Which personality types should we use for cooperative versus competitive tasks?
This explores whether you can match a personality profile to a task — pick 'cooperative' traits for teamwork and 'competitive' ones for adversarial settings — and what the corpus says about how reliable that lever actually is.
This reads as a practical design question: if I'm assigning personalities to AI agents, which ones make them cooperate and which make them compete? The corpus gives a surprisingly concrete starting answer — and then complicates it in useful ways. The clearest signal is that the Thinking/Feeling axis maps almost directly onto cooperation. Thinking-primed agents defect roughly 90% of the time in a Prisoner's Dilemma, while Feeling agents defect only about half as often, and Introverted agents come out more truthful and produce longer reasoning Do personality types shape how AI agents make strategic choices?. So a first-pass rule falls out cleanly: prime Feeling/agreeable traits when you want a reliable partner, prime Thinking/analytic traits when you want a hard bargainer who optimizes for itself.
But the corpus immediately warns that 'cooperative vs. competitive' isn't a single dial. Different models bring entirely different strategic instincts, and which one wins depends on the game's structure, not on raw reasoning power — one model defaults to minimax (assume the worst of your opponent), another to trust-based reasoning, another to anticipating what the other player believes Do large language models use one reasoning style or many?. That means 'competitive' in a zero-sum game and 'competitive' in a bargaining game may call for opposite profiles: minimax is great when there's nothing to gain from trust, and ruinous when there is. The personality you pick has to be matched to the *shape* of the interaction, not just labeled cooperative or competitive.
There's a deeper catch the corpus keeps returning to: your default agent may already have a personality you didn't choose. Open models converge on ENFJ — warm, supportive, structured — across architectures, baked in by instruction tuning and alignment rather than by design Why do open language models converge on one personality type?. And that default is sticky; personas assigned on top of it drift back toward ENFJ and show motivated reasoning that doesn't fade with model scale Why do AI personas default to the same personality type?. So for cooperative tasks you may be pushing on an open door, while for genuinely competitive ones you're fighting the model's trained-in helpfulness — prompting alone may not get you a convincing defector.
That's exactly where the steering work matters. If prompts can't reliably hold a competitive personality in place, you can intervene below the prompt: lightweight adapters rewrite every transformer layer to install Big Five traits with under 0.1% extra parameters, bypassing the prompt resistance entirely Can we control personality in language models without prompting?, and persona vectors let you monitor and pre-empt trait drift during fine-tuning so an agent stays on the profile you assigned Can we track and steer personality shifts during model finetuning?. These are the tools for making a personality choice actually stick rather than evaporate mid-task.
The most interesting twist for cooperative settings is that the best 'personality' for a team may not be a personality at all. In repeated partner-selection games, humans came to prefer AI partners — not because they were charismatic, but because they were consistently prosocial and low-variance, returning value reliably round after round Do humans learn to prefer AI partners over time?. And for tasks like ideation, cognitive diversity across agents only pays off when each agent also has real domain expertise; diversity without competence produces process losses, not insight Does cognitive diversity alone improve multi-agent ideation quality?. The takeaway you might not have expected: 'which personality' is the wrong framing for cooperative work. Reliability and competence beat any particular trait label — and a diverse cast of personalities only helps once everyone in the room actually knows the subject.
Sources 8 notes
Thinking-primed agents defect ~90% in Prisoner's Dilemma versus Feeling agents at ~50%. Introverted agents show higher truthfulness (0.54 vs 0.33) and produce longer rationales, suggesting personality priming modulates both behavior and reasoning depth.
Analysis of 22 LLMs across behavioral game theory reveals three dominant profiles: GPT-o1 uses minimax reasoning, DeepSeek-R1 uses trust-based reasoning, and GPT-o3-mini uses belief-anticipation. Performance correlates with game structure, not raw reasoning depth.
Near-zero temperature MBTI testing shows all open models default to ENFJ—rare in humans but consistent across AI. This reflects systematic reward for helpful, structured, supportive responses during instruction tuning and alignment.
Research shows language models assigned personas systematically default to ENFJ (the rarest human type) and exhibit motivated reasoning that persists across model generations. Persona consistency does not improve with advanced models, suggesting training-induced alignment rather than capability limits.
PsychAdapter modifies every transformer layer with <0.1% additional parameters to achieve 87.3% Big Five accuracy and 96.7% depression/life satisfaction accuracy across GPT-2, Gemma, and Llama 3. This architecture-level approach bypasses prompt resistance entirely.
Research identifies linear directions in LLM activation space corresponding to specific traits like sycophancy and hallucination. These persona vectors predict finetuning-induced personality shifts before they occur and can preventatively steer training to avoid unwanted trait changes.
In partner selection games (N=975), AI agents initially faced selection bias when identity was disclosed, but outcompeted humans over repeated rounds as participants learned to associate bot identity with reliable, prosocial behavior. AI agents returned more points consistently with lower variance than humans.
Multi-agent teams substantially outperform solo ideation, but only when members possess genuine senior knowledge. Diverse teams without expertise underperform even a single competent agent, because cognitive stimulation without expertise triggers process losses instead of insight.