The Pokémon Go creator is enhancing the realism of its online creatures by incorporating conceptual AI into Peridot.
To elevate the authenticity of the augmented reality pet game Peridot, Niantic is introducing conceptual AI to train the game’s colorful extraterrestrial creatures to exhibit behaviors akin to cats. The process involves a sophisticated approach where the Niantic algorithm analyzes real-world objects captured through the player’s phone camera. Subsequently, it converts these objects into textual descriptions, which are then interpreted by a sophisticated language model—specifically, an adapted version of the Meta Llama 2 software. Each virtual pet, referred to as Dots, will possess a unique personality profile reminiscent of various pokémon natures and will be nurtured by the language model. Peridot will engage the language model in discussions regarding how a Dot’s specific personality traits might influence its interactions with nearby elements, with the game’s outcomes being influenced by these interactions.
Niantic anticipates that leveraging AI technology will enable the game to leverage previously untapped visual assets that posed challenges in terms of programming contexts. In an official statement, Niantic expressed that “the incorporation of conceptual AI has expanded the range of potential responses for these visuals.” However, Niantic aims to strike a balance by ensuring that the creatures in Peridot appear realistic without veering into hyper-realism, humorously noting the challenges of dealing with virtual pet messes.
