| Abstract |
Once upon a time AI was (mostly) logic with a little statistics, the
latter as, e.g., probabilistic logic, bayesian networks and neural
networks. Logics provide an explainable way to describe the world around
us and reason about it. But, the world is largely not following the
ridgit rules of logic. With almost unlimited amounts of data available
and huge compute resources it became possible to build _statistical_
models that can perform tasks ranging from image recognition to POS
tagging and dependency parsing for natural languages to playing games.
These results are truly impressive.
Still, it is widely recognised that statistical methods have their
limitations that are hard to overcome. Notably, these methods require
huge amounts of data and generally cannot explain their conclusion. It
is also common for such models to pick up signals from unintended,
unwanted or unethical biases. At least in theory, statistical models can
profit from rule based reasoning in reducing their need for data, avoid
conclusions that simply can not be true or explain their conclusions.
I have been working in Logic Programming most of my carreer as the lead
developer of SWI-Prolog. Notably during the last year SWI-Prolog has
greatly improved in supporting truly declarative logic based reasoning.
During this WAI talk I want to outline the Logic Programming landscape
and the position SWI-Prolog takes in this landscape. My aim is to seek
opportunities for closer cooperation around these topics.
Link to slide: https://docs.google.com/presentation/d/17WZ9yZ-eBSMj-LZ8ClO1WHmi8TpqY40iPLWcOLu2TQI/edit?usp=sharing |