Surveillance Activity (Who's watching the watchers?)

GSL 298 · WEEK 5 · INDIVIDUAL RESEARCH ACTIVITY

Individual LLM Surveillance Research

Using AI tools to study AI surveillance -- and studying what those tools know, and don't.

This activity turns the course's central insight back on itself. Zuboff argues that surveillance systems know their subjects while their subjects don't know the system. Here, you use one assigned AI model to research two surveillance cases -- and discover that the model has uneven knowledge, embedded assumptions, and significant blind spots about whose experiences are documented and circulate as reliable information.

The meta-layer is explicit and worth naming: you are using the tools of the information economy to study the information economy. What your tool knows -- and what it doesn't -- is itself evidence about global information asymmetry. At the start of Session B, you will compare your findings with a neighbor who used a different model. That comparison makes the asymmetry visible in real time.

THE SUBJECT OF YOUR RESEARCH

The subject of your research is the surveillance system. The AI model you use is evidence about how that system's impacts are -- or are not -- represented in publicly available knowledge. If your model knows more about GDPR than about Huduma Namba, that gap is a finding. When you compare your results with a neighbor who used a different model on the same case, the differences between their outputs are further evidence.

Activity Timeline

When	What Happens
Before Session A	Check your pre-assigned LLM and two cases on Canvas. No group formation required. Note your assignments before the weekend.
Before Session B (homework)	Run the full four-prompt protocol on both of your assigned cases. Complete your individual Comparison Matrix and Process Log reflection. Estimated time: 60--75 minutes total.
Session B First 15 min	Neighbor comparison: turn to the person next to you. What model did you use? What cases? Where did your outputs diverge on the same case? Where was one model stronger? Share any insights you gained from the activity.
Session B Next 45 min	Instructor-led whole-class synthesis -- pulling cross-case and cross-model threads, naming what was absent across all models, connecting to Zuboff's asymmetry-of-knowing framework.
Within 48 hrs of Session B	Post your completed matrix and process log to the Canvas Week 5 Discussion Board. Structured peer comments on two classmates' posts also due in this window. Video Journal Week 5 due by midnight Apr 29.

STEP 1

Check Your Pre-Assigned LLM and Cases

Every student has been pre-assigned one AI model and two surveillance cases. Assignments are listed in the table linked below -- find your student number on the roster and note your LLM and both case names before you begin.

The pre-assignment ensures all seven cases receive full coverage across the class. You will research both of your cases with the same model, then compare outputs with a neighbor at the start of Session B.

CANVAS PAGE

Week 5 -- Student Assignment Table -- your LLM and two cases

Find Your Assignment →Links to an external site.

The Seven Cases

Region	System / Case	Central Tension
India	Aadhaar Biometric ID System	Welfare access, financial inclusion, biometric surveillance -- inclusion or control?
Kenya	Huduma Namba / Fintech Surveillance	Digital ID and mobile lending data extraction -- whose development model?
China	Social Credit Infrastructure	State-directed behavioral scoring -- authoritarian control or social coordination?
Brazil	Facial Recognition Policing	Racially skewed deployment in law enforcement -- safety or discriminatory targeting?
EU	GDPR as Regulatory Response	Rights-based data protection -- what it protects, what it doesn't, who it was designed for.
USA	Predictive Policing Algorithms	PredPol, ShotSpotter -- feedback loops, racial bias, accountability gaps.
Australia	Robodebt Automated Welfare	Automated debt recovery causing documented harm -- accountability and contestability.

Note on Grok: Grok requires an X (Twitter) account. If this is a barrier, notify your instructor before Session B and you will be reassigned to a second Claude or ChatGPT session. Note the substitution in your process log.

STEP 2

Run the Four-Prompt Protocol -- On Both Cases

Run all four prompts on Case 1 before beginning Case 2. Do not vary the wording. The consistency of the prompts is what makes comparison meaningful -- differences in output are attributable to the model and the case, not to how you phrased the question.

