Spurious normativity enhances learning of compliance and enforcement behavior in artificial agents

Köster, Raphael; Hadfield-Menell, Dylan; Everett, Richard; Weidinger, Laura; Hadfield, Gillian K.; Leibo, Joel Z.

doi:10.1073/pnas.2106028118

Cited by 34 publications

(23 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…3B, this concept can be easily captured by an agent since we provide the frequency of each solution as a feature. We speculate that passing through this sub-optimal strategy facilitates faster learning by providing additional options to copy and helping the agent learn the notion of copying much faster, similar to a recently reported phenomenon in artificial agents for social learning [35]. This is further supported by a delay in the learning process when the frequency feature is not provided (see Supplementary Information for the result where no solution frequency is given).…”

Section: Results a Default Environmentsupporting

confidence: 73%

“…Different from previous studies [32,34,35], the payoff of our work is given by a fixed landscape, not from a game between agents with a payoff matrix. Our work suggests that social learning can emerge even when explicit payoff interaction between agents is not present, which resonates with the importance of vicarious reinforcement [43] in social learning theory.…”

Section: Discussionmentioning

confidence: 99%

“…In particular, RL has shown its strength in constructing computational models of the spatiotemporal dynamics of human cooperation and anthropological problems [32][33][34][35]. Here, we model the problem of social learning by considering a group of individuals that iteratively search for better solutions in a rugged, high-dimensional landscape [6,9,10,18,19,36], where our goal is to find the optimal heuristic for individuals that yields the maximum average payoff when shared with its group (Fig.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Social learning spontaneously emerges by searching optimal heuristics with deep reinforcement learning

Ha¹,

Jeong²

2022

Preprint

View full text Add to dashboard Cite

How have individuals of social animals in nature evolved to learn from each other, and what would be the optimal strategy for such learning in a specific environment? Here, we address both problems by employing a deep reinforcement learning model to optimize the social learning strategies (SLSs) of agents in a cooperative game in a multi-dimensional landscape. Throughout the training for maximizing the overall payoff, we find that the agent spontaneously learns various concepts of social learning, such as copying, focusing on frequent and well-performing neighbors, self-comparison, and the importance of balancing between individual and social learning, without any explicit guidance or prior knowledge about the system. The SLS from a fully trained agent outperforms all of the traditional, baseline SLSs in terms of mean payoff. We demonstrate the superior performance of the reinforcement learning agent in various environments, including temporally changing environments and real social networks, which also verifies the adaptability of our framework to different social settings.

show abstract

Section: Results a Default Environmentsupporting

confidence: 73%

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Social learning spontaneously emerges by searching optimal heuristics with deep reinforcement learning

Ha¹,

Jeong²

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Although learning conformity is consistently associated with the self-awareness of university students, school policy pressure, social pressure, self-development, and academic success, research linking subcontracts of learning conformity and a specific perception of self-efficacy is more limited [22][23][24]. There is no research regarding the relationship between general self-efficacy and learning conformity in the current literature.…”

Section: Self-efficacy Researchmentioning

confidence: 92%

Association between Self-Efficacy and Learning Conformity among Chinese University Students: Differences by Gender

Xiao

Song

2022

Sustainability

View full text Add to dashboard Cite

Background: Although past research highlights the impact of self-efficacy on university students’ learning motivation, examining potential links with specific types of learning conformity is limited. The current study examined associations between Chinese university students’ perceived self-efficacy and learning conformity across different types of learning motivation.Methods: A total of 339 Chinese university students were surveyed using the General Self-Efficacy Scale and the Learning Conformity Scale. Multiple regression models were constructed to focus on the mechanisms of general self-efficacy on learning conformity. Results: (1) There are three types of learning conformity: learning abidance, learning obedience, and learning compliance. (2) General self-efficacy has a negative effect on learning obedience. In contrast, it positively affects learning abidance and learning compliance. (3) The general self-efficacy of girls is lower than that of boys. Still, girls are more likely to be motivated to learn compliance than boys. Conclusions: The study reveals that it is essential to stimulate students’ motivation to learn abidance to love learning from the inside out; to improve girls’ self-efficacy; to raise students’ awareness of self-respect and self-development; and to encourage self-approval in public institutions.

show abstract

“…'in-the-wild,' particular instances during training episodes can be analysed as quasi-experiments. For example, we can aggregate all instances where the AI system is confronted with a choice between two colours, to analyse whether an AI system has learned to associate different colours with different values (Köster et al, 2022). Such simple in-the-lab or in-the-wild tests typically report a binary pass/fail outcome.…”

Section: The Role Of Analysis In Ai Researchmentioning

confidence: 99%

Artificial moral cognition: Learning from developmental psychology

Weidinger¹,

Reinecke²,

Haas³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

An artificial system that successfully performs cognitive tasks may pass tests of 'intelligence' but not yet operate in ways that are morally appropriate. An important step towards developing moral artificial intelligence (AI) is to build robust methods for assessing moral capacities in these systems. Here, we present a framework for analysing and evaluating moral capacities in AI systems, which decomposes moral capacities into tractable analytical targets and produces tools for measuring artificial moral cognition. We show that decomposing moral cognition in this way can shed light on the presence, scaffolding, and interdependencies of amoral and moral capacities in AI systems. Our analysis framework produces a virtuous circle, whereby developmental psychology can enhance how AI systems are built, evaluated, and iterated on as moral agents; and analysis of moral capacities in AI can generate new hypotheses surrounding mechanisms within the human moral mind.

show abstract

Spurious normativity enhances learning of compliance and enforcement behavior in artificial agents

Cited by 34 publications

References 44 publications

Social learning spontaneously emerges by searching optimal heuristics with deep reinforcement learning

Social learning spontaneously emerges by searching optimal heuristics with deep reinforcement learning

Association between Self-Efficacy and Learning Conformity among Chinese University Students: Differences by Gender

Artificial moral cognition: Learning from developmental psychology

Contact Info

Product

Resources

About