Long before Coronavirus, discussions about automation and Artificial Intelligence (AI) were divided between visions of a catastrophic apocalypse and a future utopia where humans live a life of leisure. The reality is that AI has great potential to transform our lives in a very positive way, but only when security, safety, and psychological concerns are all given appropriate consideration.
This is equally true when it comes to the use of AI in online proctoring, particularly when fears are heightened by the unexpected and rapid adoption of new technology in education as a result of Coronavirus and social distancing.
The question is, how do we use AI to our benefit in online proctoring while mitigating the risks? And how do we protect the security and safety of both education providers and students, while taking understandable uncertainties about new ways of working into account?
Information and Transparency
As human beings, we are wary of what we don’t understand. It’s an evolutionary attribute that has kept us alive for thousands of years. However, it can hold us back from progress and experiencing new things. To combat this natural tendency, information and transparency are all vital when introducing anything new, including technology, for the first time.
In recent months, online proctoring that relies entirely on AI has been causing significant anxiety for students who are being asked to take important assessments online rather than in a traditional exam hall. Often, the speed of deployment has meant that important communication about how AI is being used has been missed or forgotten. That’s why, whenever online proctoring uses automated decision making – whether that’s AI, machine learning, or algorithms – facts and reasons for the approach have to be communicated to students.
Communication is critical. At PSI, we always clearly communicate that while we use AI in both our live and record-and-review proctoring solutions, this is only ever in combination with a human proctor. This enables us to utilize the latest technology to enhance security and efficiency for educational establishments while providing the added familiarity of a human being, which reassures students and staff.
In addition to providing notice when AI is being used in proctoring, it is even more critical to communicate how the outcomes of any automated decision will be used. With PSI, this might be to flag concerns during the ID authentication process to raise an alert that malpractice may have taken place during a recorded test or to recommend that a live proctored assessment be stopped or paused due to suspicious behavior.
During the check-in process for example, facial recognition algorithms are able to very quickly authenticate the identity of a test taker, comparing a selfie with a selfie from a prior exam or an ID. This minimizes the amount of students’ personal data that needs to be stored and creates an efficient process for the student. While facial recognition software provides a meaningful benefit to the student and the proctoring provider, it is known to occasionally fail to properly match a person’s image. At PSI, when AI flags a potential failed ID authentication, the images and data are provided to a human proctor who is then able to investigate in more detail.
Proctoring during the exam has also made more efficient use of AI. Review of the webcam recording of a student, their desktop, and surroundings by a human proctor can pick up any anomalies in the testing environment. But when combined with AI using pattern recognition, object recognition, and eye movement detection, this increases the likelihood that an unauthorized person entering the room or a student using a mobile phone or any other unauthorized materials will be detected.
Throughout the process, the decision about whether a test should proceed or if the results should be invalidated is never fully automated. Any suspicious activity will be flagged to a human proctor for further investigation. This approach also means that an AI can learn the behaviors to look out for and become more effective over time.
Algorithms and AI
With stories in the media about biased AI leading to poor recruitment, legal, or healthcare decisions, there is a growing public awareness around the need for appropriate controls when developing AI algorithms. These algorithms are only replicating the often-unconscious biases of the programmers that built them. But where any decision is made as a result of automation – such as to flag potential malpractice in an exam – the algorithm used should always be subject to thorough and ongoing evaluation for fairness and quality.
We take significant steps to minimize the risk of errors and prevent bias and discrimination when developing AI and machine learning algorithms. With PSI’s latest proctoring platform, every time a proctor flags a violation, that data is captured and used to enhance and train our AI. Similarly, every time AI flags particular activity for review by our proctor, that proctor serves as a check and trainer for existing AI. Important decisions about whether to stop an assessment or invalidate the results of an exam are never taken by AI alone.
Finding a Balance
While it’s important to stay at the forefront of technological developments, both the process and outcome of any exam, test, or assessment can significantly impact a student. The potential to enhance security and safety should always be weighed against student experience, as well as the psychological impact of introducing something new at what is already a stressful time.
In time, AI will become smart enough to make judgments on the severity of its own findings and take independent and appropriate action. However, this level of AI intervention should only be adopted when we are not only sure of the technology, but also that user confidence is high enough. In the meantime, clear communication about how AI is being used, along with an appropriate level of involvement by a human proctor, will create a balance between security, efficiency, and user comfort.