Contents
- General
-
Security, Ethics and Bias
- What data is passed in prompts to AI services? Is any personal or commercially sensitive data involved?
- Where is data passed to AI services processed, how is data transferred securely to these services and how do those services store data?
- How is the security of the AI services monitored and managed? E.g. do you test for prompt injection vulnerabilities?
- How was your AI model trained, and what body of data was used?
- How do you address potential inaccuracy in results produced by your AI solution? Is this made plain to users?
- How do you know that the analyses AI Insight provides are valid, accurate, and unbiased?
- Will my data be secure and private if I use AI Insight?
- Is the video data submitted by the schools used in any way to further train or enhance your AI models?
- How do you deal with potential bias in feedback?
General
-
What is AI Insight?
AI Insight is a new feature from IRIS Connect that uses artificial intelligence to analyse classroom videos, providing teachers with immediate, actionable feedback to enhance professional development
-
Can you explain more about AI Insights?
IRIS Connect's AI Insight streamlines professional development, allowing teachers, coaches, and administrators to focus on what truly matters: impactful teaching, meaningful collaboration, and continuous growth. By quickly analysing classroom videos, it delivers immediate, actionable feedback, saving valuable time and speeding up the professional development process.
When used in Pathways, AI Insight enables automated, adaptive professional development. Pathways guide teachers through reflection, collaboration, and coaching processes. When integrated into a Pathway, AI Insight tailors recommendations based on classroom teaching analysis, delivering personalised, research-based guidance within the Pathway. This level of customisation, which traditionally required significant time from coaches, administrators, and colleagues, is now available to all teachers.
-
Why did you decide to build AI Insights?
Teachers, coaches, and administrators are often stretched thin, with limited time to spend on crucial human interactions. By using AI to automate the analysis of classroom videos, we aimed to streamline this process, providing immediate, actionable feedback that saves time and accelerates professional growth. Our goal was to make high-quality, personalised professional development accessible and scalable. Traditionally, this requires significant human capital, making it costly and time-consuming.
-
How does AI Insight work?
AI Insight quickly reviews classroom videos and identifies key aspects of teaching and learning. It then provides feedback and recommendations based on this analysis, helping educators improve their instructional practices. See user guide here
-
What kind of feedback does AI Insight provide?
AI Insight offers actionable feedback on various aspects of teaching, such as classroom management, instructional techniques, and student engagement. The feedback is designed to be practical and immediately useful.
-
What are Pathways, and how does AI Insight integrate with them?
Pathways are structured programs within IRIS Connect that guide teachers through reflection, collaboration, and coaching. When integrated with Pathways, AI Insight tailors recommendations based on classroom analysis, offering personalised, research-based strategies for professional development.
-
Is AI Insight customisable?
Yes, AI Insights can be customised to align with specific priorities or programmes. If you’re interested in exploring customisation, please contact cpd@irisconnect.co.uk.
-
How will AI Insight save me time?
AI Insights help to save time by providing immediate feedback and reflections on classroom practice. This supports deeper self-reflection and helps enhance coaching and professional learning conversations, maximising the time teachers and coaches have together.
Security, Ethics and Bias
See this article for AI Security Overview and opt-out
What data is passed in prompts to AI services? Is any personal or commercially sensitive data involved?
Video transcripts. These may contain PII such as a persons name
Where is data passed to AI services processed, how is data transferred securely to these services and how do those services store data?
Where - GPT: US, Azure: Sweden
How: API
Storage: Data is not stored with the AI service
How is the security of the AI services monitored and managed? E.g. do you test for prompt injection vulnerabilities?
Prompts are hard coded within our system, users are only able to provide additional context and questions which avoids prompt injection vulnerabilities.
Security is managed by OpenAI/Azure
How was your AI model trained, and what body of data was used?
See OpenAI/Azure documentation
How do you address potential inaccuracy in results produced by your AI solution? Is this made plain to users?
The following information is displayed to users using our AI beta feature
Welcome to the AI BETA:
Use AI Insights to submit Reflections for AI completion.
The AI only gets to see video transcripts, no images of your environment or the individuals captured are processed.
Your transcript data is not stored with the AI processor in the long term and is not used to train the AI.
Only share a Reflection to an AI Insight if you are confident that it does not contain inappropriate personal data.
If you share a Reflection with another user, they will be able to submit it to Insights.
If another user has access to a Reflection you've shared and uses the same Insight, they will be able to view the Insights response to that Reflection.
You can disable Insight mode in your Reflection's privacy settings to prevent other users from submitting the Reflection to an Insight or viewing related responses.
Remember, as a BETA feature, it's likely that you will experience inconsistencies. Please share your feedback.
-
How do you know that the analyses AI Insight provides are valid, accurate, and unbiased?
AI Insight leverages advanced large language models (LLMs) developed by OpenAI to analyse classroom video transcripts. These models are designed to understand and generate human-like text based on the data they are trained on. OpenAI's technology is continuously refined to ensure accuracy and reliability. We have conducted extensive internal testing to validate the analyses provided by AI Insight. Our team has carefully crafted the AI queries to ensure that the feedback and recommendations are valid, accurate, and reliable. Our priority is to "do no harm," so we pace our feature development carefully to ensure we are providing valuable and trustworthy information to educators. Additionally, we are actively engaging with educators to test new Insights and gather feedback. This collaboration helps us refine AI Insight further and ensure it meets the high standards of quality and usefulness that educators need. Through this ongoing process, we aim to provide tools that genuinely enhance teaching and learning.
Will my data be secure and private if I use AI Insight?
Yes, IRIS Connect takes the responsibility of acting as a processor of personal data extremely seriously. AI Insight operates within the secure environment of IRIS Connect, adhering to strict data protection standards. Additionally, before analysing the videos, AI Insight attempts to remove any personally identifiable information to further protect your privacy. We constantly review our procedural, organisational, and physical security to ensure that we are offering a service that our customers can trust. We have completed multiple external audits, gap analysis, and penetration testing of our services to ensure that we meet industry best practices and the principles of GDPR.
Is the video data submitted by the schools used in any way to further train or enhance your AI models?
We prioritise the privacy and security of your data. Video data submitted by schools is never used to train or enhance our AI models. We respect the confidentiality and ownership of your content.
By incorporating these measures and policies, we aim to provide reliable, fair, and respectful AI-driven insights for all our users.
How do you deal with potential bias in feedback?
At IRIS Connect, we understand the importance of fair and unbiased feedback. Our AI Insights tool uses pre-trained models from OpenAI, which have been trained on vast and diverse datasets. While it's challenging to eliminate all biases from large language models (LLMs), OpenAI has implemented extensive measures to ensure their training data is balanced across various demographics, including race, gender, age, ethnicity, and socioeconomic status.
We want to assure our users that:
Diverse and Representative Training: OpenAI's training data encompasses a broad spectrum of perspectives to mitigate bias.
Additional Safeguards: Both OpenAI and IRIS Connect have added layers of prompt architectures to further reduce potential biases.
Transcript-Only Analysis: The AI Insights tool only analyses the transcript of audio, without explicit information about speakers' personal characteristics. While certain inferences might be made from the text, no additional external data about the speaker is provided to the model.
Bias Considerations:
Western and Liberal Bias: It is acknowledged in research that OpenAI models may exhibit a Western cultural and liberal political bias. We are continuously working to address and mitigate this.
Expert Validation: Our prebuilt AI Insights have been rigorously tested by educators to ensure consistent and expert-aligned feedback. These insights focus on factual, evidence-based questions and provide minimal, respectful qualitative feedback.
If you have any further questions or concerns, please feel free to contact our support team.