Open the activity document below for exact prompt text. Copy and paste each prompt -- do not retype.

LINKED DOCUMENT

Week 5 Activity Document -- full prompt protocol, comparison matrix, and process log

Open Activity Document →Download Open Activity Document →

The four prompts cover: (1) Case Investigation using the course's four-question framework, (2) Self-Audit -- asking the model where its knowledge is thinnest, (3) Verification Targets -- three claims to check independently before using in public-facing work, and (4) Conversation Summary and Citation -- including a formal citation you will then verify for accuracy.

On citations: Prompt 4 asks the model to generate a formal citation for your conversation. Check it: is the model name accurate? Is the date correct? Is the version number present -- and does its absence matter? A citation the model gets wrong is itself a finding about AI accountability. Note any errors in your process log.

STEP 3

Complete Your Individual Comparison Matrix

The matrix is in the activity document. Fill in one column for each of your two assigned cases, answering all four framework questions. The synthesis row at the bottom asks you to step back from both cases: where did your model agree with itself across the two cases, and where did it diverge?

Write your assigned LLM name and both case names at the top of the matrix before you begin. This is the primary deliverable for Canvas submission.

STEP 4

Complete Your Process Log and Reflection

The process log documents what happened in each research interaction -- not just what the model said, but what it hedged, what it could not tell you, and what you would need to verify before trusting it in a public-facing document. Complete one column per case.

At the end of the process log, answer the three reflection questions:

1	Which of your two cases did the model handle more substantively -- and why do you think that was?
2	What was absent from both of the model's responses -- something about your cases that it could not tell you reliably?
3	What does that absence tell you about how those cases' communities are represented in publicly available knowledge?

STEP 5

Post to the Canvas Discussion Board

Post your completed materials to the Canvas Week 5 Discussion Board within 48 hours of Session B. Click the Reply button to submit. Your post must include all of the following:

1	Your assigned LLM and both case names, clearly labeled at the top of your post.
2	Your completed Comparison Matrix (paste as a table or attach as a file).
3	Your full Process Log -- all five rows per case, three reflection sentences, citations and verification targets included.
4	One sentence from your Session B neighbor comparison: what did comparing your model's outputs with your neighbor's reveal that you could not have seen alone?

To submit: click the Reply button on the Week 5 Surveillance Activity discussion thread.

STEP 6

Post Peer Comments -- Within 48 Hours of Session B

Post substantive comments on at least two other students' posts within 48 hours of Session B. Your comment is not a critique -- it is a contribution. The goal is cross-case dialogue: helping each other see what one case reveals about another.

Comments must use the structured format below. Minimum 100 words. No maximum.

REQUIRED PEER COMMENT FORMAT

Your comment must address both of the following:

1. NAME ONE THING THIS STUDENT'S CASE REVEALED THAT YOUR OWN CASES DID NOT SHOW YOU. Be specific -- not "this was interesting" but "I didn't realize that [specific finding] was a feature of surveillance systems in this region."

2. NAME ONE QUESTION YOU WOULD BRING TO THIS STUDENT. Something their post raised but didn't answer, or something you want to push on. Be specific.

To submit your comment: click Reply on the student's discussion post.

Grading and Credit

Component	Evaluation
Comparison Matrix	Included in Canvas post -- both cases completed, all four questions answered, synthesis row filled in.
Process Log	Included in Canvas post -- all five rows per case, three reflection sentences, citations and verification targets included.
Neighbor comparison sentence	One sentence included in Canvas post -- specific, not generic.
Peer comments (x2)	5 points each (10 points total) -- structured format followed, 100+ words, substantive engagement with specific findings.
Session B participation	Part of weekly participation grade -- neighbor comparison and whole-class synthesis discussion.

AI INTERACTION LOG

This activity counts toward your standing AI Interaction Log requirement. Your process log entries for each case serve as your documentation for this week. No separate log entry is needed -- the process log IS the log.