Cambridge Analytica, Hedegreen Research, and the thin line between changing minds and steering people
I am building my first data-harvesting tool.
Not for elections. Not for ads. Not for voter profiling.
For burial mounds.
The idea is simple: people find a burial mound, photograph it from different angles, and help build a visual dataset that does not exist in the same ordinary public form.
Denmark already has official heritage data. Fund og Fortidsminder is the central register of ancient monuments and archaeological sites in Denmark; it is used both for physical planning and as a research resource, and the system also gives public access through WMS and WFS services, with metadata downloadable in formats such as MapInfo, ESRI shapefile and CSV. (slks.dk)
So the official map exists.
The coordinates exist. The administrative layer exists. The registry exists.
But a coordinate is not a field image.
A registry entry does not show how a mound looks from the edge of a winter field. A map layer does not show how vegetation hides a shape. A polygon does not show what a walker, a child, a local-history nerd, a farmer or a tourist actually sees.
That is the purpose of the tool.
It sounds innocent.
It probably is.
But it is not automatically innocent.
Because the moment a tool asks people to collect the world, it may also begin collecting the people.
1. The warning label
Cambridge Analytica died in public.
That is the easy version of the story.
The name became toxic. The scandal became shorthand. The company became a symbol of Facebook data, political targeting, psychographic claims, voter manipulation, platform negligence and the moment ordinary people realised that a quiz, a social graph and an election campaign could belong to the same machine.
The Federal Trade Commission later alleged that Cambridge Analytica, Alexander Nix and Aleksandr Kogan used deceptive tactics to harvest personal information from tens of millions of Facebook users for voter profiling and targeting; the FTC also stated that Cambridge Analytica had filed for bankruptcy and that Nix and Kogan agreed to restrictions and deletion or destruction requirements. (ftc.gov)
The UK Information Commissioner's Office described the broader issue in a way that matters even more: detailed pictures of online lives were being used by political parties to target small groups of voters with specific adverts. (ico.org.uk)
That is the real warning.
Not only that one company behaved badly.
The warning is that analysis can split in two.
One path makes systems readable. The other makes people exploitable.
2. The uncomfortable admission
Hedegreen Research is not neutral.
It should not pretend to be.
I do not write analysis because I have no direction. I write because I think some systems need to be seen differently. I want to change minds. I want people to notice patterns they missed. I want institutions, companies and public systems to become more readable.
That is influence.
There is no point lying about that.
Every serious essay, investigation, map, report, lecture, documentary, policy brief or public argument tries to change something in the reader. The ethical problem is not persuasion.
The ethical problem is hidden asymmetry.
There is a clean form of influence:
I want to change what you think by changing what you can see.
And there is a dirty form:
I want to change what you do by exploiting what you cannot see.
That is the line.
Cambridge Analytica is not only a scandal from the past. It is a warning label for anyone building analytical power.
Including me.
3. The fork in the road
There are two possible futures for Hedegreen Research.
Not because HR is dangerous now.
Because tools become dangerous when they work.
A small analysis site becomes a method. A method becomes a system. A system becomes a product. A product collects users. Users create data. Data creates leverage. Leverage creates temptation.
The danger does not begin with evil.
It begins with effectiveness.
The good path is analysis as public light.
The dark path is analysis as behavioural control.
- Good HR: Makes systems readable.
Dark HR: Makes people exploitable. - Good HR: Shows sources.
Dark HR: Hides targeting logic. - Good HR: Shows uncertainty.
Dark HR: Optimizes persuasion. - Good HR: Invites correction.
Dark HR: Avoids contestation. - Good HR: Uses public reasoning.
Dark HR: Uses private profiling. - Good HR: Changes minds through evidence.
Dark HR: Moves behaviour through asymmetry. - Good HR: Makes power readable to people.
Dark HR: Makes people readable to power.
That last line is the whole problem.
Cambridge Analytica tried to make people readable to power.
HR should make power readable to people.
4. The burial mound test
The burial mound app is a small test.
But it is a real one.
There are two possible subjects in the tool.
The first subject is the mound.
Its shape. Its visibility. Its condition. Its surroundings. Its relationship to roads, fields, forests, coastlines, paths and settlements. Its visual life across seasons and angles.
That is the good dataset.
The second possible subject is the contributor.
Where they walk. When they walk. How often they contribute. Which places they visit. How far they travel. How their behaviour changes when the app suggests a new target. Who becomes valuable because they produce rare field data.
That is the dangerous dataset.
The tool may be built to document heritage. But if it quietly begins to model the contributors, the subject has shifted.
The mound is no longer the only object being read.
So the rule has to be simple:
The mound is the subject. The user is not the product.
That rule must be designed in before the first upload button.
Not after growth. Not after launch. Not after a privacy complaint. Not after a journalist asks the obvious question.
Before.
There is also a second responsibility: the mound itself.
The 100-metre protection line around protected ancient monuments exists to protect their value as landscape elements and to take special account of archaeological layers nearby. (slks.dk)
So the app has two obligations.
Protect the contributor.
Protect the mound.
5. The gallery of afterlives
Cambridge Analytica did not have one afterlife.
It scattered.
That is the structural point.
Some of it became regulatory consequence. Some became corporate residue. Some became campaign analytics. Some became data-rights language. Some became AI governance language.
None of these afterlives are identical.
That is precisely why they matter.
The Face: Alexander Nix
Alexander Nix is where the scandal becomes paperwork.
Companies House lists Nix as disqualified from acting as a company director from 5 October 2020 to 4 October 2026, with the disqualification tied to conduct while acting for SCL Elections Limited. (find-and-update.company-information.service.gov.uk)
Public scandal is loud.
Formal restriction is quiet.
The public remembers the face.
