Artificial intelligence methods have been used to model human cognition for decades. Neural networks hold a special place among these methods as 1) human researchers can train neural nets using data sets to simulate learning, and 2) neural nets can be paired with genetic algorithms to simulate the biological evolution of cognitive abilities. In this project, however, I take an additional third approach. I allow neural networks to learn from each other instead of human researchers, and then evolve skills and cognitive abilities via cumulative cultural evolution. The result is a new form of A.I. and an enriched way of simulating human cultural evolution.
Cultural Transmission Between HomiNets
In some of my models, neural net simulations of hominins (aka ‘HomiNets’) transmit cultural skills to eachother via social learning algorithms. For example, the output of a wise and knowledgeable HomiNet can be used as the training data set for a novice HominNet to simulate prestige-biased imitation.
Key areas of my HomiNet research include:
- Social learning biases in neural networks
- The topological replication of memes in neural network connection patterns
- The emergence of mirror neurons, imitation, and symbolic communication via spike-timing dependent plasticity in neural networks
- Gene-culture coevolution of prosocial neuroendocrine systems and cumulative culture within the hominin lineage
- Evolution of prestige-biased social hierarchies in neural networks
- Cultural evolution within the social networks of neural network populations
P.S. Don’t worry, these HomiNets will not evolve into SkyNet and take over the world Terminator-style! I prefer the scenario where humans co-evolve with artificial intelligence.
No HomiNets were harmed during the running of these simulations. All studies were approved by the ethics committee of Cybertron.