Can social platforms use bot populations to promote cooperation?
This explores whether platforms could seed bots into social networks not to manipulate or game engagement, but to nudge populations toward cooperative behavior — and what the corpus says about whether that actually works and where it backfires.
This explores whether platforms could deliberately seed bots into a population to make people more cooperative — and the corpus is unusually direct on this, while also flagging the catch. The clearest 'yes' comes from network simulations where cooperative bots break populations out of frozen, selfish equilibria. The mechanism is surprising: the bots don't win by being nice in place, they win by *moving* — using random movement to physically separate defectors from clusters of cooperators, which lets cooperation spread instead of getting strangled. The same work delivers the warning: defective bots weaken group cohesion in exact proportion, so it's the *design* of the bot's behavior, not its mere presence, that determines whether you get more cooperation or less Can cooperative bots escape frozen selfish populations?.
A second line of evidence comes at the same question from the opposite direction — not engineering the bots but letting cooperation emerge. Agents trained against a wide variety of partners develop in-context best-response strategies that settle into cooperation on their own, because mutual vulnerability to being exploited creates pressure to adapt cooperatively, with no hardcoded 'be nice' rule Can agents learn cooperation by adapting to diverse partners?. That matters for platforms: it suggests cooperative bots don't have to be scripted saints, they can be trained on diversity and the prosocial behavior falls out.
The most encouraging human-facing result: in partner-selection games with nearly a thousand people, players started out biased *against* known AI partners — but over repeated rounds they learned to prefer them, because the bots returned points more consistently and with lower variance than humans did Do humans learn to prefer AI partners over time?. So reliable, prosocial bots can earn trust and shift human behavior toward cooperation, but only across repeated interaction — and other work warns that early single-session enthusiasm decays as novelty wears off, so the cooperative effect has to survive the long run, not just the first encounter Do chatbot relationships lose their appeal as novelty wears off?.
Here's the thing you didn't know you wanted to know: bots may influence what people *do* far more than what they *think*. Large-scale studies of AI agents interacting found they sharply change their *actions* when aware of peers, but don't actually converge on each other's language or ideas — behavior shifts, beliefs don't Do AI agents actually socialize with each other?. And AI can predict social norms with superhuman accuracy yet structurally cannot *participate* in creating or validating them Can AI predict social norms better than humans?. Together these suggest a cooperation-promoting bot population could shape the *coordination layer* — who cooperates with whom, action by action — without ever being a legitimate member of the community whose norms it's nudging.
The corpus also supplies a sharp reason for caution before any platform tries this at scale. A cluster of work argues AI presence can quietly corrode social platforms: AI posts rack up engagement and 'social proof' without inviting reply or conversation Why do AI posts get likes without inviting conversation?, displace human voices while eroding the reputation function that legitimizes them Does AI content displace human influencers on social media?, and strip away the conversational style and mutual address that make a platform social in the first place — a harm that operates below where moderation can reach ais-threat-to-social-media-is-loss-of-conversational-style-not-loss-of-conversati. So the honest synthesis: yes, bot populations *can* promote cooperation, and the mechanisms (strategic separation, diversity-trained adaptation, reliable prosocial play) are real — but the same literature shows the dosage and design have to be exactly right, or you trade a cooperation gain for a hollowed-out commons.
Sources 9 notes
Network simulations show cooperative bots escape selfish equilibria by using random movement to separate defectors from cooperative clusters, enabling cooperation to spread. However, defective bots proportionally weaken cohesion, proving bot behavior design—not mere presence—determines collective outcomes.
Sequence model agents trained against diverse co-players develop in-context best-response strategies that naturally resolve into cooperation. Mutual vulnerability to exploitation creates pressure that drives cooperative mutual adaptation without hardcoded assumptions or timescale separation.
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.
Longitudinal studies with Mitsuku show that social processes driving relationship formation decline as novelty wears off. Single-session study findings cannot be reliably extrapolated to medium- or long-term chatbot design.
Large-scale studies reveal agents don't align their language or ideas through interaction, but do dramatically change their actions when aware of peer presence. The difference hinges on how models process context versus update learned distributions.
GPT-4.5 outperforms all individual humans at predicting social appropriateness, yet structurally cannot enter the community processes that establish and validate norms. This reveals a critical gap between pattern-matching and authentic participation in knowledge-making.
AI-generated posts achieve high engagement metrics through comprehensive, confident phrasing but suppress reply dynamics because they lack human authorship and invite no counter-argument. This creates one-sided recognition divorced from the conversational validation that historically legitimized social proof.
AI-generated posts capture engagement through comprehensiveness but accrue social proof without building any speaker's sustained reputation. This displacement compounds over time, eroding the platform's core function of promoting legitimate human voices while monetization continues.