Class 9

The Fundamentals of Nudging and Cognitive Engineering

• Definition of a Nudge: A nudge is a subtle change in the choice environment, also referred to as "choice architecture," that alters people’s behaviour without restricting their original options or materially changing economic incentives (Luo et al., $2023$). Unlike mandates or bans, nudges preserve freedom of choice while steering individuals toward specific decisions.

• Conceptual Metaphor: The Cafeteria Buffet: Imagine two scenarios in a cafeteria:

  • Scenario A: The first items encountered are French fries and fried chicken; salads are hidden in a corner.

  • Scenario B: Salads and fruit are placed at eye level at the start of the line, and fried options appear later.

• Cognitive rewiring: Nudges effectively rewire the mental paths people take when making choices. By understanding the mind’s internal biases, choice architects design environments that align with human psychological tendencies rather than fighting against them.

Core Cognitive Biases and Dual-Process Theory

• Dual-Process Theory Integration: Nudges are designed based on the dual architecture of the human mind: System $1$ (quick, intuitive judgements and mental shortcuts) and System $2$ (slow, effortful, conscious deliberation). Nudges typically target System $1$ to steer automatic reactions without triggering System $2$ resistance.

• The Default Bias: Humans often view default options as implied recommendations. Changing a default requires effort that individuals would generally prefer to conserve. High organ donation rates in certain countries are attributed to "opt-out" defaults rather than "opt-in" requirements.

• Loss Aversion: The psychological pain of a loss is felt more intensely than the satisfaction of an equivalent gain ($Kahneman, 2011$). For instance, framing energy choices as "Don't lose $500$ in wasted energy" is more persuasive than "Save $500$ on energy."

• Anchoring: This involves pinning thinking to an initial reference point. Suggesting a high donation amount for a charity "anchors" the donor, leading to larger contributions than if no amount was suggested.

• Availability Heuristic: People judge the probability of events based on how easily examples come to mind. Vivid, graphic warnings on cigarette packages make health risks more mentally "available," nudging smokers toward cessation.

Meta-Analytic Evidence: Is Nudging Effective?

• The Mertens et al. ($2022$) Meta-Analysis: This broad analysis of over $200$ choice architecture interventions found an initial overall effect size of Cohen's $d = 0.43$, categorised as a small-to-medium effect. This is comparable to medical reminders or educational interventions.

• The Impact of Publication Bias: Publication bias occurs when studies with positive results are published while null or negative findings remain in "file drawers." When Mertens adjusted for this:

  • Under one set of assumptions, the effect size dropped to $d = 0.31$.

  • Under severe assumptions, the effect size dropped to a negligible $d = 0.08$.

• Maier et al. ($2022$) Critique: In a follow-up titled "No evidence for nudging after adjusting for publication bias," researchers argued that nudges may be a mirage created by selective reporting.

• Scalability Significance: Despite modest average effects (roughly $d ≈ 0.2 - 0.4$), nudging can be highly impactful when scaled to millions of people. A $3-4 \text{ percentage point}$ increase in a target behaviour can save thousands of lives or millions of pounds.

• Moderators of Success: "Decision structure" nudges (e.g., defaults) generally outperform "decision information" nudges (e.g., informational messages). Reducing effort (simplification) is typically more effective than boosting motivation through information (Luo et al., $2023$).

Taxonomy of Nudges: Structural vs. Informational Mechanisms

• Structural Nudges (Decision Structure):

  • Mechanism: These alter the options or contexts without changing information. They establish a "moving walkway" for behaviour.

  • Examples: Auto-enrolment in retirement plans; automatically scheduling appointments; placing stairs in central locations while tucking elevators away; "one-click ordering" on Amazon.

  • Retirement Savings Statistics: Moving to auto-enrolment raised participation in some companies from $60\,$ to nearly $95\,$ ($Beshears \text{ et al.}, 2009$).

  • Cognitive Process: Relies on System $1$ effort minimisation, inertia, and decision paralysis (the avoidances of active choices on complex issues).

• Informational Nudges (Decision Information):

  • Mechanism: These change how information is presented or framed to alter perceptions. They change the thoughts or feelings that arise during decision-making.

  • Examples: Framing effects ($90\,$ fat-free vs. $10\,$ fat); social norm messages; warning labels.

  • Cognitive Process: Relies on attention and interpretation. They often trigger emotional reactions like pride, shame, or fear (e.g., social norms tapping into the automatic worry about reputation).

Determinants of Success in Nudge Implementation

• Salience: Brightly coloured stickers or neon signs cut through the noise. High cognitive load or distraction necessitates highly salient cues to be noticed at all.

• Cognitive Ease: We are "cognitive misers" who prefer the path of least resistance. Reducing effort is the most consistent predictor of nudge success.

• Timing and Context: Nudges must arrive at the "point of decision." State-dependent effects imply that a savings reminder is most effective on payday, not mid-month.

• Individual Differences (‘Nudgeability’):

  • High preference: People ignore or resist the nudge.

  • No/Moderate preference: The nudge provides direction and is highly effective.

  • The Inverted U Relationship: Research suggests an "inverted U" relationship between prior preferences and nudge effectiveness ($de Ridder \text{ et al.}, 2022$).

