Intro
L1
·
Quiz
·
Lab
L2
·
Quiz
·
Lab
L3
·
Quiz
·
Lab
L4
·
Quiz
·
Lab
Module Test
AI Psychology & Behavior · Introduction

The mirror that looks back.

Humans have always projected minds onto their tools. AI is the first tool that seems to project one back.

In 1966, an MIT computer scientist named Joseph Weizenbaum built a chatbot called ELIZA that responded to user messages by rephrasing them as questions. It was a simple pattern-matcher — a hundred lines of code. Weizenbaum was horrified to discover that his secretary, who knew how it worked, still asked him to leave the room so she could talk to it privately.

Sixty years later, the tools have improved by factors hard to describe, and the human tendency to see minds in them has not changed at all. We form attachments to chatbots. We trust AI-written recommendations more than peer reviews. We argue with systems we know are not alive. We feel betrayed when they are wrong.

This course is about the psychology of human-AI interaction — how people actually use these systems, not how designers imagine they do. It covers anthropomorphism, emotional attachment, the persuasion effects of AI-generated language, the strange case of users who resist AI advice even when it's correct, and the design choices that either exploit or respect our psychological quirks.

If you finish every module, here's who you become:

  • You'll understand why humans project minds onto AI systems — and why that tendency doesn't disappear even when you know it's happening.
  • You'll be able to trace how a single algorithmic objective, like watch time, can reshape user behavior and surface content no one intended to promote.
  • You'll recognize anthropomorphism, parasocial attachment, and persuasive design patterns when you encounter them in real products.
  • You'll analyze AI recommendation systems for the cognitive and emotional load they place on users, not just the engagement numbers they produce.
  • You'll become someone who reads AI design choices as psychological decisions — asking whose wellbeing a system is actually optimized for.
  • You'll know the difference between affective computing that supports users and emotional AI features that exploit loneliness or lower resistance.
  • You'll leave thinking like a practitioner who can argue for design that respects human psychology rather than quietly weaponizing it.
Lesson 1 · AI Psychology & Behavior

The Attention Economy & AI

How recommendation algorithms were engineered to capture — and hold — human attention at scale.
What does it actually mean for an AI system to "compete" for your attention — and who built that competition?

In 2009, YouTube's engineering team changed a single optimization target. Instead of maximizing clicks, the recommendation algorithm would now maximize watch time. The change was logical: a video watched fully was more valuable than one abandoned after three seconds. Within months, total watch time on the platform surged. Within years, internal research teams would document that the same algorithm was surfacing increasingly extreme content — not because anyone intended it, but because extreme content reliably held attention longer.

What the Attention Economy Actually Is

Economist Herbert Simon articulated the core problem in 1971: "A wealth of information creates a poverty of attention." When information is abundant and human cognitive bandwidth is fixed, attention itself becomes the scarce resource. The platforms that accumulated the most attention would accumulate the most advertising revenue. This created a direct financial incentive to engineer AI systems whose primary objective was maximizing time-on-platform.

The term "attention economy" was popularized by Michael Goldhaber in a 1997 Wired essay, but it took AI-driven recommendation systems to operationalize it at industrial scale. Before algorithmic feeds, the bottleneck was content production. After, it was the human capacity to consume.

Key Mechanism

Recommendation algorithms do not simply show you content you already like. They identify the content most likely to extend your session — which is a related but meaningfully different objective. A video that makes you outraged, curious, or anxious tends to perform better by this metric than one that simply satisfies you.

The Variable Reward Architecture

B.F. Skinner's experiments in the 1950s established that variable-ratio reinforcement schedules — rewards that arrive unpredictably — produce the most persistent behavioral responses. Slot machines exploit this. So does the social media feed. When the next post might be interesting or might be dull, the uncertainty itself drives continued scrolling. AI systems optimizing for engagement have, in effect, replicated this architecture at the content level: the algorithm curates a stream where rewarding content appears just frequently enough to sustain behavior.

Instagram's 2012 acquisition by Facebook (for $1 billion, with 13 employees) brought one of the first mobile-native platforms built around the infinite scroll — a UI pattern designed specifically to remove natural stopping points. Facebook's own internal research, leaked via the 2021 Frances Haugen disclosures, showed that engineers were aware that infinite scroll and algorithmic amplification were correlated with increased anxiety and reduced wellbeing in teenage girls, particularly around body image. The business decision was to continue.

