Ben-Gurion University of the Negev AI researcher published their Pentagon-funded study, unveiling an AI tool capable of analyzing text and detecting violations of social norms using modern computational tools.
- The AI tool utilizes OpenAI’s GPT-3 and zero-shot text classification to identify “social norm violations” in blocks of text.
- Promising results show the tool effectively detects violations across ten social norm types, such as competence, trust, and loyalty.
May 19th marked an important date in AI history as Ben-Gurion University of the Negev AI researchers published its Pentagon’s Defense Advanced Research Projects Agency (DARPA) funded study that discusses an AI tool capable of analyzing text and detecting violations of social norms. using modern computational tools.
Using advanced language models and data parsing techniques, the tool has the potential to offer valuable insights into human behavior and decision-making.
According to the abstract, the AI tool leverages GPT-3, a programmable language model developed by OpenAI, to analyze blocks of text and identify potential “social norm violations.” The AI researcher also employed zero-shot text classification to categorize the violations and focused on two key social emotions: guilt and shame. By training predictive models on synthetic data generated by GPT-3, the researchers aimed to automatically identify norm violations in text histories.
The results are promising in detecting violations across ten different social norm types: competence, politeness, trust, discipline, caring, agreeableness, success, conformity, decency, and loyalty.
The system can analyze scenarios described in the texts and determine whether a social norm has been transgressed or adhered to. For instance, the tool identified shame in a text about a couple’s infidelity, while it recognized adherence to a social norm of academic success in a text describing scholastic achievements.
One of the challenges in using AI to identify social norm violations lies in the diversity of cultural norms worldwide. Norms differ significantly across cultures, making it challenging to create a universal standard for norm violations. While the AI researchers claim that human responses to norm violations might be consistent despite cultural differences, there is still a risk of imposing Western-centric perspectives on other societies.
The association of this research with DARPA raises concerns about privacy and surveillance. The potential application of this technology by military organizations to analyze foreign populations is unsettling, as it could infringe on privacy and civil liberties. Moreover, AI-powered norm detection in private communications could raise ethical questions regarding consent and the right to free expression.
The application of AI to detect social norm violations in texts raises ethical concerns, particularly regarding privacy, cultural sensitivity, and potential misuse. Norms can vary significantly across societies, and an AI system may not adequately account for cultural differences, leading to misinterpretations and biases. The use of such technology for surveillance or in conflict scenarios could also have alarming consequences, impacting individual freedoms and rights.
Further research and discussions are essential to ensure responsible and ethical AI applications that respect human rights and cultural diversity.
Inside Telecom provides you with an extensive list of content covering all aspects of the tech industry. Keep an eye on our Intelligent Tech sections to stay informed and up-to-date with our daily articles.