Usage examples for the AI GANESH ethical framework for AIED

Context

Dean Prof. Peters is being asked by her university to endorse the use of a technology called SmarterProctoring in the faculty's courses. She is skeptical about this tool and looks up information about it online. She finds the following description on their official website:

“SmarterProctering is an automated proctoring tool which offers a flexible, secure solution that provides on-demand testing, reduces test anxiety, and addresses privacy concerns while maintaining exam integrity. It is supported by a lockdown extension, webcam recording, and screen recording for added security. Students can take exams comfortably with 24/7 support available, ensuring both convenience and security. Our trained proctors review sessions within 48 hours of exam completion and provide incident reports, ensuring easy follow-up for faculty.

We offer proctoring that saves faculty time. With the AI Monitoring feature, suspicious activity is flagged by AI, ensuring exam integrity without constant human supervision. 24/7 availability ensures that exams can be taken anytime, anywhere, offering flexibility and convenience for busy students. With our review assurance, proctors review sessions for added security, providing detailed reports on suspicious behavior.”

Prof. Peters would like to evaluate this tool using the AI GANESH ethical framework.

Example

Prof. Peters begins by reading the introduction and identifies her role as that of an 'educational institute representative'. She then heads over to the 'Using the framework' section and understands the goal of the framework and begins with Step 1 - assessing using the Rubric. She starts with the value of Goodwill.

Code	Level 1	Level 2	Level 3	NA
EN-G1		X
EN-G2		X
EN-G3	X
EN-G4		X

For EN-G1, Level 2 is selected because such a system could cause some students to get anxious, thereby affecting their performance. So, the system does not fully consider the needs of students. For EN-G2, Level 2 is selected because there is good adaptation to AIED but the review by an external proctor might undermine the role of the educator. In the case of EN-G3, the institution promotes AIED systems with good integration into the learning environment. Some points about the ethical application are not clearly outlined or considered - such as autonomy, well-being of students, data security. However, the time of the faculty and the flexibility of the students is considered. As some ethical aspects are considered while others are not, Level 2 is selected for EN-G4.

Prof. Peters then proceeds to the value of Aptness for Education

Code	Level 1	Level 2	Level 3	NA
EN-A1		X
EN-A2			X
EN-A3				X
EN-A4		X
EN-A5				X
EN-A6				X
EN-A7		X
EN-A8			X

For EN-A1, while the proctoring system does align with the learning goal, rigorous evidence about enhancement of learning goals is not provided and hence Level 2 is selected. For EN-A2, the proctoring system is very narrow in focus and only supports the traditional metric of detection of cheating - resulting in Level 3. For EN-A3, as the choice to use this proctoring system was made by the EI without consulting the educators, it is not applicable. EN-A4 is marked as Level 2 because the tools supports online assessment but the support of other learning goals is implicit. EN-A5 and EN-A6 are marked as NA because the proctoring system is not being used for learning. For EN-A7, the use of this system seems to be based on factors such as cost-cutting and easy student monitoring during exams. There is some critical evaluation but it is partial and hence it is marked as Level 2. Lastly, for EN-A8, AI literacy does not come into the picture but students are passive users of the technology - resulting in selection of Level 3.

Prof. Peters then moves forward to the value of Non-discrimination.

Code	Level 1	Level 2	Level 3	NA
EN-N1				X
EN-N2				X
EN-N3		X
EN-N4			X
EN-N5		X
EN-N6			X

EN-N1 and EN-N2 are marked as NA because information about diversity of the development team is not available and a testing tool cannot provide education-to-industry support. For EN-N3, proctors check the AI output but no information about bias is provided here, resulting in Level 2. EN-N4 is marked as Level 3 because racial biases in the webcam recording and accessibility for differently-abled students is not considered. For EN-N5, as this is a testing tool, access is provided by the institution to all students. However, issues such as limited internet connectivity or shared resources such as shared spaces are not considered - resulting in Level 2. EN-N6 is marked as Level 3 because stakeholders were not consulted by the institution before selecting this tool.

Prof. Peters then continues to the value of Explicability.

Code	Level 1	Level 2	Level 3	NA
EN-E1	X
EN-E2		X
EN-E3		X

For EN-E1, Level 1 is selected because the role of the students, teachers and the proctors are clearly defined and transparent. For EN-E2, Level 2 is selected as information about the usage of AIED tools in other domains is unavailable. However, the usage of this tool does not come with a complete evaluation. Lastly, Level 2 is also selected for EN-E3 because the developers of the system seem to have some awareness of the impact of the tool. However, they do not fully consider the anxiety and stress that it could cause students.

Prof. Peters then continues with the value of Stewardship of Data.

Code	Level 1	Level 2	Level 3	NA
EN-S1	X
EN-S2				X
EN-S3				X
EN-S4				X
EN-S5		X
EN-S6			X
EN-S7				X
EN-S8		X
EN-S9				X

EN-S1 is grouped as Level 1 because technical safeguards for testing are implemented in the form of limited access to data and the complete recording shows the context. For EN-S5, limited access to the recording data is provided but no additional privacy measures are mentioned - resulting in a grouping of Level 2. EN-S6 is grouped as Level 3 because the AIED system is not integrated into the educational institutions' own systems but a third party system is used without integration. For EN-S8, AIED results are verified by proctors but the accuracy rates are not mentioned across different groups and scenarios - thus grouping it as Level 2. For EN-S2, EN-S3, EN-S4, EN-S7 and EN-S9, insufficient information is available to group them into any level, thus resulting in NA.

Prof. Peters then proceeds to the value of Human oversight.

Code	Level 1	Level 2	Level 3	NA
EN-H1	X
EN-H2		X
EN-H3	X
EN-H4				X
EN-H5			X
EN-H6		X
EN-H7				X
EN-H8			X
EN-H9		X

EN-H1 is marked as Level 1 because the AIED systems' decisions are verified by impartial proctors. EN-H2 is at Level 2 as the system helps with scaling online testing but leads to isolation of students during the tests. For EN-H3, Level 1 is chosen because the responsibilities of the AIED system and other human stakeholders is clearly defined with inbuilt human oversight. EN-H4 is marked as NA because insufficient information is available. EN-H5 is classified as Level 3 because the students are passive users of the tool without any meaningful control. EN-H6 is grouped as Level 2 cause end users have some information about the proctoring system but it is not clear how the system flags instances of cheating. EN-H7 is not applicable since the proctoring system is not being used for learning and content creation. For EN-H8, students are not offered an alternative option without AI - resulting in Level 3. Lastly, EN-H9 is marked at Level 2 because consent is implied and without sufficient information.

As visible from the above use of the framework, the SmarterProctoring tool fails the ethical evaluation because it has more than 2 guidelines marked at Level 3 for 3 values - Aptness for Education, Non-discrimination and Human oversight. Prof. Peters considers the Level-1 text for these guidelines and consults the educational institution about improving on these points. Additionally, she asks for the information where it is not available and requests for those guidelines to be implemented as well.