Effect of neuromodulation on performance in game playing: A modeling study

Reciprocity and Retaliation in Social Games With Adaptive Agents

“…The neural agent was rigorously tested in a previous modeling study against simple opponents with fixed strategies [20], [21]. The focus of this study was to move past using simple opponents with fixed strategies and introduce adaptive neural agents.…”

“…In prior work to assess the neural agent's performance [20], [21], we analyzed the responses of the neural agent in the Hawk-Dove game against three specific opponent strategies: 1) statistical-its opponent chose to Escalate either 25% or 75% of the time; 2) tit-for-tat (T4T)-its opponent copied the most recent move of the neural agent; or 3) win-stay, lose-shift (WSLS)-its opponent selected the same action that led to a positive payoff in the previous game (win-stay), or selected a different action from the previous game if that action led to zero or negative payoff (Lose-Shift). The purpose of these initial simulations was to develop the model of neuromodulatory decision-making against simpler models and opponents.…”

“…This allowed us to develop and analyze the model guiding the neural agent's behavior under more controlled conditions. The neural agent adapted its behavior appropriately to the environmental conditions (see Table IV) and its opponent's strategy [20], [21]. The neural agent tended to be less aggressive (i.e., chose Display more frequently) when playing in a harsh environment where the probability of a serious injury was high, or against an aggressive opponent that retaliates (T4T).…”

“…Previously, we developed a computational model of neuromodulation and action selection based on the assumptions that dopamine levels are related to the expected reward of an action, and serotonin levels are related to the expected cost of an action [20], [21]. That is, serotonergic activity might track the expected cost of an action in a similar way that dopaminergic activity is thought to track the expected reward of an action [22], [23].…”

“…That is, serotonergic activity might track the expected cost of an action in a similar way that dopaminergic activity is thought to track the expected reward of an action [22], [23]. The neural agent adapted its behavior appropriately to changes in environmental conditions and to changes in its opponent's strategy [20], [21]. The present study embedded our neural model of neuromodulation and action selection in both simulated and embodied neural agents to investigate reciprocal social interactions in games of cooperation and conflict with human subjects.…”