• Corpus ID: 233181842

Bootstrapping of memetic from genetic evolution via inter-agent selection pressures

@article{Guttenberg2021BootstrappingOM,
  title={Bootstrapping of memetic from genetic evolution via inter-agent selection pressures},
  author={Nicholas Guttenberg and Marek Rosa},
  journal={ArXiv},
  year={2021},
  volume={abs/2104.03404}
}

• N. Guttenberg, Marek Rosa
• Published 7 April 2021
• Computer Science, Biology
• ArXiv

We create an artificial system of agents (attention-based neural networks) which selectively exchange messages with each-other in order to study the emergence of memetic evolution and how memetic evolutionary pressures interact with genetic evolution of the network weights. We observe that the ability of agents to exert selection pressures on each-other is essential for memetic evolution to bootstrap itself into a state which has both high-fidelity replication of memes, as well as continuing production of new memes over time. However, in this system there is very little interaction between this memetic ’ecology’ and underlying tasks driving individual fitness — the emergent meme layer appears to be neither helpful nor harmful to agents’ ability to learn to solve tasks. Sourcecode for these experiments is available at https:// github.com/GoodAI/memes Recent advances in machine learning have refined our understanding of ways that an individual agent might learn. Could we take advantage of ideas about social or cultural learning such as Dual Inheritance Theory (Cavalli-Sforza and Feldman, 1981; Boyd and Richerson, 1988) in populations of agents to discover new kinds of learning algorithms? Whereas in an individual agent we might represent knowledge, skills, or behavioral norms as explicit memories, datasets, or neural network weights, in a social context information is kept alive via transmission from agent to agent. We investigate how message exchange in a population of (genetically evolving) agents can lead to self-replicating patterns, or ’memes’ (Dawkins, 1976), which may be capable of non-genetic adaptation. Our motivation is to discover learning algorithms for multi-agent cognitive systems (Rosa et al., 2019) based on social and cultural adaptation. We pursue a ’Learning to Learn’ style of approach (Hochreiter et al., 2001) in which the learning algorithm is discovered rather than designed. Here we focus on initial stages of this goal by trying to understand what it takes for communication dynamics to lead to adaptation in messages for the messages’ own sake. The main idea is that while finding a message that replicates regardless of the underlying agent’s network weights may be impossible in general, if the memetic degrees of free0 50 100 150 200 250 300 Step (x1000) 0