: five to fifteen years. Fundamental physics: decades to centuries. Your update frequency should match the half-life of the domain.
Step 4: Check for paradigm shifts. Has there been a Kuhn-style revolution in this field since you last updated? Paradigm shifts do not decay gradually — they invalidate entire frameworks at once. The shift from waterfall to agile. The shift from on-premise to cloud. The shift from classical to quantum computing (still in progress). If a paradigm shift has occurred, your stored knowledge may not need updating — it may need replacing.
Step 5: Apply the Lindy filter. Before discarding old knowledge as "outdated," ask: has this survived because it is genuinely temporally robust, or has it survived because no one has bothered to update it? A principle that has been tested across multiple eras and technological contexts has earned higher temporal confidence than a practice that became popular eighteen months ago.
Step 6: Note the "when" explicitly. In your notes, your decisions, your recommendations to others — always include the temporal context. Not "microservices are best practice" but "microservices were optimal for large-scale engineering organizations circa 2015-2020." Not "this treatment works" but "this treatment was the standard of care as of 2023." The explicit timestamp transforms a timeless-seeming claim into a temporally bounded one, which is what it always was.
The bridge to emotional context
In L-0165, you learned that cultural context is invisible until you cross a cultural boundary and suddenly see the assumptions you were swimming in. Temporal context operates the same way — you do not notice that your knowledge has a timestamp until something forces you to confront that the world has moved.
But there is another layer of context that is even more invisible than time, because it does not sit outside you like a date on a calendar. It sits inside you, coloring every piece of information before you consciously process it. Two people can read the same sentence on the same day in the same cultural context and derive completely different meanings — because one just received devastating news and the other just fell in love.
That is emotional context, and it is what L-0167 addresses. You now understand that the "when" of information changes its meaning. The next lesson teaches you that the "how you feel" at the moment of perception changes its meaning even more.
Sources:
- Marshall, B. J., & Warren, J. R. (1984). "Unidentified Curved Bacilli in the Stomach of Patients with Gastritis and Peptic Ulceration." The Lancet, 323(8390), 1311-1315.
- Arbesman, S. (2012). The Half-Life of Facts: Why Everything We Know Has an Expiration Date. New York: Current/Penguin.
- Stearns, P. N. (2002). "Against Presentism." Perspectives on History, American Historical Association, May 2002.
- Urminsky, O., & Zauberman, G. (2014). "The Psychology of Intertemporal Preferences." In G. Keren & G. Wu (Eds.), The Wiley Blackwell Handbook of Judgment and Decision Making. Wiley.
- Kuhn, T. S. (1962). The Structure of Scientific Revolutions. Chicago: University of Chicago Press.
- Taleb, N. N. (2012). Antifragile: Things That Gain from Disorder. New York: Random House.
- Kasai, J., Sakaguchi, K., Takahashi, Y., Le Bras, R., Asli, A., Yu, X., Radev, D., Smith, N. A., Choi, Y., & Inui, K. (2024). "Dated Data: Tracing Knowledge Cutoffs in Large Language Models." arXiv preprint arXiv:2403.12958.