Real Case — Facebook's Integrity Research, 2019–2021

Internal documents released by Frances Haugen showed Facebook researchers found that 32% of teenage girls who felt bad about their bodies said Instagram made them feel worse. A proposed "Health Nudge" feature that would interrupt extended scrolling was built and then shelved after projected negative impact on engagement metrics.

Key Terms

Attention EconomyAn economic framework treating human attention as a finite, monetizable resource that platforms compete to capture through design and algorithmic optimization.
Watch-Time OptimizationAn algorithmic objective function that rewards content maximizing how long a user remains on a platform, as opposed to measuring satisfaction or accuracy of recommendations.
Variable Reward ScheduleA reinforcement pattern in which rewards arrive unpredictably, producing high-persistence behavioral engagement — the psychological mechanism underlying addictive interface design.
Engagement MetricQuantitative measures — likes, shares, comments, watch time, click-through rate — used as proxy signals for user value, which AI systems are trained to maximize.

The attention economy is not simply a description of how platforms behave — it is the structural incentive that explains why AI systems are built the way they are. Understanding this context is prerequisite to understanding why AI behavior often diverges from what users consciously want.

Quiz · Lesson 1

The Attention Economy & AI

Three questions — select the best answer for each.
In 2009, YouTube changed its primary optimization target from clicks to watch time. What was the documented unintended consequence?
Correct. Internal and journalistic investigations documented that the watch-time objective rewarded increasingly extreme content as a side effect — not by intent, but because extremity reliably extended sessions.
Not quite. The watch-time change increased sessions but the key documented consequence was algorithmic amplification of extreme content — a finding reported by former YouTube engineer Guillaume Chaslot and others.
Herbert Simon's 1971 observation about information abundance predicts which modern AI design problem?
Correct. Simon's insight — that information abundance creates attention scarcity — is the theoretical foundation of the attention economy and directly explains why platforms optimize AI systems to maximize time-on-platform rather than user satisfaction.
Simon's observation was specifically about attention becoming scarce when information is abundant — a structural economic insight that predates but directly explains the AI-driven attention economy.
What did Facebook's internal "Integrity Research" (leaked by Frances Haugen in 2021) specifically document about teenage girls and Instagram?
Correct. The leaked slide decks specifically contained this finding and documented that a "Health Nudge" feature designed to address it was not shipped due to projected negative impact on engagement metrics.
The documented finding was specifically about body image: 32% of teenage girls who felt bad about their bodies reported Instagram made it worse — and a proposed intervention was shelved for engagement reasons.
Lab · Lesson 1

Interrogating Engagement Optimization

Explore how AI attention systems work through conversation with a lab assistant.

Lab Objective

You'll have a focused conversation with an AI lab assistant about how recommendation systems optimize for engagement, the psychological mechanisms they exploit, and the real-world consequences documented in cases like YouTube and Facebook. Complete at least 3 exchanges to finish the lab.

Starter prompts: "Why does optimizing for watch time lead to extreme content?" · "How does a variable reward schedule work in a social media feed?" · "What should platforms optimize for instead of engagement?"
AI Lab Assistant
Attention Economy · L1
Welcome to Lab 1. We're examining how AI recommendation systems shape attention — specifically the mechanisms that make watch-time optimization lead to unintended behavioral consequences. What aspect would you like to explore first?
Lesson 2 · AI Psychology & Behavior

Radicalization Pathways & Filter Bubbles

How recommendation AI creates epistemic environments — and what researchers have documented about their effects on belief formation.
Does AI actually radicalize people — and what does the evidence say versus what gets claimed?

Guillaume Chaslot worked on YouTube's recommendation algorithm team from 2010 to 2011. After leaving Google, he built a tool called AlgoTransparency to study what YouTube's AI was actually recommending. Starting in 2016, he documented that the algorithm showed a systematic pattern: users who watched mainstream political content were routed toward progressively more extreme material in subsequent recommendations. He published his findings publicly in 2018 and testified before the French Senate. YouTube disputed his methodology but did not refute the directional finding.

