| CPC G06T 17/00 (2013.01) [G06N 5/022 (2013.01); A63F 13/56 (2014.09); G06F 3/011 (2013.01)] | 20 Claims |
|
1. A method comprising:
associating, by a processor of a computing system and based on instructions stored in a memory of the computing system, a knowledge store kept in the memory with an artificial intelligence (AI) character generated by the processor based on an AI character model stored in the memory, the AI character model including a first plurality of heuristics machine learning models and a second plurality of primary machine learning models, wherein:
the AI character is a virtual character configured to interact in a virtual world with a user, the virtual world including a plurality of further AI characters present in the virtual world;
an interaction of the AI character with the user includes transitioning between time periods of a plurality of time periods of a timeline in the virtual world based on a storyline provided to the user in the virtual world; and
the knowledge store stores a plurality of data including:
world knowledge information available to the AI character and the plurality of further AI characters present in the virtual world; and
personal knowledge information available solely to the AI character in the virtual world;
the AI character model is set to generate one or more responses of the AI character based on the world knowledge information and the personal knowledge information associated with the AI character;
determining, by the processor, context parameters for the AI character, wherein the determining the context parameters includes:
collecting a plurality of streams of inputs associated with the interaction of the AI character with the user, the plurality of streams including a first stream of a user input received from the user during the interaction, a second stream of environment parameters associated with the virtual world, and a third stream of events triggered by the user to occur in the virtual world during the interaction;
pre-processing the plurality of streams of inputs to transform the first stream of the user input, the second stream of environment parameters, and the third stream of events into word embeddings, the word embeddings including representations of a plurality of words, wherein a first word and a second word of the plurality of words having related meanings are represented by a first word embedding and a second word embedding, wherein representational characteristics of the first word embedding correspond to representational characteristics of the second word embedding; and
passing the word embeddings through the first plurality of heuristics machine learning models to produce outputs of the first plurality of heuristics machine learning models, the outputs including the context parameters;
determining, by the processor, a current time period of the plurality of time periods associated with the storyline;
based on the context parameters and the current time period, defining, by the processor, a scope of knowledge available to the AI character from the knowledge store;
based on the defined scope of knowledge, constricting, by the processor, access of the AI character to the world knowledge information in the knowledge store by:
selecting, based on the scope of knowledge, a portion of the plurality of data in the knowledge store, the portion of the plurality of data being selected from the world knowledge information, the portion of the plurality of data including information to be possessed by the AI character, wherein the portion consists of facts solely pertaining to the virtual world; and
setting the second plurality of primary machine learning models to generate the one or more responses of the AI character based on the portion of the plurality of data selected from the world knowledge information;
generating, by the processor, using the second plurality of primary machine learning models, the one or more responses of the AI character based on the portion of the plurality of data selected from the world knowledge information, the personal knowledge information, and the context parameters produced by the first plurality of heuristics machine learning models;
monitoring, by the processor, changes in the context parameters;
detecting, by the processor, transitioning from the current time period to a further time period of the plurality of time periods associated with the storyline; and
dynamically adjusting, by the processor, the scope of knowledge based on the changes in the context parameters and the transitioning to the further time period.
|