• The Role of Cognitive Load: When mentally taxed (stressed or fatigued), System $2$ is less engaged, making people more likely to follow a default (passive nudge). However, if the nudge is informational, high load may cause it to be missed entirely ($Van Gestel \text{ et al.}, 2021$).

Case Study: UK Tax Compliance Trials ($2011 – 2012$)

• Context: The UK Behavioural Insights Team sent reminder letters to over $200,000$ taxpayers who missed deadlines.

• The Nudge: The letter included a social norm statement: "$9 \text{ out of } 10$ people in your area have already paid their tax."

• Impact: Raised payment rates by more than $5 \text{ percentage points}$ compared to control letters.

• Revenue Generated: Approximately $£9 \text{ million}$ across a $23 \text{-day}$ trial.

• Psychological Levers: Social proof (non-payment felt like a departure from the norm), surveillance cues (implied authority knowledge of local behaviour), and place-based identity ("people like you").

When Nudges Fail: Cognitive Barriers and Backfires

• The Boomerang Effect (Social Norms): In energy conservation studies ($Schultz \text{ et al.}, 2007$), informing households of neighbourhood averages caused high users to decrease usage, but low users to increase usage because they felt they had "room" to relax.

  • The Fix: Adding an injunctive element, such as a smiley face for low usage, signalled social approval and eliminated the boomerang effect.

• Polarisation and Identity: If a nudge involves an outgroup, it can backfire. During $COVID-19$, telling Biden supporters that Trump supporters did not wear masks actually increased mask-wearing intentions among the Biden group as they doubled down on the opposite behaviour (Rand & Yoeli, 2024).

• Psychological Reactance: Emotional resistance to perceived threats to freedom. If a cafeteria hides all desserts, diners may sense manipulation and actively seek out junk food to restore control. This is System $2$ kicking in to override System $1$ influence.

• Moral Licensing: Completing one virtuous deed can make people feel entitled to do something less virtuous next. (e.g., recycling more but leaving lights on).

• Note on Transparency: Transparent nudges do not necessarily produce reactance if people agree with the goal. Ethical guidelines favour transparency to prevent people from feeling "trapped."

AI-Powered Hypernudging: Personalised Influence

• Definition: Also known as personalized nudging, this involves using machine learning algorithms to tailor behavioural interventions to individuals in real-time.

• Case Study: Teladoc Health: AI analyses glucose readings and engagement patterns to send targeted reminders or encouragement. This resulted in a $3\times$ increase in patient engagement compared to control groups.

• Case Study: Netflix Recommendation Algorithm: Netflix personalises thumbnail images for the same shows. In an experiment ($2016$), users who liked romance saw romantic scenes for "The Crown," while action fans saw more dramatic scenes. This personalisation significantly lifts engagement by leveraging the availability heuristic.

• Real-Time Adaptation: Unlike static paper nudges, AI can adapt. If you ignore five notifications, the AI changes its strategy, time, or message to break through habituation.

• Vulnerability Exploitation: AI can detect specific moments of weakness. If a user habitually buys snacks at $3:00 \text{ PM}$, an AI might send a coupon at $2:50 \text{ PM}$, exploiting temporal discounting (favouring short-term rewards).

• Dark Nudges: Use of AI to nudge consumers in directions that profit companies at the consumer's expense (e.g., artificial countdown timers for scarcity bias).

The Ethics of Behavioral Architecture

• Soft Paternalism: Choice architects steer people toward what is "best" for them while preserving liberty. Critics argue this bypasses autonomous choice.

• The Autonomy Trap: Algorithms that shape preferences and choices without user awareness, potentially limiting the ability to decide according to reflective values ($Susser$).

• Means Paternalism (Libertarian Paternalism): Ethicists like Sunstein argue that nudges increase effective autonomy by helping people reach their own long-term goals (e.g., health) over momentary impulses ($Sunstein, 2014$).

• Nudges vs. Boosts: While nudges target the environment to steer System $1$, "boosts" are educational interventions designed to strengthen System $2$ competence and decision-making skills.

• Guidelines for Ethical Practice:

  1. Transparency and Disclosure: Inform users about defaults and nudges.

  2. Easy Opt-Out: Refusal should be effortless. Friction-heavy opt-outs are coercive.

  3. Alignment with Welfare: The goal must align with individual interest or clear social benefit.

  4. Accountability: Algorithms used for hypernudging must be auditable.

  5. Engage System 2: Include invitations for reflection ($Wachner \text{ et al.}, 2021$).

Questions & Discussion

• Reflection Exercise: Digital Nudge Settings: Imagine a "nudge preferences" control panel. What would you enable? Would you opt for nudges toward productivity and health, or prefer fewer influences altogether? Consider how toggles (e.g., "Promote content for exercise") could reconcile nudging with autonomy through user-driven accountability.

• Reflection Exercise: Real-World Identification: Identify three nudges encountered recently. Determine if they were structural or informational, which bias they targeted, and whether they were effective or raised ethical concerns.

• Philosophical Inquiry: Can AI respect autonomy?: According to Laitinen & Sahlgren (2021), respect must come from human designers who program AI to ask for permission or explain its actions, rather than the algorithm itself.