Generative AI
Generative AI (GenAI) is a subset of Artificial Intelligence. GenAI generates new content such as text, audio, code, images, or videos based on vast amounts of “training” data, typically derived from the internet. Users are able to request and refine specific content via prompts and context setting: directions, examples, and queries submitted to the model. Such technology can enhance research, content development and design, and approaches to teaching and learning, offering new creative educational tools.
It is important to address the ethics of AI. Ethical AI means developing and deploying these technologies with a steadfast commitment to fairness, transparency, and accountability, ensuring they positively impact society. This is crucial, as biases in AI training data can inadvertently perpetuate discrimination. Whether these biases are intentional or not, these tools and their outputs require rigorous scrutiny.
Moreover, the trustworthiness of outputs from GenAI systems is a critical concern, as GenAI models produce “hallucinations,” false or misleading information that appears correct. Though hallucinating is the popularized term, GenAI tools are not sentient and cannot “imagine” things. GenAI outputs are a synthesis of the most likely words to answer the prompt, which does not always correlate to fact and GenAI systems are not necessarily designed to check the generated string of words for accuracy. As such, GenAI outputs require careful review to prevent the dissemination of misinformation. The quality of GenAI outputs is thus dependent upon the user’s expertise, AI literacy, and critical evaluation.
CPS plans to integrate a variety of GenAI tools into our daily operations. District-ready Generative AI products and tools come from either CPS internal development or external vendors, which require usage and data handling to be governed by strict contracts that align with our District’s standards and policies. Unapproved, third-party GenAI products are blocked on the CPS network to maintain our robust AI Governance, protect our District, mitigate risk, and ensure responsible adoption. For more information on available tools and privacy restrictions visit the Approved Generative AI Tools section of the website.
As CPS embraces the transformative potential of GenAI in education, we are excited about the possibilities this technology offers for enhancing teaching and learning. We recognize the importance of maintaining the integrity of our educational practices and the originality of human thought. Our intent in creating this guidance is to enable stakeholders to use GenAI to innovate and expand their capacity for teaching and learning through safe, meaningful, and responsible exploration and adoption of AI. We are committed to preparing our students for a future in which they seamlessly integrate technology and human creativity.
Models vs Tools
Key vocabulary to understand is the difference between foundational GenAI models and GenAI products and tools. Think of the GenAI product as a car, and the foundational model as the engine that powers it. Many GenAI vendors license the popular models such as Google’s Gemini, OpenAI’s ChatGPT, or Anthropic’s Claude to power their products. Even if CPS has a contract with the company that produced the foundational model being used in a GenAI product (say an EdTech tool is powered by Gemini), that doesn’t guarantee that the product has a data agreement with the model that adheres to CPS AI Governance.
For more information on data agreements, see the Approved Generative AI Tools section of this website. For more key vocabulary, check the Glossary.
Memory vs Training
A common misconception about conversations with a generative AI chatbot is how it “trains” on your data. Some people may believe that as they engage with the chatbot, it is actively learning from their answers. It is important to understand that a chatbot retaining your conversation is not the same as the GenAI model training on your data.
For example: You tell Gemini that your birthday is in March. If, in a later response or conversation, you ask Gemini where you should go to dinner and it produces a response like “Since you’re a March Aries, and Aries is a fire sign, you should try a hibachi restaurant where the food is cooked on an open flame in front of you. Here are some options in your area…” This is an example of the model referencing previous data you gave it and using it to tailor responses accordingly, often called the tool’s memory. This doesn’t impact any conversation other than your own, and if you start a new conversation window, or erase the memory, it won’t impact your future conversations either.
Let’s say you go on to respond that you don’t like hibachi. If Google saves your conversation and uses it to determine patterns that inform how the model responds to anyone in the future, this is the model training on your data. This difference is why we can have agreements with Google, and other generative AI tools, that allow you to still have memory-enabled conversations while maintaining data privacy.
For more information on data agreements, see the Approved Generative AI Tools section of this website.