Adaptable, Stochastic and Context-aware framework for Detection of Bias, Discrimination and Stereotyping

PDF view of the paper entitled Ascend-Bds: Adaptable frame, random and context shield to detect bias, discrimination and stereotypes, by Rajiv Bahl and 13 other authors

PDF view

a summary:The rapid development of large language models (LLMS) has transformed the treatment of natural language but raises decisive concerns about the prejudices inherent in their publication and use through various linguistic, social and social contexts. This paper displays a frame called Ascend BDS (adaptable, random and tip frame to detect bias, discrimination and stereotypes). The frame shows an approach to discovering bias, discrimination and stereotypes through different groups such as sex, age, age, disability, social and economic situation, linguistic differences, etc., using an adaptive, random and shield in the context. Current frameworks are highly dependent on the use of data groups to create scenarios to detect bias, discrimination and stereotypes. Examples include data sets such as civil comments, Winno, Vinobia, Bold, Crows PAIRS and BBQ. However, such an approach provides points solutions. As a result, these data groups provide a limited number of scenarios for evaluation. The current framework overcomes this restriction by having advantages that allow adaptation, stochastic, and context awareness. Awareness of context can be allocated to any nation, culture or sub -culture (for example the unique culture of the organization). In this paper, consciousness was created in the context in the Indian context. Content from the 2011 Indian census has been used to obtain common classification denominations. Working framework has been developed using category, sub -category, stem, X factor, synonymous with enabling adaptation features, randomness and awareness of context. This framework was described in detail in Section 3. Generally 800 legs, 10 categories, 31 unique sub -categories were developed by a team of consultants at the Saint Fox Consultance Private Ltd. The concept was tested in SFCLABS as part of the product development.