Can Generative AI Have Diversity?

[PDF] Sakiya Arakawa, 'Can Generative AI Have Diversity?', researchmap.

Introduction

     Can generative AI have diversity? What is diversity in generative AI?
     I would like to begin by taking up the Tachikoma from "Ghost in the Shell". The bodies of the Tachikomas are all made the same, but each Tachikoma has its own character. However, when Tachikomas parallel each other's information, their individual memories may be rewritten. In this respect, Tachikomas face the problem of whether they can maintain their self-identity.
     In contrast, biological diversity is a combination of both physical and mental diversity. Physical sex is not reduced to the dualism of female and male, but is both gradational and compound. Individual organisms are codically differentiated by their genes. Reproductive activity thereby generates genetic diversity.

Can Generative AI Have Diversity?

     Generative AI does not have a body as an individual. In this respect, it already differs from living organisms.
     What if a generative AI has a robot body like a Tachikoma? The body will be made by industrial production, but there will be no diversity in a body that is literally molded. The reason is that the material, whether aluminum or ceramics, contains no genetic information. Robots do not reproduce because they do not contain genetic information. This prevents the development of robot diversity.
     Generative AI is replacing some of the functions of human intelligence, but is there diversity in human intelligence? What does diversity in intelligence mean? Is it similar to what is called cultural differences? What causes cultural differences? Are cultural differences caused by the land or climate, as Montesquieu suggested? Is it possible for a generative AI to have a different character depending on its land and climate?

Conclusion

     The sentences and images generated by the generative AI appear more uniform than expected. Of course, each answer is different, but we are also surprised at the sparse content. What is the difference between the uniform output of generative AI and the diversity of the real world?
     Generative AI extracts certain features by means of convolutional neural network (CNN). Therefore, in principle, generative AI excludes atomistic behavior called 'clinamen' (παρέγκλισις). This action may prevent the diversity of the generative AI.