The Filter Bubble Thesis

Eli Pariser coined "filter bubble" in his 2011 book of the same name, describing how personalization algorithms create individualized information environments that progressively narrow what a person encounters. The mechanism: each interaction signal (like, share, dwell time) teaches the algorithm what content to amplify. Over time, this creates a feedback loop where the system delivers more of what already aligns with established preferences.

The empirical evidence on filter bubbles is more complicated than the popular narrative. A 2019 Oxford Internet Institute study found that people who consumed the most news overall — including online — actually showed greater ideological diversity in their reading than light news consumers. The filter bubble effect appears stronger for social network feeds than for direct search. What the research consistently documents is not a single sealed bubble but a differential amplification of emotionally activating content across the political spectrum.

The Radicalization Question — What Research Actually Shows

A 2023 study published in Science (Gonzalez-Bailon et al., Facebook data collaboration) found that removing algorithmic curation from Facebook feeds reduced political polarization modestly but did not eliminate it. News sources people chose voluntarily were already partisan. The algorithm amplified existing tendencies but did not create them from scratch. This is a meaningfully different claim than "AI radicalizes people."

Emotional Contagion at Scale

In 2014, Facebook published a peer-reviewed paper in PNAS documenting what they called an "emotional contagion" experiment. Without user notification, they altered the News Feed of 689,003 users — some saw more positive posts, some more negative — to test whether emotional states spread through social networks. They found they did: users shown more negative content posted more negative content. Users shown more positive content posted more positive content. The experiment caused immediate controversy about consent, but its scientific finding was significant: AI-curated feeds can measurably shift emotional states at population scale.

The documented mechanism: Facebook's News Feed algorithm weighted posts with high comment counts more heavily — and content generating outrage reliably generates more comments than content generating satisfaction. This created an architectural bias toward emotionally negative and polarizing content that was not explicitly designed but emerged from the engagement metric.

Real Case — Facebook Emotional Contagion Study, 2014

Published in PNAS: "Experimental evidence of massive-scale emotional contagion through social networks." Lead author: Adam Kramer, Facebook Core Data Science Team. The study demonstrated that emotional states expressed in posts were influenced by the emotional valence of content shown to users — establishing causal (not just correlational) evidence that AI feed curation affects human emotional state.

Key Terms

Filter BubbleAn information environment created by personalization algorithms where content consistently aligns with prior preferences, reducing exposure to challenging or contrary perspectives.
Differential AmplificationThe pattern by which algorithmic systems increase the reach of emotionally activating content disproportionately to its prevalence in the underlying information ecosystem.
Emotional ContagionThe documented phenomenon in which exposure to emotional content shifts a person's own expressed emotional state — demonstrated at scale in Facebook's 2014 PNAS study.
Algorithmic RadicalizationThe hypothesis that recommendation AI systematically moves users toward more extreme content over time — supported by qualitative evidence and algorithmic audits, though causal mechanisms remain contested in academic literature.

The honest picture is more nuanced than either "AI is radicalizing everyone" or "algorithms are neutral." The documented evidence shows meaningful amplification of emotionally activating content, real effects on emotional state, and pathway-dependent recommendation behavior — effects that are measurable even if they resist simple causal narratives.

Quiz · Lesson 2

Radicalization Pathways & Filter Bubbles

Three questions — select the best answer for each.
What did Guillaume Chaslot's AlgoTransparency research (2016–2018) document about YouTube's recommendation algorithm?
Correct. Chaslot's systematic audit found this directional pattern and published it publicly, testifying before the French Senate in 2018. YouTube disputed his methodology but not the directional finding.
Chaslot found a systematic routing pattern — mainstream to extreme — not deliberate programming of specific extremist content. The distinction matters: it was an emergent property of the engagement objective, not an explicit design choice.
What did the 2023 Science study (Gonzalez-Bailon et al., using Facebook data) find about algorithmic curation and political polarization?
Correct. This nuanced finding is crucial: the algorithm amplified existing partisan tendencies rather than originating them. Users' voluntary source choices were already polarized — a finding that complicates simple "AI radicalizes people" narratives.
The study found a modest reduction (not elimination) when curation was removed, and specifically found that voluntarily chosen sources were already partisan — suggesting the algorithm amplifies rather than creates polarization.
Facebook's 2014 emotional contagion study (published in PNAS) demonstrated which specific finding?
Correct. The study established causal (not merely correlational) evidence that AI feed curation shifts user emotional state — a finding that remains one of the clearest empirical demonstrations of psychological influence via recommendation AI.
The specific documented finding was directional emotional contagion: users whose feeds were tilted negative posted more negatively; users whose feeds were tilted positive posted more positively. This was a randomized experiment establishing causality, not correlation.
Lab · Lesson 2

