This short guide outlines important safety and security practices for users of Cogniti to create, manage, and use generative AI in educational settings.
Authentication and personal information #
Cogniti does not store any passwords. It uses third party authentication services (such as Microsoft and Google through OAuth 2.0) to log you in. These services provide your name and email address to Cogniti so that you can be given access to the right areas of the platform. You must ensure that you keep your account secure.
More information can be found in our Privacy Policy and Terms of Service.
AI security and safety #
When using the OpenAI AI models like GPT-4, we recommend using the Microsoft Azure hosted versions because they are protected by Microsoft’s Azure AI Content Safety. This protects prompts and completions from offensive or inappropriate content.
We have also provided recommendations on the educational guardrails that you can provide to your AI agents within Cogniti.
AI output verification and evaluation #
An important aspect of AI literacy for teachers and students is to appreciate that the current generation of AI does ‘hallucinate’ at times, which means that it can output incorrect information.
We have provided recommendations to reduce hallucinations, and also recommend that teachers regularly inspect the conversation histories for their agents to verify and evaluate AI outputs.
AI transparency #
All users are informed that they are working with a generative artificial intelligence through a note adjacent to user input areas. It is critical for human users to be aware that they are interacting with AI.
Student safety and wellbeing #
As teachers and educational institutions, student safety and wellbeing are paramount. We strongly recommend that you instil in students key tenets of AI literacy including a functional understanding of AI hallucinations, bias, and human agency. The UNESCO AI Competencies for Students is a good resource to consider.
We recommend that teachers regularly inspect the conversation histories for their agents to review messages between their students and AI.
Follow local policies #
Your institution and jurisdiction will likely have local policies, guidelines, and other rules that specify how you should (and should not) engage with generative AI. Ensure you follow these when using Cogniti. This may include not entering sensitive information into generative AI tools including Cogniti.
Data that is provided to Cogniti is handled according to our Privacy Policy and Terms of Service. Your data is not used to train any AI models.
