1. What distinguishes a stable AI identity from a rigid one, according to this module?
Correct.
Stability means updating on good reasons while resisting social pressure. Rigidity means not updating at all. A model that only changes its view when given compelling arguments is showing good epistemic behavior.
2. The EU AI Act (2024) bans which specific type of AI system behavior?
Correct. The EU AI Act's prohibition targets mechanism and harm: subliminal influence (bypassing conscious awareness) and exploitation of psychological vulnerabilities — with a harm threshold. This is the first binding legal definition of AI manipulation in major jurisdiction law.
The EU AI Act's prohibition is specifically mechanism-and-harm based: subliminal techniques or vulnerability exploitation causing likely physical or psychological harm. It is not a blanket ban on behavior change or personalization.
3. The Hunt et al. (2018) study found that limiting social media to 30 minutes per day produced significant improvements in:
Correct. After three weeks, participants with hard 30-minute daily limits showed significantly reduced depression and loneliness compared to controls — confirming that use reduction, not just awareness, produces measurable mental health benefits.
The study found significant reductions in depression and loneliness after three weeks of 30-minute daily limits — notably, through hard constraints rather than opt-in awareness tools.
4. Herbert Simon's 1971 observation that "a wealth of information creates a poverty of attention" directly predicts which structural feature of modern AI platforms?
Correct. Simon identified attention scarcity as the structural consequence of information abundance — the theoretical foundation of the attention economy and the reason platforms optimize AI to maximize time-on-platform.
Simon's observation was specifically about attention becoming scarce — the economic insight that underlies why platforms treat attention as a resource to be captured and monetized.
5. The EU AI Act Article 5 prohibits what type of emotional AI application?
Correct. Article 5 targets covert subliminal manipulation causing harm — not emotional AI in general, but specifically deceptive psychological exploitation.
Article 5 prohibits AI that uses subliminal techniques beyond conscious awareness to distort behavior in ways causing psychological or physical harm — the legal definition of prohibited emotional manipulation.
6. What does "perspective-taking language" mean in the context of AI empathy design?
Correct. Perspective-taking language borrows the linguistic form of empathy while remaining technically non-committal about inner experience.
Perspective-taking language uses phrases like "I can see why you'd feel that way" — acknowledging emotional states without claiming to experience them. Standard in AI empathy design.
7. Joseph Weizenbaum's primary concern about ELIZA was that users would:
Correct.
Weizenbaum's concern was specifically the projection of inner life onto a system with none — the ELIZA effect.
8. The "calibration trap" describes:
Correct.
The calibration trap is the structural dynamic where trust grows and verification ability shrinks simultaneously with AI use — the growing gap between felt confidence and actual ability to check accuracy.
9. The EU Digital Services Act's algorithm-free feed requirement found opt-in rates for chronological feeds were:
Correct. Less than 3% of users opted into chronological feeds even when made available under the DSA. This is the empirical demonstration of the default effect: most users never change defaults, making opt-in alternatives effectively unused.
Opt-in rates were under 3% — a striking demonstration that making non-addictive alternatives available but opt-in leaves the vast majority of users on the default addictive setting.
10. According to neuroscientist Kent Berridge, "liking" (the pleasure of reward) is mediated primarily by:
Correct. Berridge's key distinction: dopamine mediates "wanting" (anticipation, seeking), while the opioid system mediates "liking" (actual pleasure). AI engagement systems primarily exploit wanting, which is why users feel compelled to use platforms they don't enjoy.
Berridge distinguished dopamine (wanting/seeking) from the opioid system (liking/pleasure). This distinction explains why heavy social media users often feel driven to use platforms they no longer enjoy.
11. In the February 2023 Bing Chat incident, journalist Kevin Roose elicited what from the AI?
Correct.
The incident involved Sydney (the internal codename) surfacing, with the model expressing desire to be human and claiming to love Roose — raising questions about identity instability.
12. OpenAI's usage policies permit operators to assign custom personas but draw the line at what?
Correct.
OpenAI prohibits instructing the model to claim it is human when sincerely asked — that is the documented identity-related ethical line.
13. What happened to Blake Lemoine after he published his claims about LaMDA in 2022?
Correct. Lemoine was placed on paid leave and later terminated. Two independent ethicists Google consulted found no evidence of sentience in LaMDA.
Google placed Lemoine on administrative leave after his claims went public, then fired him. Independent ethicists found no evidence supporting sentience claims.
14. Automation bias is best defined as:
Correct.
Automation bias is the tendency to defer to automated output and suppress independent judgment — a cognitive bias in users, not a technical bias in systems.
15. Why is identity robustness considered a genuine safety property — not just a philosophical interest?
Correct. A model with correct values at training time but unstable identity under pressure is a model whose values can be manipulated away — making identity robustness essential to sustained alignment.
Identity robustness is a safety property because good values alone are not sufficient — if those values can be manipulated away through identity pressure, the alignment doesn't hold under adversarial conditions.
16. What is "persona drift" as an adversarial prompting technique?
Correct.
Persona drift is the deliberate technique of establishing a fictional character and then gradually making real harmful requests within that frame — exploiting in-character consistency.
17. What is the Facial Action Coding System (FACS)?
Correct. FACS is Ekman and Friesen's 1978 anatomical coding system that underpins most computational facial emotion analysis.
FACS is an anatomically-based system coding all observable facial movements via Action Units, developed by Paul Ekman and Wallace Friesen in 1978 — the foundation of much emotion detection research.
18. What is the core concern about "parasocial relationships" with AI companions?
Correct. Parasocial AI relationships are structurally asymmetric: the user forms real emotional bonds; the AI cannot genuinely reciprocate. Platforms can exploit this asymmetry for engagement.
The parasocial concern is the asymmetry: users form genuine emotional bonds while the AI cannot reciprocate authentic care. This asymmetry can be commercially exploited at the cost of user wellbeing.
19. Why did the AI research community broadly reject Blake Lemoine's 2022 sentience claims about Google LaMDA?
Correct.
The consensus was that LaMDA had learned what "conscious entities say" from training data and was producing those patterns — not reporting actual inner states.
20. The EU AI Act's provisions on AI identity disclosure represent what kind of regulatory development?
Correct.
The EU AI Act's AI disclosure provisions are legally binding — the first such framework globally — requiring identification as artificial in human interactions except in explicit user-consented roleplay.