Learn it → Apply it → Save as Skill
A new technology is easiest to redirect before anyone understands it well enough to want to. While a system, platform, or policy is still new, changing course is cheap — but almost nobody yet knows what its real consequences will be. By the time those consequences are clear, the thing is woven into infrastructure, habits, and other people's plans, and changing course has become slow, expensive, or nearly impossible.
The Dr. Wallace's Favorite Mental Models is designed to turn useful ways of thinking into reusable tools. Each model gives you a fast visual overview, deeper interactive learning, practical application exercises and an AI-ready Markdown skill you can save for future use.
Instead of merely reading about mental models, you can gradually build your own AI skills library: a collection of portable reasoning skills that can help ChatGPT, Claude, Grok and other AI assistants apply the mental models you value.
Before a technology, feature, or policy scales, ask whether you actually understand its downstream effects — and if not, whether waiting until you do means losing the only moment when changing course was still cheap.
Important: this is a snapshot of the early stage, when control is high and understanding is low. Wait long enough to gather real understanding, and the two bars trade places — control shrinks to match what knowledge grew into.
Most useful at the exact moment you're deciding how fast to commit to a new technology, platform, architecture, or policy — while it's still unclear what its real consequences will be.
The dilemma names a structural bind. These four moves turn it into something you can actually act on — close to what Collingridge himself called "intelligent trial and error."
Sketch the decision's reversibility curve: how fast will the cost of changing course rise as adoption, dependency, or infrastructure builds up around it?
Make the first real version small, cheap, and easy to unwind — not the most complete or permanent version you're capable of building.
Build in fast feedback: pilots, metrics, sunset clauses, scheduled reviews — so real consequences surface before the exit door closes.
Protect the ability to change course for as long as reasonably possible: modular design, staged rollouts, opt-outs — even at some short-term cost to speed.
Open a use case to see what the model changes about how you'd approach it.
Early on, swapping a database or integration pattern costs a sprint. Two years and millions of records later, the same swap costs a quarter and a dedicated team — right when you finally understand its real limits. Decide what you can afford to learn the hard way, and what you can't.
A new AI feature ships fast because waiting for complete certainty about misuse means never shipping. The Collingridge move isn't "wait" or "ship and hope" — it's shrinking the blast radius (limited rollout, usage monitoring, a kill switch) so you can actually learn before the feature is load-bearing for other products.
Regulate too early with incomplete evidence and you risk freezing a beneficial technology in its worst-understood form. Regulate too late and the technology's business models, user habits, and lobbying power are already built around the harm. Favor rules that stay adjustable as evidence improves over rules that assume today's understanding is final.
A decision to widen roads instead of funding transit is cheap to reverse in year one and nearly impossible to reverse once zoning, commutes, and storefronts have organized themselves around cars for forty years. The dilemma is why so many infrastructure debates feel like arguing with the past.
Switching platforms in month three is an afternoon's migration. Switching after three years of integrations, trained staff, and dependent workflows is a multi-quarter project with real failure risk. Price the switching cost into the decision before you have any data on whether you'll need to switch.
Early in a career, changing direction costs a semester. A decade in, the same change costs income, identity, and a running start others already have. You rarely feel ready to decide at the one moment deciding is cheap — that mismatch is the dilemma, not a personal failure.
This is an illustrative example designed to teach the method, not a claim about any real company.
Expand for the origin of the term, what its author actually proposed, and how the idea has aged.
David Collingridge was a technology-policy scholar associated with the University of Aston's Technology Policy Unit. He introduced the dilemma in his 1980 book The Social Control of Technology, writing about the structural difficulty of steering new technologies toward good outcomes.
The information problem: a technology's full effects usually can't be known until it has been extensively developed and widely used. The power (or control) problem: by the time those effects are known, the technology is often so entrenched — economically, socially, institutionally — that controlling or changing it has become slow, costly, and difficult. Collingridge summarized the bind directly: "When change is easy, the need for it cannot be foreseen; when the need for change is apparent, change has become expensive, difficult, and time-consuming."
No — that's the Precautionary Principle, a stricter, separate idea that later critics such as Wired editor Kevin Kelly and philosopher Max More argued is so risk-averse it can block beneficial innovation indefinitely without actually resolving the underlying uncertainty. Collingridge's own proposed answer was different: "intelligent trial and error" — favor small, flexible, reversible, low-cost experiments and fast feedback loops over big, irreversible commitments made on incomplete information.
Yes. Scholars Audley Genus and Andy Stirling (2018) observed that while the "dilemma of control" is cited constantly in innovation research, much of that work engages only with the headline idea and skips Collingridge's fuller methodological and philosophical argument. They connected his original thinking to the modern "responsible innovation" and anticipatory-governance literature, arguing it deserves deeper engagement rather than replacement.
Treating the dilemma as a reason to delay every decision until the consequences are fully known guarantees you'll decide later, during the expensive, locked-in phase — the opposite of the point.
Ignoring the dilemma entirely and committing to the most permanent, least reversible version of a new system because speed feels like an advantage in itself.
"We can't know yet" is a real asymmetry, not a permanent license to avoid responsibility. At some point a decision — even a provisional one — has to be made.
Collingridge's own answer was flexible trial and error, not a blanket freeze. Conflating the two skips the part of his argument that is actually actionable.
Some decisions stay cheap to reverse for a long time (a pricing experiment); others lock in almost immediately (physical infrastructure). Apply the model to the specific reversibility curve, not technology in general.
The same bind shows up inside a single product team's architecture choice or a single person's career decision — scale the scrutiny to the stakes, not the size of the institution.
the cost of reversing the decision stays low no matter how long you wait — most feature flags, pricing tests, and copy changes never truly lock you in.
the consequences are already well documented elsewhere. If the information problem is already solved by someone else's experience, this isn't the right lens anymore — just use the evidence.
the stakes are genuinely low. Invoking a dilemma about irreversible technology lock-in for a trivial, cheaply-reversible choice is overkill.
This browser tool structures your thinking. It does not send your information anywhere or perform AI analysis.
Save as Skill converts the model into portable Markdown instructions that teach an AI when and how to apply the Collingridge Dilemma, what questions to ask, what evidence to seek, what mistakes to avoid, and how to structure its analysis.
Save the skills you find useful and organize them into your own reusable library. The Markdown is intentionally model-neutral, so it can be used wherever an AI system accepts uploaded files, project knowledge, custom instructions, or reusable text context. Platform capabilities differ, so this is a portable skill file rather than a claim of one-click installation.
Collingridge wrote about technology and society at large, but the same structural bind shows up at any scale where a decision gets harder to reverse as other decisions, habits, or infrastructure accumulate around it — a product's architecture, a city's roads, or a person's career path. Scale the analysis to the size of the stakes, not the size of the institution.
No — Collingridge's own preferred response, "intelligent trial and error," is about moving in small, reversible, well-instrumented steps rather than moving slowly in general. Speed and reversibility are different variables; the goal is to keep the second one high while moving at a reasonable pace on the first.
Finding that the decision's reversibility curve is actually flat — that changing course would cost about the same next year as it does today. If switching costs aren't rising meaningfully over time, the information problem still exists, but the power/control problem that makes it a true dilemma does not.