The BigBrotherAward 2021 in the “Education” Category goes to Proctorio GmbH, München-Unterföhrung, for their AI-based exam software, also known as Proctorio.
The name Proctorio is derived from proctoring, which is another word for supervision or monitoring during exams. Proctorio is about online-proctoring of exam candidates at universities via the Internet.
For this business model, the outbreak of the Corona pandemic in early 2020 was an absolute stroke of luck: due to extensive lockdown-measures and the “shutdown”, lecturers had to stop giving presence exams overnight. Using public transport to arrive for a written exam was as big a health hazard as sitting in a lecture hall for hours on end. Students were worried about the continuation or conclusion of their studies, and not without reason.
In this situation, viable technical solutions for conducting online exams were desired almost as badly as an effective vaccine against the SARS-CoV-2 virus. This was the perfect time for the Proctorio software, which is claimed to offer “fully automatic and secure supervision of online exams”, and which is at the same time, according to marketing, “scaleable, cost-effective and GDPR-compliant”1. According to the company website, the software enables the cheat-protected conduct of online exams in the student’s own home with their own devices, without the need for direct human supervision and monitoring.
But at what cost?
The Proctorio software severely compromises the integrity of the students’ personal devices. In order to participate in exams, they have to install the software on their computers and surrender control of their devices to Proctorio for the duration of an exam.
Use of the Proctorio software also requires the Google Chrome browser to be installed on the students’ computers. The Proctorio software also activates third-party cookies in order to provide “proactive chat support” – via the US-based company OLARK.2
The only way to participate in an online exam is to allow Proctorio access to the webcam, which has to be turned on for the entire duration of the exam. Examiners can decide if they want to monitor the candidates themselves, or if they have the software do it for them. And the examiners can instruct Proctorio to prevent the start of applications and downloads, or block extensions and personal settings. Even copy-and-paste can be disabled completely. If the examiners so choose, a “room scan” has to be performed before the start of the exam, where the candidates have to show the entire room to the camera. This “room scan” needs to be repeated on request during the exam.
But it gets worse. Universities and lecturers are lured in by Proctorio as potential customers with the argument that written exams can be conducted under “fully automated supervision”. To this end, incoming video signals are evaluated by the software using artificial intelligence, which is claimed to be able to detect the presence of an additional person the room.
On top of that there is an eye-movement analysis called “face detection”3 by Proctorio. The company describes this as follows: “The system detects anomalies from repeated glances into a certain direction, and flags these incidents as potentially suspicioius.”4 A few sentences later it says, “this does not mean that you cannot look away when taking a break for thinking. If you don’t use any aids, you don’t have to worry.”5 How generous of the Proctorio software to allow pause for thought! But if the software somehow still finds that suspicious, the candidates bear the full risk.
In fully automatic mode, the observed behaviour of the candidates is compared to patterns that are stored in the software as “normal behaviour”. If the software thinks the behaviour is fine, whatever that means, then the recordings are routinely deleted by Proctorio after 30 days. If there was something suspicious, however, the examiners can take a look at the respective videos. This means that in automatic mode, Proctorio’s artificial intelligence alone decides if there is suspicion of cheating. All by itself.
These kinds of automated evaluations of human behaviour are always questionable. In the realm of schools and universities, this revocation of the principle of presumption of innocence by any form of “automated grounds for suspicion” should be absolutely taboo, especially in view of the educational goals and basic values to be conveyed.
Many students will suffer from increased exam stress levels when they have to work eye-to-eye with a camera for hours on end, without knowing if their behaviour has just raised the suspicion of a supervisor or a piece of software.
I have been a university teacher for 25 years, and during this time I have proctored many written tests and given many oral exams. That is how I know that for many students exam situations can be very stressful, independent of the actual level of subject knowledge, and may often cause anxiety. During exams in personal presence, examiners can often alleviate the students’ stress and anxiety with just a few friendly, encouraging words. And I never find it suspicious when someone lets their eyes wander while thinking, as long as the gaze does not fall onto the neighbour’s text. The same is true for oral exams, where one can quickly figure out whether students really don’t know the answer, or are just too nervous to find the right words. Examiners are able to reduce the pressure when physically present, to build bridges, or to encourage the candidate by giving permissible hints. A machine, on the other hand, can only work through its algorithms, but cannot perceive if someone just lets the eyes wander, lost in thought, or with the intention of copying the neighbour’s solution.
Based on that knowledge, I would have strongly disagreed, before the Corona pandemic, with conducting monitored mass online exams, as would many of my colleagues. That we still permitted online exams to go ahead in this special situation, despite strong reservations, does not mean that these should now be considered normal. For all standard situations universities will have to return to presence exams “after Corona”. And in cases where online written or oral exams can make sense, they must take place under the supervision of humans, who will not just see a glance, but also understand it.