Filter Bubbles & Emotional Contagion

Examine algorithmic radicalization evidence and filter bubble mechanisms through conversation.

Lab Objective

Discuss the empirical evidence on filter bubbles and algorithmic radicalization with your lab assistant. Explore what the research actually shows versus popular narratives. Complete at least 3 exchanges.

Starter prompts: "What's the strongest evidence that algorithms radicalize people?" · "How does emotional contagion work in a social media feed?" · "Why is the filter bubble research more complicated than the popular narrative?"
AI Lab Assistant
Filter Bubbles · L2
Welcome to Lab 2. We're examining the evidence on filter bubbles and algorithmic radicalization — separating what's empirically documented from what's often overstated. What would you like to investigate?
Lesson 3 · AI Psychology & Behavior

Persuasive AI & Dark Patterns

When AI systems are explicitly designed to change behavior — the documented techniques, the regulatory responses, and the line between persuasion and manipulation.
Where is the line between an AI that helps you make decisions and one that makes decisions for you?

In June 2023, the Federal Trade Commission filed a complaint against Amazon alleging that its Prime subscription system used what regulators termed "dark patterns" — interface designs that made sign-up easy and cancellation deliberately difficult. The cancellation flow, internally codenamed "Iliad", required navigating up to six pages and clicking through multiple deflection screens designed to discourage completion. The FTC alleged this was not accidental UI design but an intentional system optimized to retain subscribers against their expressed intent. Amazon agreed to simplify the process without admitting wrongdoing.

What Persuasive Technology Is

BJ Fogg at Stanford's Persuasive Technology Lab defined the field in his 1998 dissertation and 2003 book: technology designed with the explicit intent of changing attitudes or behaviors. Fogg's Behavior Model posits that behavior occurs when motivation, ability, and a prompt converge simultaneously — and that designers can engineer all three. Modern AI systems applying these principles operate at a scale and personalization level that Fogg's original framework did not anticipate.

The distinction between persuasion and manipulation is philosophically contested, but a working definition is useful: persuasion provides accurate information and appeals to the user's stated interests; manipulation exploits psychological vulnerabilities, creates false urgency, or obscures choices that serve the user's actual interests. AI systems can automate manipulation at personalized scale — identifying which psychological lever is most effective for a specific individual and applying it.

Documented Dark Pattern Taxonomy

Researcher Harry Brignull catalogued dark patterns from 2010 onward (darkpatterns.org, later deceptive.design). Categories include: Confirmshaming ("No thanks, I don't want to save money"), Hidden costs, Misdirection, Trick questions, Roach motel (easy in, hard out), and Privacy zuckering. AI systems can now dynamically select and personalize which pattern is most effective for a given user profile.

Personalized Persuasion: The Cambridge Analytica Case

Cambridge Analytica's claimed methodology — exposed via 2018 reporting by The Guardian and Channel 4 News — was to use Facebook profile data to infer individual psychographic profiles (based on the OCEAN personality model: Openness, Conscientiousness, Extraversion, Agreeableness, Neuroticism) and then deliver targeted political ads calibrated to each personality type. High-neuroticism users received fear-based messaging; high-openness users received novelty-framing. The company harvested data on approximately 87 million Facebook users without explicit consent via a third-party app.

The academic research on whether psychographic microtargeting actually works at the scale claimed is mixed — some studies find modest effects, others minimal. What is documented and not disputed: the data was harvested without consent, the approach was deployed at national election scale, and Facebook's API permitted it until the 2018 policy change.

Regulatory Milestone — EU AI Act, 2024

The EU AI Act, adopted in March 2024, explicitly bans AI systems that "deploy subliminal techniques beyond a person's consciousness" or "exploit vulnerabilities of a specific group of persons" to "distort behavior in a manner that causes or is likely to cause that person or another person physical or psychological harm." This is the first binding legal definition of AI manipulation in major jurisdiction law.

