Arguments for Understanding our Complex World: Integrating Knowledge and Data Technology in Artificial Intelligence

2^nd November 2016, 13:00 Ashton Lecture Theater
Dr. Bart Verheij
Artificial Intelligence, Rijksuniversiteit Groningen
Bernoulliborg
Postbus 407
9700 AK Groningen
The Netherlands

Abstract

Artificial Intelligence (AI) is currently in the center of attention, by high-profile applications in autonomous driving, big data analytics, and expert-level game play. These appealing examples of AI inspire dreams about the coming age of humanoid robots, and simultaneously lead to concerns about autonomous warfare, the demise of privacy, and human-robot competition in the labour market.

In this talk, I contend that realizing these dreams and countering these concerns require the same innovation in AI: the development of argumentation technology. Argumentation is a successful human activity: whether in science, in politics or in everyday life, our deepest insights have required extensive debates about different hypothetical options. In recent years, a lively community of AI researchers has started the development of argumentation technology. I claim that argumentation technology can integrate two successful sides of AI: knowledge technology, founded on formal logic, and data technology, founded on probability theory. Such an integration is long sought for, and a prerequisite for the computer-supported understanding of the complex world we live in.

As an example, I present results in an NWO Forensic Science project that I am leading. In criminal investigation, it proves hard to combine statistical information, for instance DNA evidence, with qualitative information, such as witness testimony. In the project, techniques have been developed integrating probabilistic and logical modeling tools (Bayesian networks, scenario analysis, formal argumentation). I explain how the lessons of this project pave the way to new argumentation-driven AI integrating knowledge and data technology.

Once robots can argue about the balanced reasonable choices that we know are needed for human-level skills, they become much less dangerous---not more than the rest of us.