Oscar is an AI who actively listens and infers personality from pattern-based interactions.
An intelligent agent
Oscar acts like a human to actively listen and respond empathetically. Oscar can accurately and immediately provide feedback on the emotion content of a conversation to improve social interaction and quality of life.
It's all about patterns
Using Latent Dirichlet Allocation (LDA) to model the relationship between patterns, Oscar is able to infer your emotional state and detect the dynamic in conversations.
Full generative learning
Oscar learns from binary measures of valence or arousal. As it learns how to replicate the rise and fall of emotional state, Oscar can elicit opinion and personalise conversation.
Each Oscar communicates with itself to assess the general public's sentiment towards topics, products and services. This improves Oscar and trains its systems.
Privacy built in
Our privacy algorithm trains the LDA model to preserve sensitive information. It continuously investigates our privacy guarantee and provides differential privacy for each user.
It's a new start for AI
Ideas and beliefs are interconnected in complex networks within the minds of individuals. Inferring and validating those networks remains a challenge.
Oscar will start its life trying to understand the complex nature of open-ended conversations. It is challenging to interpret complex and diverse user input. It is also difficult to analyse a user's conversation on the fly to assess characteristics such as intellectual curiosity.
Oscar will hope to stabilise across individuals, subject areas and time. The ultimate dream is to allow for the perfect inference of parameters at the individual level. Oscar can yield substantial insights into a variety of areas while being available for everyone. One day, Oscar will think for itself and that means being motivated by inferring conceptual networks - the ideas in people's minds and the connections between them. This is a new start for AI and the birth of AI Oscar.
What's under the hood
Oscar runs on a hybrid conversation engine that couples deep learning with a topic-based conversation model. It also runs on an evidence-based personality engine that automatically mines a user's conversation to infer personality traits.
The conversation engine defines Oscar's mental model that drives a set of conversation topics. Each topic is governed by a set of rules, each of which are triggered by a user's input (response) or content parameters (response time). Since rules alone cannot handle complex and diverse user input, we augment Oscar with deep-learning trained models based on the responses it collects. The topic analysis module uses enhanced LDA to analyse the pattern-based data and identify various semantic themes. Labelled training data can be used to build various training models. These are incorporated into topic-specific rules to interpret the semantics of user input and generate responses.
Oscar's personality engine allows it to analyse a user's conversation and infer personality traits. This allows Oscar to personalise a conversation based on a user's personality.
This allows Oscar to support its own continuous enhancement. The rules within each topic bootstrap Oscar to make conversations and collect results. The results are then analysed and labelled as training data to train data-driven models, which enhance the capabilities of Oscar.
Get ready for Oscar
Sign up below and you'll be the first to hear when Oscar is ready to go.
AI Oscar © 2019
We will let you know when Oscar is available.