Module 9 · Game design theory

Player psychology: motivation and biases

Two questions sit underneath everything else: why people play and who plays. Intrinsic motivation rests on three needs (autonomy, competence, relatedness); on top of that run cognitive biases (loss aversion, the endowment effect, sunk costs) — and whether you use them to build meaning or to exploit is where design ethics lives.

~17 min🎲 design + 🧭 ethics

The gist in 30 seconds

Why people play — Self-Determination Theory (Deci and Ryan): intrinsic motivation rests on three needs — autonomy (meaningful choice), competence (mastery with clear feedback), relatedness (connection to others or to the world). Satisfy them and the game is its own reward; lean on extrinsic rewards (cosmetics, ladders) and you risk the overjustification effect: an external reward crowds out the internal one ("pay me for the painting → now it is work, not art"). Who plays — taxonomies: Bartle (4 types from MUDs: Achiever/Explorer/Socializer/Killer) was the first model; Quantic Foundry (Nick Yee, 400K+ surveys) is the data-driven extension: 12 motivations across 6 domains. Games also lean on biases: loss aversion (a loss lands roughly 2× harder than a gain), the endowment effect, sunk costs ("I've put in 100 hours — I'm not quitting"). The ethical line: use biases for meaningful engagement (mastery, story) rather than for obligation (dailies, FOMO deadlines).

The mechanism: why they play and who plays

SDT: the three needs of intrinsic motivation

Autonomy is a sense of control: the player chooses how to play, and tasks have several solutions (BG3: pass a check through dialogue, stealth or combat). Competence is a sense of mastery: challenge matched to skill, with clear feedback (Celeste: learn A, then A+B, then under a timer). Relatedness is connection to others or to the world: social interaction, narrative immersion (the relationships in Hades, co-op in FFXIV). Satisfy all three and the game becomes its own reward.

Intrinsic vs extrinsic and overjustification

Intrinsic: you play because it is interesting (Balatro — the act of assembling the deck is itself pleasurable). Extrinsic: you play for an external reward (cosmetics, rank, an achievement). The trap: extrinsic rewards can undermine intrinsic motivation (the overjustification effect, Deci's classic 1971 experiment): if you love drawing and someone starts paying you for a picture, drawing suddenly feels like work rather than art. In games: unlocking cosmetics is pleasant, but once the cosmetics run out players leave, because they were playing for them and intrinsic motivation had died by that point. The principle: intrinsic first.

✗ Live-service failure

A strong core loop (intrinsic) → the publisher adds a battle pass and cosmetics (extrinsic) → players grind cosmetics instead of playing for enjoyment → the cosmetics are limited and expensive → it feels like milking → they leave (extrinsic does not motivate forever, and intrinsic is already dead).

✓ Elden Ring success

No battle pass, no cosmetic grind. Pure intrinsic motivation: challenge, exploration, discovery. Players come back for years.

Who plays: Bartle and Quantic Foundry

Bartle's taxonomy (from his MUD research, 1996) has four archetypes: Achievers (goals, points, progression/ladders), Explorers (secrets, lore, understanding the world), Socializers (guilds, conversation, doing things together), Killers (dominance, PvP). Most players mix types; the model reflects the MUD era and simplifies. Quantic Foundry (Nick Yee, author of "The Proteus Paradox"; running the GMM since 2015) is the most data-driven extension: 12 motivations across 6 domains, based on 400,000+ surveys:

Domain	Motivations
Action	Destruction, Excitement
Social	Competition, Community
Mastery	Challenge, Strategy
Achievement	Completion, Power
Immersion	Fantasy, Story
Creativity	Design, Discovery

Its value in 2026: it maps onto feature priorities more concretely than a four-type model, it has data (correlations between motivations, regional differences — Korea scores higher on Competition, the West on Discovery), and for monetization a motivation profile predicts willingness to pay and churn risk better than demographics. Aim at 2–3 motivations; trying to cover all of them dilutes everything.

Cognitive biases (and ethics)

Loss aversion (Kahneman–Tversky): a loss feels roughly twice as strong as an equivalent gain.

| v (- x) | \approx λ \cdot v (x), λ \approx 2.25

Hence the design: losing progress hurts (respawning locally while keeping upgrades vs restarting the whole game; roguelikes treat permadeath with permanent progression). The endowment effect: you value what you own more highly — cosmetics attach you to your character, and artificial scarcity (limited-time items) raises the sense of ownership. Sunk costs: "I've put in 100 hours, I'm not quitting" — long progress bars and limited-time events hold people with login streaks. The ethical line: good is engagement that feels meaningful (progress = mastery, story); bad is engagement as an obligation (daily chores, battle-pass deadlines, FOMO). The same levers, in detail and from the monetization side, are in gacha and battle passes.

🕹 What to play — and what to notice

BG3 / Celeste / FFXIV the three SDT needs

BG3 = autonomy (you solve a check by dialogue/stealth/combat). Celeste = competence (a clean ladder of mastery plus clear feedback). FFXIV/Animal Crossing = relatedness (community, doing things together).