The register remembers the consequence.
The Bridge: Aleksandr Kogan
Aleksandr Kogan is the bridge between research culture, app-mediated collection, platform data and political targeting.
The FTC described Kogan as the developer of the GSRApp, also known as "thisisyourdigitallife," and alleged that the app collected Facebook data from users and their friends. The FTC said that information was used to train an algorithm that generated personality scores, which Cambridge Analytica, Kogan and Nix then matched with U.S. voter records for voter profiling and targeted advertising services. (ftc.gov)
The University of Cambridge also described Kogan as joining its Department of Psychology as a lecturer in 2012, establishing the Cambridge Prosociality and Well-Being Lab, and later creating Global Science Research as a commercial enterprise. (cam.ac.uk)
The lesson is not that research is bad.
The lesson is that research methods can be absorbed by institutions with very different incentives.
That is why Kogan belongs in this article.
He is not here only as a Cambridge Analytica figure. He is here because the same structural risk exists at a smaller scale in any tool that collects data through public participation.
A method built to understand personality became part of a system for political targeting.
A method built to document burial mounds could, if badly designed, become a system for profiling contributors.
The parallel is not moral equivalence.
The parallel is structural drift.
The original purpose of a dataset does not permanently control its future use. The design does.
The Continuity Node: Matt Oczkowski and Data Propria
Matt Oczkowski is the cleanest afterlife question.
The Associated Press reported in 2018 that Data Propria was run by former Cambridge Analytica officials, that at least four former Cambridge Analytica employees were affiliated with it, and that Oczkowski led the firm; AP described Data Propria as specializing in voter and consumer targeting work similar to Cambridge Analytica's work before its collapse. (apnews.com)
This is the sharpest case because it does not require drama.
The question is obvious:
When the company dies, does the method simply re-enter the market?
Not necessarily the same illegal data. Not necessarily the same brand. Not necessarily the same wrongdoing.
But the same market hunger remains:
Behavioural data. Voter targeting. Consumer targeting. Persuasion analytics.
That is the afterlife problem.
The Inversion: Brittany Kaiser
Brittany Kaiser is where the language flips.
Reuters described Kaiser as widely known as a Cambridge Analytica whistleblower and reported that, after leaving Cambridge Analytica, she worked on privacy, data protection and encryption-law issues. Axios later described her as co-founder and chair of the Open Source AI Foundation, while her Own Your Data Foundation profile places her in data rights and digital-literacy work.
This is not a simple story of guilt or redemption.
It is more interesting than that.
The same historical wound now produces the vocabulary of reform:
Data rights. Transparency. Open systems. Privacy. Auditability. Governance.
That may be sincere.
It may also be useful.
Both can be true.
The Corporate Residue: Emerdata
Emerdata is where the company register remembers what public memory forgets.
Companies House lists Emerdata Limited as incorporated on 11 August 2017, currently in administration, with its business activity classified as "data processing, hosting and related activities." (find-and-update.company-information.service.gov.uk)
Among Emerdata's current active directors, Companies House lists Jennifer Mercer and Rebekah Mercer, both appointed on 16 March 2018; it also lists Alexander Nix as a resigned director appointed on 23 January 2018 and resigned on 28 March 2018. (find-and-update.company-information.service.gov.uk)
This does not prove a conspiracy.
It proves something colder and more useful:
The public remembers the scandal as collapse.
The register remembers the transition.
6. The structural point
The weak version of this article would say:
Look, the bad people are still around.
That is too easy.
And probably unfair.
The stronger version says:
Analytical power rarely disappears. It changes vocabulary.
That is the point.
Not that every person remains guilty of the same thing. Not that every new role is corrupt. Not that every data project is Cambridge Analytica reborn.
The point is that methods survive reputational death.
They survive as consulting. They survive as campaign infrastructure. They survive as governance language. They survive as ethical AI. They survive as corporate filings. They survive as lessons learned.
Sometimes that survival is useful.
Sometimes it is dangerous.
Usually, it is both.
7. The HR constraint
An oath is not enough.
Good intentions are not enough.
Most dangerous systems do not begin with a villain speech.
They begin with a useful tool, a real problem, a clever method, a small dataset, a growing audience and a reason to collect just a little more.
That is why the line cannot live only in an essay.
It has to live in the design.
For Hedegreen Research, the burial mound app cannot simply promise to respect users. It has to be built so that respect is the default.
The mound must be the subject.
The contributor must not become the product.
That means minimal collection. Visible consent. Separation between heritage data and contributor data. No hidden route profiling. No unnecessary device metadata. No behavioural scoring disguised as engagement. No later model-training without saying so. No "we may use your data to improve services" fog.
If the tool needs data, it should say what data.
If it stores data, it should say why.
If it benefits from contributors, it should say how.
If it changes purpose, it should ask again.
That is the difference between an ethical statement and an ethical system.
Cambridge Analytica is useful here because it is not only a scandal.
It is a stress test.
It shows what analysis can become when the people being read cannot see the reading, cannot contest the model and cannot know who benefits from the output.
Hedegreen Research has to choose the opposite direction.
Not because HR is pure.
Purity is not a method.
The method is this:
Make systems readable. Make sources visible. Make uncertainty explicit. Make correction possible. Make extraction difficult. Make refusal real.
Analysis may change minds.
It should.
But it should change minds by increasing what people can see, not by exploiting what they cannot see.
Cambridge Analytica tried to make people readable to power.
Hedegreen Research should make power readable to people.
That is not an oath.
That is a design constraint.
Note
This article does not claim that every post-Cambridge Analytica role represents wrongdoing. It examines how analytical capacity, methods, language and corporate structures can survive reputational collapse.
— Dennis Hedegreen, trying to see the structure
