When Muriel Grenier had access to the DALIA software dashboard, the individual educational assistant (API) who works at the Bois-de-Boulogne college initially felt “great pleasure”, but also a certain distrust.
Before his eyes, red, orange and green color codes predicted the risks of students dropping out and failing one or more courses. Gone, then, are the long, grueling hours and weekends manually compiling data into an Excel file from forms filled out by teachers and searching student records.
The artificial intelligence tool, used since last winter at the college, now does the work. However, a number of questions came to mind at the API. “I told myself that this tool doesn’t know CEGEP, doesn’t know everything that years of experience have taught me,” says Muriel Grenier. “I found it a bit odd,” adds the woman who feared that a “machine” would replace her.
She quickly felt reassured: the tool does not make decisions for her and does not replace her interventions. This offers a neutral, and above all faster, way of identifying students in difficulty and giving them the necessary help, Muriel Grenier now believes.
DALIA is used by around fifteen CEGEPs in Montreal and elsewhere in Quebec. Ultimately, an agreement provides for the gradual deployment in around forty establishments, specifies Optania, a Saguenay company which developed the software in partnership with Skytech, which is, for its part, behind the creation of the Omnivox portal. .
The tool uses dozens of already existing data, specifies the president of Optania, Louis-Raphaël Tremblay. Information from past students has already been compiled at the CEGEP registrar’s office for as long as 40 years, he specifies. “When a CEGEP comes on board and activates DALIA, we take data from the last 20 years. »
“100% anonymous” “training data” creates predictive models. They are then compared to data from a current student, revealing whether any characteristics fit a pattern. “If so, that’s where a mechanic comes in to drill into the model,” he said. The more we drill into the model, the more we establish a level of risk. »
A group of success experts was set up with CEGEPs to determine the most relevant data. High school grades are used, and predictions are refined with data from each college course. Pooling data from CEGEPs that use the tool allows for more accurate predictions.
“Detecting students was done in the sixth or seventh week of the session,” says Édouard Taza, president and CEO of Skytech, DALIA’s supplier. We had to wait almost halfway through the session before meeting the students to raise their awareness. He could, for example, start attending the French help center only five weeks before his final exam. »
Student concerns
Students are informed by their institution that they can sign, or not, a consent form for the software to estimate their probability of success and perseverance. They can withdraw their consent at any time. At Ahuntsic and Bois-de-Boulogne colleges, more than 95% agreed to sign it.
“We wanted to go for a two-phase deployment, with the first step being to verify that consent was well understood and integrated by the student population,” explains Isabelle Mailloux, director of continuing education and academic progress at Ahuntsic College.
Only APIs have, for the moment, access to DALIA, specifies the Regroupement des cégeps de Montréal, which has 12 establishments and which is a driving force in the implementation of the tool. “This could give a favorable or unfavorable prejudice to the teacher,” underlines Stéphane Godbout, general director. However, it does not close the door to possible access.
An avenue that students refuse. “We must avoid the Pygmalion effect. The more we believe in the success or failure of the student, the greater the chances of it happening,” indicates Laurence Mallette-Léonard, president of the Quebec College Student Federation.
We must avoid the Pygmalion effect. The more we believe in the success or failure of the student, the more likely it will happen.
She urges the authorities to exercise caution. “We do not at all want it to be used for admissions, because it can reinforce discrimination processes,” she adds. A scenario which is firmly ruled out by the speakers with whom The duty spoke.
Everyone assures that ethics is at the heart of the process. “All partners, including the supplier, have signed the Montreal Declaration,” mentions Stéphane Godbout. An ethics committee, composed in particular of philosophy professors, was also created.
“It is always the student who is the one and only owner of his own data,” emphasizes Édouard Taza. “Impossible” for the company to sell them, he said, and a CEGEP must obtain the student’s consent if it wants to distribute its data externally.
How effective?
For the moment, CEGEPs are not yet able to quantify whether the tool has, in fact, made it possible to reduce failures and dropouts. At the Bois-de-Boulogne college, the director of studies nevertheless has the impression that the software is effective. “Tests and analyzes were carried out before use,” emphasizes Sébastien Piché. He was able to predict almost 88% of failures. It’s still efficient. »
For her part, Muriel Grenier emphasizes that her approach remains the same as before the implementation of DALIA: the API contacts a student who might have more difficulties by simply inviting them to come see it. “If he says yes and comes into my office, the conversation starts with how he’s doing and his session,” she describes. He’s the one who’s going to call me out if there’s something he’s uncomfortable with. »
These meetings sometimes make him realize that a student comes from a less fortunate background and must work more, which is detrimental to his studies. Or that someone else is less well psychologically, and that she can refer him to services. And now that Excel spreadsheets are out of her life, she has more time to meet with students.