A software like Proctorio has no understanding of students, but instead tests if their behaviour is consistent with what other people have defined to be “normal”. Students in the USA, where Proctorio is in use at a number of universities, have reported in the Washington Post6 that the program grew suspicious if they merely moved their head, eyes or mouse unusually frequently during an exam. It was also deemed suspicious if they scrolled, clicked or resized windows “too often” – whatever the relevance of these insights may be. Finishing the exams too quickly, or too slowly, was also classified as “deviant”, “abnormal”, or “conspicuous”. Students also reported they did not dare leave their room to visit the bathroom for fear of being accused of cheating. In their own home.
Those who had the misfortune of having to take the exam in a room with lots of external noise, a slow Internet connection, poor lighting or a flickering camera were also flagged to the examiner by the software. To eliminate such factors, Proctorio advises students in their “FAQ for students” concerning the selection of a suitable room: “Early enough beforehand, consider in which room you would like to take your exam. You may also take your exam in a place where you can ensure adequate silence (e. g. an office).”7 Considering the tough financial situation many students have been facing since the start of the pandemic, this seems really cynical.
During normal times we would expect to see students taking to the streets to protest the planning of such a “complete and total control”. But we are in a pandemic. Protests and assemblies are hard to organise. And of course the students’ primary goal is to get on with, and finish their studies. So they will accept online exams, especially when there are no alternatives.
But perhaps students in the USA are just somewhat ahead of us. There were multiple protests directed at Proctorio and posted on social media. In Germany, on the other hand, our awardee, Proctorio GmbH in Munich, tries to present itself as being praised unanimously by universities, teachers and students. If we are to believe the company’s press releases there should be armies of students hungry for online exams, calling out in unison, “all is well as it is. Honest students are rewarded, and cheaters will be uncovered by the software.”
I don’t want to deny, as a university teacher, that online written exams at home may be more attractive and convenient to students than sitting in a university lecture hall. But whoever makes online exams the norm will have to accept the loss of educational fairness and equal opportunity. Those who are forced to write their online exams in a small room of a shared flat on a slow and old notebook while being bombarded with loud music from the room next door won’t have the same chances of success as those who can use a well-equipped study in a quiet neighbourhood. This is especially true when the Proctorio software considers nervous glances or loud external noises suspicious, prompting examiners to take a closer look.
What we have learned
Use of the Proctorio software for supervising and monitoring online exams is a severe incursion into the integrity of the students’ personal devices.
Permanent video surveillance of the candidates during the exam is a severe violation of their privacy and their private spaces, especially if a room scan is performed.
The automatic analyses of their behaviour performed by the “AI” software are intransparent to students. The general presumption of innocence, which applies to all citizens, is effectively suspended by the algorithms and the resulting intransparent control.
Gestures and particularly eye movements are monitored and evaluated continuously. The software can use these to draw negative conclusions, which in turn increases the pressure and stress for the students.
online exams at home are probably considered convenient by some students. But due to the diversity of living conditions, they are a threat to equal opportunity, which is to be ensured when conducting “presence” exams.
The potential for savings that Proctorio promises for automated proctoring makes the software attractive for cost-conscious universities. But the costs of these savings are borne by the students who are more comfortable with conventional presence exams than with online exams under the watchful eyes of a piece of software.
Compliance with data-protection legislation is already doubtful because of the absence of robust statements about a legal basis, and it seems unlikely that “freely given consent” can be ensured for students in an exam situation.
This is sufficient justification for the BigBrotherAward in the category “education”.
Congratulations, Proctorio GmbH.
1 There are numerous valid doubts about the “GDPR compliance”: For instance, there is no conceivable way for students freely to give their valid consent to the processing of their data by the Proctorio software, given the situation of a compulsory exam.
The processing of personal data of students for reasons of “product development” by the data processor (cf. Proctorio GmbH Datenschutzinformation, p. 1 (PDF) (German), (English) [Content no longer available]) lacks sufficient legal foundations.
Referring to “legitimate interests” of the Proctorio company for processing personal data according to GDPR, Article 6, paragraph 1, point (f) ignores the overriding nature of interests or fundamental rights and freedoms of the students, especially in cases where there is no alternative for them to the use of the Proctorio software.
2 Proctorio FAQ für Studenten (Proctorio FAQ for students), p. 8 (PDF) [Content no longer available]
3 op. cit., p. 5
4 op. cit., p. 5
5 op. cit., p. 5
6 washingtonpost.com: Cheating-detection companies made millions during the pandemic. Now students are fighting back (Web-Archive-Link)
7 Proctorio FAQ für Studenten (Proctorio FAQ for students) op. cit, p. 3 [Content no longer available]