AI Transparency Notice
AI Compass for Nonprofits™
Visit: ai4sp.org/transparency for details.
This notice provides transparent information about the AI system you are using, outlining the system’s capabilities, limitations, biases, data usage practices, and safety measures. We want to empower you with the knowledge needed to understand how this AI works, what it can and cannot do, and how your data is handled.
By providing this transparency, we enable you to make informed decisions about your interactions with our AI technology.
Intended Use and AI Features
The AI Compass for Nonprofits™, created and maintained by AI4SP, uses Artificial Intelligence (AI) to generate personalized reports with recommendations to help nonprofits, associations, and foundations start their AI journey. It analyzes user inputs to benchmark organizations against 15,000 others and provides practical ideas for implementing AI solutions.
AI Used
- Baseline Model: We use GPT-4o from OpenAI or Claude Sonnet 3.5 from Anthropic.
- Our Application: We’ve built upon these base models to create AI Compass for Nonprofits™.
How the AI is Trained
Baseline Model
GPT-4o and Claude Sonnet 3.5 are trained on large amounts of publicly available data from the Internet. For detailed information on how these models were created, please visit the official websites of OpenAI and Anthropic. It’s important to note that there are ongoing discussions about data sourcing practices and potential copyright implications of using publicly accessible data for training large language models.
Our Application
We fine-tuned the model using proprietary insights from 15,000 organizations on their technology use and AI adoption levels. We also combine insights from our analysis of 10,000 Generative AI Tools in our private database. We also use publicly available data from leading tech companies about the benefits of their software and AI features. Our AI training never uses or accesses Personally Identifiable Information (PII).
Data Usage
| Aspect | Details |
|---|---|
| Use of Personal Identifiable Information (PII) | None |
| PII security | Not Applicable (not collected) |
| Data used for AI Training of our models | YES (aggregated data, non-PII) |
| Data shared with AI providers and used by them for training their models | NO |
Notes:
- No personally identifiable information (PII) is collected or used. Our app uses only drop-down menus and sliders, preventing accidental PII input.
- We collect sector, employee count, and self-assessment responses; responses are used for report generation and cannot be tied to individuals.
Known Biases
Potentially Inherited Biases
Our solution may inherit the following biases from its baseline model GPT-4o and Claude Sonnet 3.5:
- Language bias favoring English-language content and Western perspectives.
- Demographic bias reflects the data available on the internet.
- Temporal bias due to the knowledge cutoff date of the baseline models.
- Representation bias in business practices, potentially favoring larger or tech-centric companies.
- Geographic bias, possibly overrepresenting practices from regions with high internet usage.
- Industry-specific bias, potentially favoring sectors with more online presence or documentation.
Inherited Biases Mitigation Efforts
We have taken steps to mitigate many of these biases:
- Our proprietary dataset includes information from 15,000 organizations and an analysis of 10,000 AI tools from global innovators.
- We have a balanced sample of organizations of all sizes across over 15 industries in the private sector and 17 in the nonprofit sector.
- Our geographic coverage, while still evolving, includes data from Australia, Brazil, Canada, Colombia, France, Germany, India, Mexico, Netherlands, New Zealand, South Africa, Spain, the UK, and the US.
Known Biases Specific to Our Application
- Potential bias towards common AI adoption patterns in surveyed organizations.
- Possible recency bias in AI technology recommendations.
We continuously work to identify and further mitigate these biases in our application.
Limitations
Our AI has some constraints:
- Scope: Designed to generate AI adoption recommendations for nonprofits, associations, and foundations. By design, this tool cannot offer any other type of advice.
- Knowledge cutoff: The model’s knowledge was updated on Jul 15, 2024.
- Context: It may not fully capture the unique nuances of your specific business situation as the model relies only on sector and company size as references to personalized advice.
The AI Compass for Nonprofits™ provides general guidance and should be a starting point for your AI strategy, not a definitive plan.
Hallucinations & Accuracy
While generally reliable, our AI can sometimes produce incorrect or inconsistent information, a phenomenon often called “hallucinations” in AI:
- It might generate plausible-sounding but inaccurate recommendations.
- Outputs may not always reflect current AI technologies or business practices.
- Recommendations may be influenced by common practices in surveyed organizations.
We have mitigated hallucinations with our proprietary and patent-pending methodology.
Human Oversight
Our team regularly reviews and updates the AI’s recommendations based on real-world feedback and new developments in AI technology. We conduct regular focus groups and gather input at workshops, keynotes, and events where we collect direct feedback from users of our tool. This continuous engagement with our user community ensures that our AI system remains relevant, accurate, and aligned with the evolving needs of businesses across various sectors.
User Actions
- Interpreting Biases: Consider that recommendations may be influenced by common practices in surveyed organizations and may not perfectly fit your unique situation.
- Seeking Human Review: For critical decisions about AI adoption, we recommend consulting with AI experts or our team for personalized guidance.
Continuous Improvement
Our AI model is regularly updated to enhance performance and address issues. We update this notice when significant changes occur.
User Feedback
To provide feedback, click on the following link: ai4sp.org/#contacto