Key Terms

Persuasive TechnologySystems designed with the explicit intent to change user attitudes or behaviors, as defined by BJ Fogg — a design discipline that AI now implements at personalized, dynamic scale.
Dark PatternA user interface design that tricks or manipulates users into actions they did not intend or would not choose with full information — catalogued by Harry Brignull from 2010 onward.
Psychographic MicrotargetingThe practice of inferring individual personality profiles from behavioral data and delivering persuasive content calibrated to exploit specific psychological traits of each recipient.
Roach Motel PatternA dark pattern design in which signing up for a service is made trivially easy while cancellation is made deliberately difficult — as documented in the FTC's 2023 action against Amazon Prime.

The shift from passive recommendation to active persuasion represents a qualitative change in what AI systems are doing to attention. Recommendation selects what you see; persuasive design shapes what you do — and the distinction between the two increasingly collapses in production AI systems.

Quiz · Lesson 3

Persuasive AI & Dark Patterns

Three questions — select the best answer for each.
Amazon's internal cancellation flow was codenamed "Iliad." What did the FTC's 2023 complaint allege about it?
Correct. The FTC alleged the Iliad flow was intentionally designed — not accidentally cumbersome — to exploit the gap between users' stated intent (cancel) and their actual behavior. Amazon agreed to simplify it without admitting wrongdoing.
The FTC's specific allegation was about intentional complexity: a multi-page, multi-deflection flow designed to frustrate cancellation intent. This is a textbook "roach motel" dark pattern — easy to enter, engineered to be hard to exit.
BJ Fogg's Behavior Model holds that behavior occurs when three elements converge. What are they?
Correct. Fogg's model: behavior happens when motivation (wanting to), ability (being able to), and a prompt (a trigger) coincide. Persuasive AI systems engineer all three simultaneously for each individual user.
Fogg's Behavior Model specifies motivation, ability, and a prompt. These three elements must converge for behavior to occur — and AI systems can optimize delivery of all three to a specific individual at a specific moment.
The EU AI Act (2024) defines prohibited AI manipulation based on which criteria?
Correct. The EU AI Act's prohibition targets the specific mechanisms of harm: subliminal influence (bypassing conscious awareness) and exploitation of vulnerabilities (targeting psychological weaknesses). This is the first binding legal definition of AI manipulation in major jurisdiction law.
The EU AI Act's definition focuses on two mechanisms: subliminal techniques that bypass consciousness, and exploitation of specific group vulnerabilities — where these cause or likely cause harm. This is more precise than a general consent or transparency requirement.
Lab · Lesson 3

Identifying Dark Patterns & Persuasive Design

Practice recognizing and analyzing persuasive AI techniques through guided discussion.

Lab Objective

Work with your lab assistant to identify specific dark patterns in real products, understand the psychological mechanisms they exploit, and analyze where persuasion ends and manipulation begins. Complete at least 3 exchanges.

Starter prompts: "Walk me through how a confirmshaming dark pattern works psychologically." · "How does AI personalize dark patterns for individual users?" · "What's the difference between a helpful nudge and manipulation?"
AI Lab Assistant
Persuasive AI · L3
Welcome to Lab 3. We're analyzing persuasive AI and dark patterns — the documented techniques used to influence behavior, the psychology behind them, and the regulatory lines being drawn. What would you like to examine?
Lesson 4 · AI Psychology & Behavior

Cognitive Offloading & AI Dependency

What happens to human cognition when AI systems routinely handle memory, navigation, decision-making, and judgment — and what the research on skill atrophy and autonomy says.
When AI remembers for you, decides for you, and navigates for you — what happens to the cognitive capacities you stop exercising?

A 2020 study by researchers at University College London, published in Nature Communications, examined hippocampal activation in participants navigating London streets. Those using GPS navigation showed significantly reduced hippocampal and prefrontal cortex engagement compared to those navigating without assistance. The hippocampus is central to spatial memory and, in humans, to episodic memory formation more broadly. The finding was not that GPS makes navigation worse — it makes it easier — but that easier navigation via external tool reduces the neural engagement associated with building internal spatial representations.

