Case study: optimising school outcomes

This case study is based on the third case from the Princeton Dialogues on AI and Ethics Case Studies. It was initially translated to Dutch and has now been translated back to English.

1. BACKGROUND INFORMATION

OS Minerva high school has achieved a depressing record: it has the highest dropout rate in the country. 9% of its students leave school without a diploma! Furthermore, only 55% of its students graduate without repeating a year. The education inspectorate demands improvements and threatens to close the school if results do not improve. However, the school board sees no easy solutions. The students do not seem motivated, and the teachers are also reluctant to enter a compulsory improvement process after all the negative assessments.

The rector of OS Minerva, Ms Vulcani, starts looking for a solution and, in consultation with the inspectorate, a plan emerges to use the data the school has about students. OS Minerva collects data on students' academic performance, of course, but in addition to this, every student also has a pass for the library and canteen and the use of the Wi-Fi network is also tracked. Ms Vulcani wonders if it might be possible to use this data to help students and to improve school results.

Hephaestats, a local company, is asked to start analysing the data. Ms Vulcani, together with Hephaestats, sets some goals for the project and the inspectorate and the rector agree. The students' data can be shared with Hephaestats.

2. WHAT ROLE DOES AI PLAY IN THIS CASE? AND HOW DOES THIS AI WORK?

Hephaestats uses AI techniques to predict the risk of dropping out of school. Examples of the information include demographic information (gender, address, home situation, ethnicity), academic information (grades, averages, attendance, detention) and teachers (diplomas, amount of failing grades, years of experience). Hephaestats uses this data to create a new, derived dataset that has similar statistical properties to the original data. The dataset is no longer linked to specific students.

Teachers are shown the profiles of at-risk pupils, along with recommendations on how to help each specific pupil best. Teachers who follow the recommendations see an immediate improvement in pupils' results. The school board uses an overview of the recommendations to make changes to timetables and the canteen.

After a few years, OS Minerva's results have improved tremendously. 85% of students now graduate without repeating a year and only 5% leave school without a diploma.

3. IN WHAT WAY WERE AGREEMENTS MADE BETWEEN THE RELEVANT PARTIES IN THIS CASE STUDY?

Between OS Minerva and Hephaestats, the following goals were established for the project:

• Identify predictive factors for distraction as indicators of a school dropout risk. Apply these factors to identify pupils at risk.
• Provide teachers with detailed information in order to target at-risk pupils with specific interventions, such as talking, a modified schedule or a conversation with their parents.
• Be transparent in the use of the system.

However, the school does not communicate its collaboration with Hephaestats to parents and students. Indeed, the rector claims that the parents' council never agrees with the board, but the problem is too urgent to do nothing. Furthermore, the rector feels that this project is within the school board's mandate.

4. WHAT ARE THE RISKS INVOLVED IN THIS CASE?

• Privacy

Data is analysed from the pupils of a secondary school, almost all of whom will be minors. This analysis will be carried out without the consent of the parents or children, and they will also not be informed about the project until after it is completed. Schools have a legal accountability regarding the processing of their pupils' personal data.

• Dual interests

While it is undoubtedly better for students to finish high school with a diploma, this project has been started not only for the students but also for the school's self-interest. Furthermore, decisions are taken without parental or pupil participation.

• Test subjects

OS Minerva students feel like test subjects. They are only being helped with this system for the purpose of improving the results of the school.

• Transparency

The school's approach is not transparent. It is not clear how the risk profiles are created, and the source code is not public.

• Board-teacher relationship

Some of the teachers are frustrated that in previous years their recommendations to help students were not listened to, and now the school board is listening because an AI system makes the recommendations. Hephaestats is now getting all the praise while teachers are doing the heavy lifting.

• "Artificial Intelligence"

One of the mathematics teachers argues in an opinion piece that Hephaestats' AI system is just a matter of statistics sold in a pretty package. She warns against the tendency to blindly rely on this system because it would not be future proof.

5. HOW ARE THESE RISKS MITIGATED IN THIS CASE STUDY?

• Privacy

Although pupils' data are anonymised for the analysis process, they are still processed when determining risk profiles. The school cites its legal duty to provide good education as the basis for the processing, but it would be much better if they had sought permission from the participation council for this. After all, it is not a given that the school's duty to provide good education is enough to process its students' data in this way.

• Dual interests

By not informing parents and pupils, the school has not been able to dispel the idea that they are doing this project out of self-interest. If they had involved parents and students earlier this would not have been a problem.

• Test subjects

Due to OS Minerva's lack of information about the approach, students are treated not as an interest group but more like test subjects. They are assessed without their knowledge and although they may benefit themselves, their fundamental rights are not taken into account.

• Transparency

The school's approach is not transparent. It should have stipulated with Hephaestats that the source code would be made public. Hephaestats, after receiving due public criticism, did create a programme that allows students to view the influence of individual variables. Besides technical transparency, a school must also be transparent about the protection of its students' personal data, which was not done here.

• Board-teacher relationship

Although the changes have a positive impact on students, it is dangerous to rely blindly on an AI system. It is a well-known phenomenon that results can change just by the focus of measurement. While there are indeed positive results for students, this may also be because there is finally support from the board for major changes.

• "Artificial Intelligence"

Since Hephaestats will not release its source code, it is not possible to verify how the results and recommendations are created. This also makes it impossible to check how sophisticated the techniques are. Nor is it a foregone conclusion that a school should be allowed to unleash an AI system on its students, even if the education inspectorate would give permission.

CONCLUSION

Although the collaboration between Minerva and Hephaestats undoubtedly helped students in achieving better school results, the school did violate the rights of its students to do so. The school board has focused on its duty to deliver education, but that must be done in a safe learning environment. By consulting with parents, pupils and teachers earlier, the project could have been equally effective while respecting pupils' rights and teachers' motivation.