Prof. Yair Neuman from Ben Gurion University accepted DARPA's challenge and succeeded in developing an artificial intelligence that would recognize social norms and the Parthen. The research findings were published in the prestigious journal Scientific Reports
When the US Secretary of State, Anthony Blinken, met in 2021 with his Chinese counterpart in Alaska, there was a tense atmosphere. This tension can be attributed both to the strategic conflict between the superpowers, but also to the cultural gaps and the understanding of social norms. The American government has come to realize that failures in intercultural communication are an obstacle to achieving the goals of the United States and may even become a threat of war.
To this end, a project was launched within the framework of the American Defense Research Agency (DARPA), which aims to develop automatic tools for cultural understanding. Prof. Yair Neuman from the university responded to the challenge and succeeded in developing an artificial intelligence that would recognize social norms and understand them. The research findings were published in the prestigious journal Scientific Reports.
Social norms define standards of social behavior and these vary as a function of culture and context. For example, in Chinese culture humility is expected in interpersonal interactions, which may be perceived in other cultures as a lack of self-confidence and even a violation of a proper social norm.
To teach the computer to recognize social norms and break them, he presented Prof. Yair Neuman from the department of cognitive and brain sciences at the university, about 5000 social situations before the artificial intelligence and asked it to explain which norm was violated and why. The explanations of the artificial intelligence were analyzed and the insights were fed into another artificial intelligence system through which rules were automatically identified (automatic rule discovery) that make it possible to distinguish between situations of norm violation and confirmation.
The heart of the system is in the automatic recognition of social emotions. Prof. Neuman Shaar that although there is a huge number of norms and that they change in different cultures, the violation of a social norm is associated with feelings of shame, guilt and embarrassment, while obedience to a social norm is associated with feelings of pride and gratitude. Therefore, the system relies on the automatic detection of social emotions related to the violation of social norms or their confirmation.
The system was tested on 24,000 cases, half of which involved violating a social norm. When the classification ability of the artificial intelligence system was tested, it became clear that it successfully identified 82% of the norm violation cases. In a second experiment, the system was tested on another data file, close to 12,000 situations, half of which involve non-normative activity. In this case the system was able to identify with 89% of success what type of situation it is.
"The algorithm we developed knows how to derive emotions within situations through the use of new models of language processing. Artificial intelligence today knows how to analyze things in a fast and effective way that its performance does not fall short of human performance. These new tools are so powerful that we do not know how to sufficiently appreciate the magnitude of the revolution", Prof. Neumann concluded.
The research group included: Prof. Yair Neuman and engineer Yohai Cohen.
The research was supported by the Defense Advanced Research Projects Agency (DARPA). The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.
More of the topic in Hayadan: