New research highlights the need for inclusive design of AI-based information systems essential for users with low literacy skills.
In a rapidly evolving information search environment A.I, research from Canada’s Triangle Lab and Italy’s Università degli Studi di Milano Bicocca focused on an important issue: the accessibility of generative information systems for users with literacy challenges. This research, presented at the 2024 ACM SIGIR Conference on Human Information Interaction and Retrieval, highlights the urgency of developing inclusive AI technologies that meet the full range of literacy levels of users.
The findings point to urgent concerns within the industry. A generative model such as ChatGPT, Bing Chat, etc. generate content primarily at the university level. This unintentionally excludes important demographics who struggle with reading and comprehension. This paper by Adam Roegiest and Zuzana Pinkosova meticulously analyzes responses from popular large language models (LLMs) and reveals potential biases in training methodologies that may favor users with higher literacy skills.
The research methodology involved evaluating the readability of the generated system using a widely used command fine-tuning dataset. The data set showed that the system tended to produce sophisticated prose geared toward college-educated users, potentially leaving out those struggling with cognitive and literacy issues. The key message of this study is the call for comprehensiveness in systems that integrate generative models to make them accessible to individuals with diverse cognitive needs.
The implications of this research are critical to the AI, blockchain, and cryptocurrency industries, given the increasing reliance on AI-based interfaces for user interaction. As these technologies continue to permeate our daily lives, increasing accessibility has become not only an ethical imperative, but a business imperative. The potential of AI to revolutionize many fields is limitless, but if the literacy gap is not addressed, there is a risk that significant segments of the population will be left behind.
In response to the study, industry experts are now advocating a holistic approach to AI development. This involves designing a system with multiple “ideal” responses of varying complexity while maintaining accuracy. Companies behind leading LLMs, such as OpenAI and Google, should consider the findings in future model training and implement strategies that account for the full range of user abilities and needs.
The challenge goes beyond English to encompass a variety of language forms such as pidgins, creoles and dialects that are integral to cultural identities around the world. These linguistic variations represent more than just a communication tool. It is a fundamental aspect of people’s heritage and daily life. The findings highlight the need for generative models that accommodate these diverse linguistic expressions to ensure that users’ communication preferences are not only understood but also respected.
In conclusion, while AI and information systems have made significant progress in improving our ability to access and process information, this research serves as an important reminder of the work that still needs to be done. To build a digital environment that benefits all users equally, it is essential to build systems that are fair, accountable, transparent, secure and accessible.
Image source: Shutterstock