Prof John Domingue

AIEd researchers have for decades been motivated by Bloom’s famous 1984 paper which found that students taught 1-to-1 performed 2 standard deviations better than students taught in a standard classroom setting, with the average 1-to-1 student performing better than 98% of the students taught in the control group. AI offers the potential to unlock low-cost personalised teaching to massively assist both students and teachers and dramatically improve learning outcomes. The advent of Generative AI has brought this dream closer and has captured the attention of many educators. For example, a recent study of more than 1,000 K-12 teachers and 1,000 students found that 51% of teachers reported using ChatGPT - 40% weekly. Interest and take-up is similar in the Higher Education Arena. For example, an analysis of eight Russell Group universities found that there were over 1 million visits to the ChatGPT site during December 2022 and January 2023. All of the above use is via a platform provided by a US company which has not been transparent on how the underlying Large Language Model was trained and for which there are a number of issues including: • Misinformation - responses may not be truthful. • Copyright - text may be generated which is similar to existing copyrighted content. • Bias - bias in training data may mean that responses replicate existing biases and prejudices. In SAGE-RAI we will build upon a number of existing proof of concept and pilot education-oriented Generative AI tools, created by the partners, to target Bloom’s 2 Sigma problem through Responsible AI whilst also addressing the issues raised above. In particular we will build a platform that can support assessment and provide student guidance through Responsible Generative AI.

tbc

STFC Centre for Doctoral Training in Data Intensive Science

A major obstacle to increasing the efficiency, effectiveness and quality of education in Europe is the lack of widely available, accessible, multilingual, timely, engaging and high-quality educational material (i.e. OpenCourseWare). The creation and maintenance of comprehensive OpenCourseWare (OCW) is tedious, time-consuming and expensive, with the effect, that often courseware employed by teachers, instructors and professors is incomplete, outdated, dull and inaccessible to those with disabilities. By using the open-source SlideWiki platform (available at SlideWiki.org) the effort of the creation, translation and evolution of highly-structured remixable OCW can be widely shared (i.e. crowdsourced). In a similar way to Wikipedia for encyclopaedic content, SlideWiki allows users to: 1. collaboratively create comprehensive OCW (curricula, slide presentations, self-assessment tests, illustrations etc.) online in a crowdsourcing manner, 2. semi-automatically translate this content into more than 50 different languages and to improve the translations in a collaborative manner and 3. support engagement and social networking of educators and learners around that content.