Artificial cultural belief evolution – ACBE

Social aspects of human beings have induced their remarkable cognitive capabilities. Cultural evolution is considered to have taken over biological evolution in their development. Yet, current work in artificial intelligence, besides one-to-one dialogues, does not take it into account. We aim at bringing cultural evolution to machines, like artificial intelligence aims to do with intelligence. Hence ACBE is a project of artificial cultural evolution. Although this applies globally to all cultural artifacts, ACBE focuses on knowledge and beliefs, considered as a main driver of agent behaviour. ACBE will provide (a) models of cultural evolution games and agents, (b) a simulator for running experiments according to these models, (c) a framework for formally describing experiments to ensure their reproducibility, (d) experiments demonstrating the benefits of the approach, and (e) artificial cultural evolution awareness-raising initiatives. Numerous experiments simulating cultural evolution have been carried out, but they are designed in a one-shot fashion preventing to reuse and compare them. In particular, they do not isolate explicitly the evolutionary mechanisms of variation, selection and transmission. This prevents them from being easily identified and controlled in experiments and thus from assessing their influence on knowledge quality and diversity. ACBE will design a general model for games and agents involved in cultural evolution experiments (a). It will make explicit these parameters. Such a model should be prone to computer simulation and theoretical analysis. It will be implemented in a simulator (b) which allows agents to play different games and games to be played by heterogeneous agents in the same experiment. There exist no such environment yet. Experiment will be described formally (c) so that they can be easily modified, reproduced and their results compared. This will improve experiment reproducibility and permit searching and globally analysing an experiment repository. ACBE will perform experiments (d) demonstrating the benefits of the proposed approach. These will involve different agent populations and generations, controlling the evolutionary mechanisms intertwined in agents behaviour and involving different games and heterogenous agents. In particular, we aim at assessing the necessity and relative influence of each evolutionary mechanisms across different games. We also plan to establish the capability of heterogeneous agent populations, including some based on generative artificial intelligence, to evolve shared knowledge and beliefs. Finally, we will promote the proposed approach towards cultural evolution research communities and particularly those from social sciences and the humanities (e). The long-term prospect of ACBE is to understand how artificial systems can find their place in our societies.