How can we ensure that generative artificial intelligence is genuinely useful to teachers, leaders and school administrators and not just an overhyped fad?
The Department for Education (DfE) in England are currently consulting on this very topic (here) and we look forward to the thoughts of the industry. We hope that a detailed position paper such as the one from the Unites States Office of Educational Technology will be produced, with a UK context, to help steer future development and keep safeguarding, privacy and ethics to the fore. We’d encourage anyone with an interest to read the report ‘Artificial Intelligence and the Future of Teaching and Learning’.
Here at Bromcom, we believe that the benefits of AI are amplified when applied to the data that you already have in your school MIS (introspection) and combined with the training data and large language models hosted externally (extrospection.)
Phase 2 of the Bromcom AI that we demonstrated at the Shard event in June successfully, demonstrated the concept when applied to data held in the MIS on students and their attendance. Questions such as ‘Who are my lowest attenders in year 6 who have special educational needs?’ are powerful because they can be expressed in natural language rather than through reports, dashboards and filters. Questions such as ‘Which students with below 80% attendance in year 5 are likely to have low attendance next year in year 6?’ take the pattern matching and analysis capabilities of the AI to facilitate forecasting and therefore early intervention or preventative work. The next question ‘Which strategies to improve attendance are most likely to be successful with this cohort’ then takes this to the next level. Nothing startling has been recommended by the AI, but all useful in reminding us of options without having to consult the research or a tick list.
Significant technical design work has taken place to ensure that Bromcom AI is secure and appropriate for school use. Firstly, the Microsoft Azure Open AI does not use any input data to retrain the language models, has enhanced security and filters content and monitors for abuse. Secondly, no personal or private information is sent to Chat GPT. In simple terms, the language model in Chat GPT is being used to interpret the prompt given by the user (the intent) and turn this into queries that are run entirely within the Bromcom database, when school data is being accessed.
We are excited about Phase 3 that takes the data domains that can be interrogated by Bromcom AI and adds in the ability to ask questions on student behaviour, assessment data, assignments, interventions and communication. We are also adding in the ability to analyse information relating to staff such as timetable, cover, attendance and performance management. The Large Language Models are particularly sensitive to how the question is phrased, so we are working on prompt engineering and building a set of prompts that will produce useful and reliable results. The possibilities for exploring interrelationships between student attendance, behaviour and academic outcomes will be fascinating.
Phase 4 will embed the power of Bromcom AI at key parts of the workflow. Rather than having to bring up the chatbot window, this will be integrated into regular tasks. This has the benefit of time saving but also hiding the complexity behind prompt engineering, as the AI simply provides support to the user in completing routine tasks.
Our first example is to take the letter writing abilities of Chat GPT and integrate this into the workflow of sending a quick message to parents. This combines the existing templates with content that is generated by AI and can be edited, regenerated and tweaked before sending to parents. Additional features such as expand or shorten the letter, make tone more formal or translate into different languages are all single clicks of a button – Click here to try it out!
A second example is to save teachers time in creating self-marking quizzes and assignments on any topic. The AI can quickly generate a range of questions on any given topic and supply answers, and again, questions could be simplified or made more challenging with single button clicks. Bringing this back into the quiz editor prepopulated the questions and answers, and the teacher can change the order, edit the questions or add an external resource before assigning to students – Click here to try it out!
We have launched a user focus group on Bromcom AI to help to steer the future development of the product and ensure that it meets customer needs. Please get in touch with us if you would like to join this group or have suggestions for future developments, via the following email address – BromAIfeedback@bromcom.com
*You can watch Fergal’s full Bromcom AI: Phase 2 presentation from our 2023 Shard event by clicking the image below*