Towards Human-Level Intelligence
To help AI achieve the most complex human-like tasks, we are powering advances in computer-based sensing, understanding, and action.
Mastering language is a perfect example. Human language is rich, expressive, and fraught with ambiguity and complexity. While computers have made significant progress in understanding language, they are very far from fluent. IBM Research AI is improving AI language skills via efforts in semantic representation of language, information extraction, knowledge modeling, question answering and reading comprehension, summarization, language generation, machine translation, and more.
Reasoning is yet another fundamental capability of humans. While machine learning provides a foundation for building understanding, via induction of models from data, it cannot provide deep explainability or make inferences from higher-level knowledge. We are exploring the new kinds of massive knowledge representations and algorithms which are needed to achieve this, by pushing beyond learning to reasoning.
Humans also excel at applying what they have learned in one domain to new tasks. AI also needs to leverage efficient learning techniques to support an extremely large number of tasks across a wide range of domains and industry use cases. We are exploring methods for effective and continuous capture, re-use and transfer of learned models. We are also focusing on increased exploitation of contextual information in the form of multiple modalities (vision, audio/speech, language) and end-to-end learning of complex tasks across them.