🎮 Do: for a game you love, name which of the three SDT needs it satisfies most strongly and through what exactly. Then find a game that bored you quickly — almost always one of the three needs is sagging (no choice / no growth / lonely).

Your own motivation profile Quantic Foundry

The free Gamer Motivation Profile test shows your top motivations out of the 12. Useful both for yourself and for seeing how far your design intent matches the real motives of your audience.

🎮 Do: take the test at gamerprofile.quanticfoundry.com and look at your top 3 motivations. Then guess the profile of your target audience for Novgorod and check it: which 2–3 motivations are you actually aiming at?

Cosmetic grind / a FOMO banner overjustification + biases

A game where you ground for cosmetics and quit once they ran out is overjustification in the flesh. A limited-time gacha banner with a countdown is loss aversion plus scarcity.

🎮 Notice: remember a game you started playing "for the reward" rather than for the process — and how fast you burned out once the rewards dried up. Catch a deadline in any live-service game (an event or pass expiring) and name which bias it presses on: loss aversion, scarcity or sunk costs.

Deep end · the value function and prospect theoryskippable

The asymmetry of losses and gains

Prospect theory (Kahneman–Tversky, 1979): people evaluate outcomes not by absolute utility but relative to a reference point, and the value function $v$ is concave over gains, convex over losses and steeper on losses ( $λ \approx 2.25$ ). This explains why FOMO works: "missing" a banner feels like a loss (×2) rather than "not acquiring" — the same item is framed as a decrement from your current reference point. Design shifts the reference point (first "give" you the item or discount → now taking it away is a loss) and triggers a disproportionate reaction.

Reference dependence in retention

Streaks and progress bars create a reference point ("I have a 40-day streak") whose breaking = a loss → a powerful anchor. It is ethical when the progress reflects real mastery or meaning; it becomes a dark pattern when the streak's only function is to retain through fear of loss with no value inside. The line is engineering-thin: the same mechanism (permanent progression in a roguelike) can either soothe the pain of permadeath or forge a sunk-cost trap — the difference is whether it serves the player or the metric.

Deep end · design: SDT as a checklist, targeting motivations, intrinsic-firstskippable

SDT as a review checklist

For every major system, ask: does it provide autonomy (is there meaningful choice, or an illusion of choice?), competence (is growth visible and the feedback clear?), relatedness (is there a connection to the world or to people?). A system failing usually means one of the three failed (grind with no choice kills autonomy; unclear feedback kills competence).

Targeting 2–3 motivations

Like the 8 kinds of fun, the Quantic Foundry motivations are goals rather than "all at once". A Mastery+Immersion game (a soulslike) and an Action+Social one (a hero shooter) want a different loop, economy and communication. The motivation profile → feature priorities and, separately, the monetization model (cosmetics monetize Achievement/Creativity, a pass monetizes Completion, PvP ranks monetize Competition).

Intrinsic-first as a defense against Goodhart

Optimizing extrinsic proxies (engagement metrics, cosmetic grind) degrades the true goal (enjoying the game) — that is Goodhart from the human side. Intrinsic-first is the defense: first a loop that is its own reward, and only then a careful external layer that does not replace it.

Analogy

Motivation is like a campfire. Intrinsic motivation is a self-sustaining flame: autonomy, competence and relatedness are the fuel, the air and the heat that keep it going on their own; the player plays because the game is already the reward. Extrinsic rewards (cosmetics, badges) are like splashing on petrol: a bright flare, but it burns out, and worse, it drowns the real fuel — train someone to play for the flare and the self-sustaining flame goes out (overjustification). Good design feeds the fire itself (mastery, meaning); predatory design keeps pouring petrol until the player burns out and the fire dies.

Why it matters

This is the "why" underneath everything else: retention, monetization and feel all come back to motivation. Knowing SDT, knowing who your players are and which biases you are engaging is at once a design tool and an ethical responsibility. And the pairing "intrinsic vs extrinsic, the proxy destroys the goal" is the reward-hacking/Goodhart theme seen from the angle of human incentives.

🔁 Beyond games — where this transfers

The lesson is about incentives: what actually drives behavior and how an external reward destroys an internal one.

ML / AI (your domain): intrinsic vs extrinsic ⇄ intrinsic-motivation / curiosity-driven RL (an internal reward for novelty or competence against an external task reward), and the overjustification effect ⇄ reward crowding / over-optimization: a badly specified extrinsic reward kills the behavior you wanted (reward hacking from the incentive side; RLHF over-optimization degrades base capability). Loss aversion / prospect theory ⇄ models of human preference are nonlinear: RLHF learns a human value function with biases baked in (reference dependence, asymmetry) — so preference data ≠ true utility. Motivation taxonomies (Quantic Foundry clustering 400K surveys) ⇄ unsupervised segmentation of user behavior, and profile → churn/willingness-to-pay prediction ⇄ predictive user modeling. "Intrinsic first" ⇄ the alignment lesson: optimizing a proxy (the extrinsic) degrades the true goal — Goodhart from the human side. And the ethics of dark patterns ⇄ responsibility in systems that optimize engagement (recommenders) not to exploit the same biases.

