A crucial experience during my time at university— computer science (with focus on AI) and linguistics—was the documentary “Maschinenträume” (1988) by Peter Krieg. It features the long-term AI project “Cyc,” in which Doug Lenat and his team try to represent common sense knowledge in a computer. When Cyc started, in 1984, it was already known that many AI projects failed due to the machine’s lack of common sense knowledge. Common sense knowledge includes, for example, that two things cannot be in the same place at the same time, or that people die, or what happens at a children’s birthday party. During the night, while the researchers are sleeping, Cyc tries to create new knowledge from its programmed facts and rules. One morning the researchers were surprised by one of Cyc’s new findings: “Most people are famous.” Well, this was simply a result of the researchers having entered, besides themselves, only celebrities like, for example, Einstein, Gandhi, and the U.S. presidents. The machine-dreaming researchers, however, were in no way despondent about this obviously wrong finding, because they figured they would only have to enter the rest of the population, too. The underlying principle behind this thought is that it is possible to model the whole world in the form of ontologies. The meaning of the world can be captured in its entirety in the computer. From that moment the computer can know everything that humans know and can produce unlimited new insights. At the end of the film Peter Krieg nevertheless asks: “If one day the knowledge of the whole world is represented in a machine, what can humans do with it, the machine having never seen the world.”
Discovering Interpretable Topics by Leveraging Common Sense Knowledge