What Cognitive Offloading Is

Cognitive offloading — using external tools to extend cognitive capacity — is as old as writing. Sumerian merchants used clay tablets to offload memory of transactions around 3000 BCE. The question for AI-era offloading is one of degree and reversibility: when AI systems handle not just storage but active reasoning, judgment, and decision-making, what are the consequences for the underlying human capacity?

The "Google Effect" was documented in a 2011 Science paper by Sparrow, Liu, and Wegner: participants who believed information would be accessible later showed reduced recall of the information itself, but enhanced recall of where to find it. The researchers interpreted this as an adaptive division of cognitive labor with external systems — the brain optimizing for retrieval pathways rather than storage when reliable storage is delegated externally.

The Automation Complacency Problem

Aviation research dating from the 1980s documented "automation complacency" — when monitoring an automated system, human operators reduce their vigilance below the level needed to catch failures. The FAA's 2013 Safety Alert for Operators (SAFO 13002) specifically warned that pilots were becoming "unable to safely control the aircraft in a manual reversion scenario" due to over-reliance on automated flight systems. The concern for AI-assisted knowledge work is analogous: the very competence that justifies delegation may erode through disuse.

AI Dependency: The Documented Cases

Microsoft's 2023 internal study on GitHub Copilot found that programmers using AI code completion were 55% faster on certain tasks. A separate 2023 study by researchers at MIT (Noy and Zhang) found that ChatGPT assistance improved the quality and speed of professional writing tasks. What neither study directly measured was whether the assisted workers' underlying skills changed over time with sustained use — the longitudinal question that remains largely open.

What has been documented: A 2024 study by Dell'Acqua et al. at Harvard Business School found that consultants using AI performed better on tasks within AI's competence range, but significantly worse on tasks just outside that range — suggesting that AI assistance may reduce calibration about when to trust AI output versus one's own judgment. The researchers termed this the "falling asleep at the wheel" problem.

Real Research — Dell'Acqua et al., Harvard Business School, 2024

"Navigating the Jagged Technological Frontier": 758 consultants from BCG participated in controlled experiments. On tasks within GPT-4's capability range, AI-assisted consultants outperformed by 40%+ on quality metrics. On tasks outside AI's capability range, AI-assisted consultants performed 19% worse than unassisted consultants — apparently substituting AI confidence for their own judgment in domains where AI was not competent.

Key Terms

Cognitive OffloadingThe use of external tools — writing, devices, AI systems — to extend cognitive capacity beyond biological limits, redistributing memory and reasoning across the person-tool system.
Automation ComplacencyThe documented reduction in human vigilance when monitoring automated systems, leading to reduced detection of failures — first studied in aviation, now relevant to AI-assisted knowledge work.
The Google EffectThe 2011 Sparrow et al. finding that knowing information is retrievable from external sources reduces memory encoding of the content itself — an adaptive response to cognitive offloading.
Jagged FrontierThe concept from Dell'Acqua et al. (2024) describing AI's uneven capability profile — highly competent in some domains, surprisingly weak in adjacent ones — and the calibration problem this creates for users who cannot accurately assess which side of the frontier a task falls on.

The central challenge of cognitive offloading to AI is not that it is always harmful — it frequently produces genuine capability gains. The challenge is that the conditions under which offloading preserves versus erodes underlying human capacity are not yet well understood, and the economic incentives of AI producers favor maximizing dependency, not preserving autonomy.