Product/HR: autonomy-mastery-purpose (Pink, "Drive") is the same trio for employee motivation and user retention; bonus schemes frequently run into overjustification.

Incentives in systems: any KPI or bonus is a proxy that agents (people, services) optimize literally; build in defences against gaming the metric.

Principle: build on intrinsic motivation; dose external rewards carefully — they easily crowd out the thing they were supposed to amplify; use biases honestly.

🔧 Try it on and design

🎮 Profile + SDT audit ~25 min

Take the Gamer Motivation Profile (Quantic Foundry) and find your top motivations. Then take 2 games (one you love, one you got tired of) and run them through the SDT checklist: where autonomy/competence/relatedness are present and where they sag. Connect "got bored" to the sagging need.

🧭 Ethical breakdown ~15 min

Find one engagement mechanic in any live-service game (a streak, a pass, a limited-time item) and classify it: does it give meaningful engagement or obligation? Which bias does it press on (loss/endowment/sunk cost)? How would you rebuild it as an intrinsic version for your own Novgorod.

Checklist: found your motivation profile; ran 2 games through SDT; classified a mechanic by ethics and bias; designed an intrinsic-first alternative; connected it to intrinsic-reward RL / Goodhart.

Connections

foundation

MDA and the core loop — "aesthetics" = which motivations the game aims at; SDT and motivations set the target.

adjacent

Gacha and battle passes — the same biases, but from the monetization (and more predatory) side.

adjacent

Retention — motivation is the engine of retention; intrinsic holds longer than extrinsic.

Game feel — competence needs immediate, clear feedback, and feel is what provides it.

Questions worth asking

Can an extrinsic reward really kill intrinsic motivation?

Yes, it is a reproducible effect — overjustification. The classics: Deci (1971) and Lepper–Greene–Nisbett (1973, the "overjustification effect" with children and drawing). People who did something for enjoyment started being paid or rewarded — and when the reward was removed they did it less than those who were never paid at all. The mechanism: the brain re-attributes the cause ("I do this for the reward, not because I like it"), and as soon as the external cause disappears so does the motive. In games you see it when a player grinds cosmetics or a battle pass instead of enjoying the loop — and when the rewards run out they leave entirely, because intrinsic interest has been crowded out by then. A nuance: not every reward is harmful — unexpected rewards and rewards for quality (rather than for mere participation) do less damage or help; what kills intrinsic motivation is an expected, contingent reward for something that was already interesting. Hence the rule: the self-rewarding loop first, the external layer on top and carefully.

Bartle or Quantic Foundry — which should you use?

Bartle (1996) is excellent intuition and a shared vocabulary (achiever/explorer/socializer/killer), but it is a model of the MUD era, made of four coarse types, without large-scale data, and real players mix types. For actual product decisions in 2026 take Quantic Foundry: 12 motivations across 6 domains, built on 400K+ surveys, with correlations and regional data. The practical difference: Bartle says "you have explorers" (and then what?), while Quantic Foundry says "your audience scores high on Fantasy+Story and low on Competition", which translates directly into feature priorities (invest in the world and narrative, not in ranked PvP) and into monetization (cosmetics and story will sell, a competition battle pass will not). Use Bartle to think and explain; use Quantic Foundry to decide and target. In both cases the same rule applies: aim at 2–3, not at all of them.

Is using loss aversion and sunk costs unethical?

The mechanism itself is neutral — the question is what it serves. The line: does the mechanic give the player real value, or does it exist only to retain them through fear or guilt? Permanent progression in a roguelike uses loss aversion to soften the pain of permadeath and give a sense of growth — that serves the player. A streak that expires at midnight and whose only function is to force a login exploits the same bias against the player. The dark-pattern tests: (1) would the player be better off if the mechanic were removed (then it was pressuring rather than giving)? (2) is it built on a deadline or fear rather than on enjoyment? (3) are the odds and costs transparent (gacha with an opaque pity system is worse)? (4) does it target the vulnerable disproportionately (whales, minors — see privacy/ethics)? The line is engineering-thin, but the compass is simple: a mechanic should work for the player, and "engagement" through obligation is a loan against future burnout and reputation.

How does this map onto intrinsic-reward RL and RLHF?

Closely. Intrinsic motivation in RL is the direct analogue: the agent gets an internal reward for novelty, prediction error or competence (curiosity, RND, empowerment) on top of a sparse external task reward — exactly as SDT's competence and autonomy feed a player beyond external points; it fixes exploration where external reward is nearly absent. Overjustification ⇄ reward over-optimization: a badly specified extrinsic reward crowds out the desired behavior (reward hacking), and RLHF pushed too hard on a proxy reward degrades base capability — "breaking things by taking an external incentive too literally". Prospect theory ⇄ preference models: RLHF learns a human value function, and that function is nonlinear and biased (reference dependence, loss asymmetry λ≈2.25), so preference data represents biased preferences rather than true utility, and that has to be kept in mind when interpreting a reward model. And the through-line alignment lesson is the same as "intrinsic first": optimizing a proxy destroys the true goal (Goodhart) — build on the real objective and dose the external signals.