In this article, we discuss our overall approach to using generative AI (artificial intelligence) in the Tribal Habits platform. There are also links to dedicated articles for each AI related feature. AI features are being added to our platform regularly, so this article will be updated regularly with new links and information.
NOTE: Access to AI features is plan dependent. Some plans may not be able to access any or all AI features. The dedicated article for each AI related feature outlines the plan requirements to access that feature.
What AI platform does Tribal Habits use?
We use Amazon (Amazon Web Services) Bedrock. Amazon Bedrock is a fully managed service that provides Tribal Habits with access to multiple AI models from leading AI companies. This allows us to select different AI models for different AI features so we can maximise performance, security and cost. Our cost-optimised use of Amazon Bedrock is a key reason we can offer selected AI features even on our Starter plans at no additional cost for our customers.
What data is shared with the AI? Is our content used to train AI models?
A key advantage of Amazon Bedrock is that is offers access to popular AI models in a local, isolated manner. The AI models we access via Amazon Bedrock are:
Located with AWS in Sydney, Australia. So all data transfer occurs within Australia.
Isolated from commercial AI models. The AI model running in Bedrock is not the commercial model running from the AI provider. It is an isolated clone of the AI model running natively in Amazon Bedrock in Australia. This means no content is shared with the AI model provider (Amazon only shares aggregated usage data with each AI model provider).
Pre-trained models. This means that your content is not used to train any of the Amazon Bedrock models. In fact, any content sent to a model is not stored or retained. It is a once-off, transactional service only.
As a result, your usage of our AI features is extremely private and secure. No data storage, no content transfer outside Australia, no training of AI models from your content. Our Terms and Privacy policy reiterate our commitment to responsible and private AI usage.
What generative AI models does Tribal Habits use?
Through Amazon Bedrock, we can access models from a range of providers. We select an appropriate model for the AI feature based on performance and cost. Our choice of AI models may change over time as newer models become available, but currently include:
Amazon Polly. Polly is a text to audio conversion AI provided directly by AWS. Amazon Polly drives the text to audio conversion in our Narration, Audio and other elements.
Anthropic Claude and Haiku. The Anthropic LLMs are among the world's safest and most capable large language models (LLMs). We utilise both the Claude and Haiku models for a wide variety of content regeneration features.
How does Tribal Habits make generative AI safe to use?
In addition to our use of Amazon Bedrock to manage your content and data, we also take several other steps to improve the safety of your AI usage.
First, if desired, your organisation can completely disable the use of AI in your Tribal Habits portal. On the Admin/Account/Subscription page, Super Admins can toggle the AI features off entirely within your platform.
Second, we apply 'guardrails' on AI usage within Amazon Bedrock. These guardrails monitor prompts and responses for a variety of harmful content. If such content is detected, the guardrails may override the prompt and prevent a response from being generated (with a notification that our guardrails were triggered). We monitor guardrail usage to help fine tune the parameters for this feature. If you feel our guardrails are being triggered incorrectly, please notify us with the prompt and content that triggers the guardrails and we will investigate.
Third, our AI features are driven by your content. We do not allow for new content to be created from random prompts. Our AI focus is on "transformation" of your existing content. For example:
Generating a quiz based on your process document
Generating a summary introduction of a part of your module's content
Generating flip card content from many paragraphs of text
Generating a quote to highlight a key point from your content
And in the future, analysing your content or learner submissions to provide summary data and analysis.
At this time, we do not believe generative AI is capable of creating raw training content from a basic prompt that is reliable, engaging or useful. Such AI generated training content often suffers from incorrect facts (hallucination) or is so generic / basic that it is of little value. It often requires more time editing than it would have taken to just create content directly. Since our focus is on using your content in tightly controlled situations, this dramatically reduces the chances of AI hallucination.
Fourth, even with all these measures, our AI features still include disclaimers to remind your users to carefully check all AI generated content for accuracy and appropriateness.
Do we have to use AI features in Tribal Habits?
No. These features are entirely optional.
What AI features are currently available in Tribal Habits?
Our AI features are collectively grouped as Sage AI. Sage represents our AI 'personality', designed to help make the process of creating and managing online training approachable for any user regardless of their experience with elearning.
Our initial Sage AI features include:
Sage AI Narration. Used to generate audio from text in elements like Narration, Audio, Interact and Hotspot.
Sage AI Summary. Generates a summary (30-50 word introduction) of each point (part, step, idea, rule, fact) in your topic.
NOTE: AI Models are extensively trained on US English language data. As a result, our AI features are purely English language based at this time. In addition, while our prompts request UK English responses, the AI Models often still respond in US English. This process is not within our control so you may need to edit some US to UK English at times.
What AI features are planned for Tribal Habits?
We are currently developing a range of additional Sage AI features for release in early 2025. Please follow our Change Log for progress updates on these features.
Sage AI Transform. Sage Covert will be a new 'element' in our content creator. It will allow creators to input content from several sources and have that content automatically converted into one of our other content elements.
Sage AI Assess. Sage Assess will grade Q&A submissions based on marking guides from assessors. Sage Assess will be able to independently assess Q&A submissions or provide human assessors with an initial review of a submission to accelerate their manual grading.
Sage AI Advice. Sage Advice will represent an improved version of our existing topic analysis for content creators, providing more insightful comments and advice on content created in training modules. Sage Advice will be the next generation of help for anyone creating online training modules.
Sage AI Analyse. Sage Analyse will examine answers and responses from learners in training modules to provide creators and admins with summary analysis and trends from learners. Sage Analyse will help make sense of large volumes of training submissions and data.