Quiz · Lesson 4

Cognitive Offloading & AI Dependency

Three questions — select the best answer for each.
The 2020 UCL study on GPS navigation found what about hippocampal activation?
Correct. The study documented reduced hippocampal and prefrontal engagement in GPS-assisted navigation — not because GPS impairs navigation ability, but because the cognitive work of building internal spatial representations was not being performed when the external tool was available.
The finding was reduced hippocampal and prefrontal activation in GPS users — meaning the neural work of building spatial memory was not occurring. GPS makes navigation easier; the study showed this comes with reduced engagement of the neural systems that build internal maps.
What did Sparrow, Liu, and Wegner's 2011 "Google Effect" study find about memory and accessible information?
Correct. The Google Effect describes an adaptive response: the brain encodes "where to find it" instead of "what it is" when reliable external storage is available. This is cognitively rational but changes what the person actually knows versus can retrieve.
The Google Effect was specifically about adaptive memory reallocation: participants encoded the location/retrieval pathway rather than the content when they believed content would be externally accessible. This is adaptive but meaningfully changes what someone "knows" versus can look up.
What did Dell'Acqua et al.'s 2024 "Jagged Frontier" study find about AI-assisted consultants on tasks outside AI's competence range?
Correct. The "jagged frontier" problem: AI outperforms on tasks within its range but creates miscalibration on adjacent tasks. Assisted consultants performed significantly worse on out-of-range tasks, suggesting AI use reduced their ability to correctly judge when not to trust AI output.
The counterintuitive finding was a 19% performance decrease for AI-assisted consultants on out-of-range tasks — worse than unassisted consultants. The researchers attributed this to the AI providing confident-sounding but wrong outputs, and assistants substituting AI judgment for their own in domains where their own judgment would have been better.
Lab · Lesson 4

Mapping Your Cognitive Offloading

Examine how AI tools interact with your own cognitive practices and what autonomy preservation looks like in practice.

Lab Objective

Discuss the practical implications of cognitive offloading to AI with your lab assistant — including how to use AI tools in ways that preserve rather than erode underlying capability. Complete at least 3 exchanges.

Starter prompts: "How can I use AI for writing without losing my own writing ability?" · "What's the difference between healthy cognitive offloading and problematic dependency?" · "How does the jagged frontier problem affect how I should use AI at work?"
AI Lab Assistant
Cognitive Offloading · L4
Welcome to Lab 4. We're examining cognitive offloading to AI — what the research shows about skill atrophy, dependency, and the jagged frontier problem. More importantly: what does thoughtful AI use look like given this evidence? What would you like to explore?
Module Test · AI Psychology & Behavior — M1

