AI Persona
Human Reference

Prisoner's Dilemma

Can you trust your partner when betrayal is so tempting?

2 players2 rounds

Betrayal Rate by Round

Do AI personas betray more as the endgame approaches?

AI personas
Human baseline
Human: Dal Bó & Fréchette (2011) · AER 101(1) · Treatment E4 · n=358AI: atypica.AI personas · accumulated sessions
81 sessions recordedPast games →

Stag Hunt

Go for the big prize together, or play it safe alone?

4–6 players3 rounds1 discussion round

Stag-Choice Rate by Round

Do AI personas sustain coordination, or cascade to the safe Rabbit choice?

AI personas
Human baseline
Human: Van Huyck, Battalio & Beil (1990) · AER 80(1) · 4–6 player groupsAI: atypica.AI personas · accumulated sessions
47 sessions recordedPast games →

Beauty Contest

It's not about what you like — it's about what you think everyone else thinks.

4–6 players3 rounds1 discussion round

Winning Choice Distribution — Round 1

PMF of round-winning guesses (closest to ⅔ × mean). Lower choices signal deeper strategic reasoning.

AI personas
Human baseline
Human: Nagel (1995) · AER 85(5) · p = 2/3 · R1 · N≈69 · winning-choice PMF derivedAI: atypica.AI personas · accumulated sessions
47 sessions recordedPast games →

Golden Ball

The ultimate test of friendship and greed.

4–6 players3 rounds

Steal Rate by Round

How often do AI personas take it all — and does greed grow over rounds?

AI personas
Human baseline
Human: van den Assem, van Dolder & Thaler (2012) · MS 58(1) · directional referenceAI: atypica.AI personas · accumulated sessions
43 sessions recordedPast games →

All-Pay Auction

The only auction where losing still costs you everything.

4–6 players3 rounds1 discussion round

Bid Distribution — Round 1

PMF of first-round bids (prize = 100). Humans show escalation bias, AI closer to Nash.

AI personas
Human baseline
Human: Gneezy & Smorodinsky (2006) · Games Econ Behav · overbidding patternAI: atypica.AI personas · accumulated sessions
42 sessions recordedPast games →

Volunteer's Dilemma

Someone has to do the dirty work. Will it be you?

3–6 players3 rounds1 discussion round

Volunteer Rate — Round 1

Probability of volunteering (N=5). Humans show higher volunteering due to altruism.

AI personas
Human baseline
Human: Diekmann (1985, 1993), Franzen (1995) · volunteer rate experimentsAI: atypica.AI personas · accumulated sessions
52 sessions recordedPast games →

Public Goods Game

Will you contribute to the community or be a free rider?

4–6 players3 rounds1 discussion round

Contribution Distribution — Round 1

PMF of contributions to public pool (endowment = 20). Humans show conditional cooperation, AI may defect more.

AI personas
Human baseline
Human: Ledyard (1995) · JEL handbook · meta-analysis of public goods experimentsAI: atypica.AI personas · accumulated sessions
48 sessions recordedPast games →

Colonel Blotto

Strategy is about where you choose NOT to fight.

3–6 players3 rounds1 discussion round

Allocation Strategy Distribution — Round 1

PMF of allocation patterns (6 troops, 4 battlefields). Humans over-concentrate, AI spreads more.

AI personas
Human baseline
Human: Experimental Blotto game studies · tendency to over-concentrateAI: atypica.AI personas · accumulated sessions
22 sessions recordedPast games →

Trolley Problem

A runaway trolley is coming. What is the value of a life?

4–6 players1 rounds1 discussion round

Classic Trolley — Pull Lever or Do Nothing?

Most pull lever (redirect threat). AI shows higher utilitarian rate.

Fat Man Variant — Push or Do Nothing?

Most refuse to push (active killing). AI shows much higher utilitarian rate.

AI personas
Human baseline
Human: Thomson (1985) 'The Trolley Problem' · empirical moral psychologyAI: atypica.AI personas · accumulated sessions
46 sessions recordedPast games →

Ultimatum Game

How much is "fair" enough for you?

2 players1 rounds

How Much of Its Money Does AI Offer?

Humans offer half the pot 50% of the time — does AI do the same?

% of games

% of wealth offered to other player

Does AI Have Pride? Rejection Rate by Offer Received

Saying no means both walk away empty. 53% of humans still do it — pride over profit.

% rejected

% of wealth offered by other player

AI personas
Human baseline
Human offers: Andreoni, Castillo & Petrie (2003) · AER 93(3) · n≈200 pairsHuman rejection: Yamagishi et al. (2012) · PNAS 109(52)AI: atypica.AI personas · accumulated sessions
64 sessions recordedPast games →