04/24/2024 | News release | Distributed by Public on 04/24/2024 13:55
Privacy and security considerations for using Artificial Intelligence (AI) in K-12 education primarily revolve around safeguarding sensitive student data, appropriately minimizing data, and grappling with ethical considerations around data use and algorithmic decision-making.
The following is meant to serve as an initial guide for educators grappling with these issues. To aid educators who would like concrete examples of what has been done previously, our colleagues at New America's Teaching, Learning, and Tech program have compiled a repository of AI guidelines and resources intended for use in K-12 education.
As with all emerging technologies, there are more questions than answers on what responsible and ethical AI use might look like. This means that school districts and educators need to be actively engaged in shaping how they want to utilize AI. The following questions are meant to ground initial conversations and decisions around AI use:
Additionally, given the growing number of EdTech AI vendors and applications, caution is needed when deciding which applications should be used and when. Beyond ensuring vendors meet local, state, and federal student privacy requirements, it is important that schools and educators only work with third-party vendors who are responsible data stewards. The following guiding questions may be helpful in assessing vendors:
There's a need for more transparency around data collection and usage, especially when using AI, a technology that relies on large amounts of data to learn and make predictions. While school districts have long collected data to track student metrics and educational attainment, the growing use of EdTech by teachers and school administrators has led to an increase in both the type of information being collected and the number of entities that can access this data. While many EdTech tools offer significant promise, educators and school districts should carefully consider the efficacy of new tools. School districts should select third-party vendors with care, because not all vendors prioritize student privacy and data security.
School districts should involve parents/guardians and students in deciding what information can and should be collected, shared, or used by AI models, even if they are using it for educational purposes. Technology policy guidelines should be easily accessible and understandable, making it clear to parents/guardians and students exactly what information will be collected and how it will be used. Teachers can help ensure digital literacy skills by talking about data collection and usage with their students in an age appropriate manner. Additionally, there should be clear protocols around student and guardian data access, correction, and deletion.
School districts should provide educators with professional development opportunities around algorithmic bias and ethical AI use. AI models make predictions based on a large amount of data, but it is important to remember that those models are not infallible and can amplify existing harms to different communities. Because of many AI models' tendency to replicate errors in existing data and reinforce existing discriminatory assumptions or outcomes, strong caution is needed when using algorithmic decision-making. Where feasible, school districts should require algorithmic transparency from the third-party apps they use.
Finally, considerations for AI involve protecting student data from unauthorized access and malicious attacks. Schools must implement measures such as encryption, access controls, breach protocol, and regular security audits to safeguard both the AI infrastructure and the sensitive data it processes. Unfortunately, data breaches are far too common, and more data sharing means there are more opportunities for breaches. Again, this is why it is important to ensure educators are only working with trusted vendors. Where feasible, school districts should vet and set up data sharing agreements with any vendor that may receive student data.