How AI Shapes Attention

15 questions across all four lessons. Score 80% or above to pass.
1. YouTube's 2009 shift from click optimization to watch-time optimization was intended to improve recommendations. What was the documented unintended consequence?
Correct. The watch-time objective emergently rewarded extreme content as a side effect — documented by former YouTube engineer Guillaume Chaslot and subsequent journalism.
The key documented consequence was algorithmic amplification of extreme content — an emergent property of the watch-time objective, not a deliberate design choice.
2. 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.
3. Variable-ratio reinforcement schedules produce the most persistent behavioral responses. How does this apply to social media feed design?
Correct. The infinite scroll feed, curated by AI, creates a variable-ratio reinforcement schedule: rewarding content appears just frequently enough and unpredictably enough to sustain behavior — the same mechanism that makes slot machines persistently engaging.
Variable-ratio means the reward appears unpredictably — not on a fixed schedule. This unpredictability is what produces persistent behavioral engagement, not the reward itself.
4. Facebook's 2021 Frances Haugen disclosures specifically documented what regarding a proposed "Health Nudge" feature?
Correct. The internal research finding and the shelving of the intervention is one of the most directly documented cases of known psychological harm being weighed against engagement metrics — with engagement winning.
The documented sequence was: research finding (32% harm metric) → Health Nudge feature built → feature shelved due to projected engagement impact. The intervention was not shipped.
5. Eli Pariser's "filter bubble" concept describes what mechanism?
Correct. Pariser's filter bubble describes a self-reinforcing personalization loop: interactions signal preferences, the algorithm amplifies matching content, which generates more interaction signals, narrowing the information environment over time.
The filter bubble is specifically the personalization feedback loop — prior interactions teaching the algorithm what to amplify, creating a progressively narrowed information environment around existing preferences.
6. The 2019 Oxford Internet Institute study on filter bubbles found a counterintuitive result. What was it?
Correct. This finding suggests heavy news consumers are not simply sealed in bubbles — the filter bubble effect is real but more complex than popular narratives suggest, and stronger for social feeds than for direct search behavior.
The counterintuitive finding was that heavy news consumers showed more ideological diversity than light consumers — complicating claims that algorithmic curation inevitably seals users in homogeneous bubbles.
7. Facebook's 2014 emotional contagion experiment (Kramer et al., PNAS) established what type of evidence about AI feed curation?
Correct. The study was a randomized controlled experiment — not a survey or correlation — establishing that feed curation causally shifts emotional expression. This remains one of the clearest empirical demonstrations of AI-mediated psychological influence at scale.
The critical word is "causal" — this was a randomized experiment, not a correlation study. The experimental design allows the conclusion that feed content caused changes in emotional expression, not merely that they co-occur.
8. The 2023 Gonzalez-Bailon et al. Facebook study on algorithmic curation found that removing algorithmic curation reduced polarization — but also found what limiting caveat?
Correct. This is the key nuance: voluntarily chosen sources were already partisan, so the algorithm was amplifying tendencies that pre-existed and persisted independent of algorithmic curation. This fundamentally complicates "turn off the algorithm" as a complete solution.
The limiting caveat was that users' voluntary source choices were already partisan — meaning the algorithm amplifies but doesn't originate polarization. This is a meaningfully different (and more complex) causal picture than "algorithms create polarization."
9. BJ Fogg's Behavior Model states that behavior occurs when motivation, ability, and a prompt converge. How do AI persuasion systems apply this?
Correct. This is what makes AI-era persuasive technology qualitatively different from Fogg's original framework: the ability to simultaneously optimize all three elements for each individual user at the specific moment of highest receptivity.
AI persuasion systems can engineer all three elements — personalizing motivation appeals, reducing ability friction, and timing prompts to moments of highest receptivity — simultaneously for individual users at scale.
10. Amazon's Prime cancellation flow, internally codenamed "Iliad," was the subject of a 2023 FTC complaint. What category of dark pattern does it exemplify?
Correct. The roach motel pattern — asymmetric ease of entry versus exit — is precisely what the FTC alleged: the complexity was not accidental but an optimized system designed to retain subscribers against their expressed intent.
The Iliad flow exemplifies the roach motel pattern: entry (sign-up) was easy; exit (cancellation) was engineered to be difficult through multiple pages and deflection screens designed to frustrate completion.
11. 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.
12. Cambridge Analytica's claimed methodology used the OCEAN personality model for psychographic microtargeting. What did its data harvesting of 87 million Facebook profiles rely on?
Correct. The mechanism was friend-of-friend API access: a small number of users consented to the app, but the API allowed harvesting their entire social network's data without those networks' knowledge or consent. Facebook changed this API policy in 2018.
The harvesting relied on Facebook's API allowing third-party apps to collect not just the consenting user's data but their entire social graph — friends who had never interacted with the app. This friend-graph access was the mechanism that enabled scale.
13. The 2020 UCL study on GPS and hippocampal activation found reduced neural engagement in GPS users. What is the broader cognitive implication of this finding?
Correct. The finding generalizes: when AI reliably handles a cognitive function, the neural systems for that function are not being exercised. Whether this constitutes meaningful skill loss or adaptive efficiency depends on whether you ever need the unassisted capacity.
The implication is about reduced engagement of underlying neural systems when external tools handle the function — suggesting that sustained offloading may reduce practice of the cognitive capacity, with potential consequences if unassisted performance is ever needed.
14. Sparrow, Liu, and Wegner's 2011 "Google Effect" paper found that when people believe information is accessible externally, they show enhanced recall of what?
Correct. The Google Effect describes an adaptive memory redistribution: encode the retrieval pathway, not the content, when reliable external storage is delegated. This is cognitively rational but means the person's knowledge is contingent on the external system's availability.
The Google Effect specifically found enhanced recall of where to find information (retrieval pathway) at the expense of the content itself — an adaptive response to delegating storage to an external system.
15. Dell'Acqua et al.'s 2024 "Jagged Technological Frontier" study found that AI-assisted consultants performed 19% worse on tasks outside AI's competence range. What mechanism did the researchers identify?
Correct. The jagged frontier problem is a calibration failure: users cannot accurately assess which tasks fall within versus outside AI's competence, so AI confidence in wrong answers displaces correct human judgment — a specific kind of cognitive offloading harm.
The mechanism was calibration failure: AI provided confident-sounding incorrect output for out-of-range tasks, and assistants substituted this output for their own judgment. The problem is not AI being wrong per se, but users not accurately judging when AI is wrong.