18389 Chap.9 The Collingridge Dilemma and Adaptive Regulation
The Collingridge Dilemma and Adaptive Regulation
The dilemma
The final part of Lecture 2 imagines 2031, five years after a pilot of autonomous taxis was approved. The service is now part of daily life, the developer has invested heavily, and cyclists have evidence of near-misses. The regulator finally has the evidence but may no longer have the power.
This is the Collingridge dilemma, associated with David Collingridge's 1980 book on the social control of technology.
Two problems
Early in a technology's life there is an information problem: its uses, markets and social effects cannot yet be known. Later there is a power problem: it is embedded in infrastructure, habits and institutions, and strong interests resist change.
A chart shows knowledge rising as the ability to change falls, so the point where both are high is never reached.
Adaptive design
Tools exist for each problem, pilots, sandboxes and monitoring for learning, and staged deployment, conditions, review and sunset clauses and temporary permissions for keeping control, but each can fail.
The chapter's model answer to the lecture's TfL exercise ties every measure to a harm, a monitoring method and a trigger for change.
What this chapter covers
- 01
The 2031 scenario
- 02
Knowledge rises while power to change falls
- 03
The information problem
- 04
The power problem
- 05
Tools for learning and for keeping control
- 06
Adaptive regulation and the design exercise
Worked example · free
Design a review clause for an AI triage pilot
- 1Information problem: run the pilot in one department with every AI priority checked by a clinician.
- 1Monitoring: compare waiting times and outcomes with a matched department without the tool.
- 1Power problem: grant a temporary permission that expires after twelve months.
- 1Trigger: suspend use if missed urgent cases exceed the matched department's rate.
- 1Explain that each measure names its harm, its monitoring and its trigger.
Key terms
- Collingridge Dilemma
- The Collingridge Dilemma is the structural tension between learning about a technology's effects and keeping the power to change it.
- Sunset Clause
- A Sunset Clause ends a rule or permission at a set date unless it is renewed.
- Temporary Permission
- A Temporary Permission is an authorisation granted for a limited period so that the regulator must decide again whether it continues.
The Collingridge Dilemma and Adaptive Regulation FAQ
What is the Collingridge dilemma?
It is the problem, associated with David Collingridge's 1980 book The Social Control of Technology, that early on we lack information about a technology's effects, while later, when we know more, the technology is embedded and powerful interests resist change. It is a structural tension to be lived with.
How can regulation keep the power to change course?
The Lecture 2 notes list staged deployment, conditions, review clauses, sunset clauses and temporary permissions. They warn that these are not magic: a review clause achieves nothing unless someone is able and willing to act on what it finds, so design matters as much as the label.
Does adaptive regulation mean experimenting on the public?
No. In the notes it means stating what the regime is meant to learn, capping how many people are exposed while doubts remain, and making sure the rules can still be stopped or redirected. It is a disciplined way of learning, not endless trial and error.
What is the difference between a review clause and a sunset clause?
Both preserve the power to change course. A review clause requires the rule or permission to be reconsidered at a set point, while a sunset clause ends it automatically unless it is renewed. Either only works if the regulator is able and willing to act on what the review or the expiry date reveals.
Assessment move
Draw the two curves from memory and label the information problem and the power problem. List the tools for each problem and write one way each tool can fail. Then attempt the lecture's TfL exercise yourself, writing every measure as harm, measure, monitoring and trigger, before comparing with the model in this chapter. Return to the 2031 scenario and write what each actor would say at a review meeting.
Compare your design with a classmate's and check whether each measure has a trigger.
Working through The Collingridge Dilemma and Adaptive Regulation in 18389? Sia is AskSia’s AI Law tutor — ask any 18389 The Collingridge Dilemma and Adaptive Regulation question and get a clear, step-by-step explanation grounded in how 18389 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.