In 2015, David Ferrucci – the award-winning artificial intelligence (AI) researcher who led the development of IBM Watson — which won the television quiz show Jeopardy in 2011 against two of the game’s top champions — noticed that most AI systems failed to understand the meaning behind language. That meant they couldn’t provide rich, reasoned explanations for any output.

That’s when Ferrucci founded New York City-based AI research and technology company Elemental Cognition, to tackle one of the most difficult challenges facing the future of AI: Developing the ability to reason and understand beyond statistical machine learning and data analytics, overcome bias and provide intelligence at scale.

“Elemental Cognition’s mission is to go beyond traditional AI to assist humans in understanding complex content, reasoning about it and finding and explaining plausible answers,” Ferrucci told VentureBeat.

Now, the company is applying those efforts to cancer research. For example, last week, the company announced a partnership with the Philadelphia-based neuro-oncology research center, Penn Medicine Brain Tumor Center, to accelerate therapeutic drug development.

Seven years in the making, the company’s hybrid AI platform combines natural language understanding, machine learning, explicit knowledge models and automated reasoning to enable a new class of AI applications capable of learning and delivering explainable intelligence. What makes the platform interesting is its three-pronged approach to doing this.

From IBM Watson to CORA

First is Cogent, Elemental Cognition’s expert Engagement solution that captures the knowledge essential for powering expert applications and allows users to share, maintain and extend expertise in natural language.

Then, the Cordial engine introduces “reasoning” to the platform. The scalable APIs can empower chatbots, websites, or applications with interactive problem-solving intelligence and deliver expert interactions at a fraction of the cost. This can help users customize and evaluate alternatives, allocate resources, optimize schedules, diagnose failures, apply and explain policies and more — allowing them to manage overwhelming situations in real time.

The company’s Collaborative Research Assistant (CORA) is a software-as-a-service (SaaS) application built on a hybrid AI platform that aims to accelerate research by getting fast, unbiased and logical answers to complex research questions. It uses natural language understanding and transparent logical reasoning to interactively help discover, evidence, and connect knowledge.

CORA supports researchers in aggregating information across various data sources by performing a semantic search over concepts and relations that are automatically induced from the data. It codifies expert knowledge, creates repeatable research templates, and develops reasoning over the data to surface novel insights, surfaces evidence to support or refute any findings, and automatically summarizes the findings logically with the evidence and references.

Though CORA can be deployed across multiple industries, it specifically adds great value to biomedical research with its ability to analyze content, identify linkages across sources, and leverage domain expertise to accelerate the discovery of life-changing translational cancer therapeutic strategies.

Natural language understanding for cancer drug development

Cancer drug development is an area with long lead times for bringing life-saving drugs to market. Traditional methods of doing pre-clinical research are inherently slow, requiring in-depth insights into complex, open-ended questions.

Manual search requires deep background knowledge and can be slow to understand and find key linkages between concepts, weigh different answers, and detect potential biases. A traditional search engine using keywords alone can be too narrowly focused, biased, misleading and incomplete. Industry experts can spend days, weeks, or months searching, navigating, curating and synthesizing evidence to generate and justify good answers.

“Understanding and resolving meaning in context and through time is the deeper challenge that results in semantically integrated content,” said Ferrucci. “CORA enables researchers to rapidly explore open-ended, complex questions, considering that the ‘facts’ are highly qualified by their context and that they evolve.”

Natural language understanding (NLU) transforms unstructured corpora into rich, semantically integrated, semistructured data sources. Then, it uses expert knowledge and logical reasoning to guide researchers to discover plausible answers that connect evidence across, not just within, sources.

CORA helps reduce bias by automatically finding both supporting and refuting evidence and tracking it over time for each link in a multistep chain of inference.

Nduka Amankulor, director of the Penn Brain Tumor Center and chief of neurosurgical oncology at Penn Medicine — along with his research team — conducted eight years of genomic research to identify novel immune adjuvants for the treatment of IDH mutated brain cancer. His team was interested in using AI to accelerate their research efforts and turned to Elemental Cognition for help.

Elemental Cognition aims to use CORA to help the PennMed team find therapeutic strategies not easily identified with currently available tools. The team’s research is focused on shortening the period between the discovery and development of therapeutics using a combination of proprietary datasets generated by Amankulor’s lab in conjunction with existing public domain research data.

For Ferrucci and Elemental Cognition, this work is just another milestone in their journey towards unraveling AI’s true potential.

“We are pursuing partnerships like PennMed where our differentiated hybrid-AI complements and partners with human intelligence to improve processes and outcomes,” Ferrucci said.

The company’s north star, he explained, is to develop AI that is transparent, interactive, and helps people solve problems and make better, more informed, and responsible decisions by scaling access to answers and the reasons why in ways that can be evidenced, understood and trusted.