The students of the TEN-7006 Instructional and Training Systems Design course received the visit of Apprentx’s two co-founders Jean-Philippe Bradette, President, and Julie Castonguay, Director of Experience. They shared their insights and achievements in personalized learning and artificial intelligence through their B12 application.
Here are some of the questions they were asked.
How does the B12 application enable personalized learning?
Julie: Personalization is done with artificial intelligence (AI). Personalizing a learning experience involves taking an interest in each person individually in the program to choose the best activities for them. Obviously, it’s a huge job to do when you go beyond a small class; however, it’s all the more relevant with a large pool of learners as more data is connected and improves the predictions of the algorithm.
The AI detects patterns of what is happening in the participants’ learning. The program will offer additional activities or adapt them according to the learner’s path. For example, adapting the level of difficulty: not asking questions that are too easy or too difficult helps to avoid dropping out and maintain motivation. The AI thus calculates success rates, and offers content and topics according to them.
Obviously it is also always possible to segment by hand, by organizing cohorts without using AI for example.
Jean-Philippe: As a company, we talk a lot with executives who are looking for performance and have employees with very busy schedules. The AI optimizes the time spent in training, thus reducing costs by seeking the right level for each learner, including those who would sometimes be neglected in less personalized contexts.
Is there a basis to the choices of the algorithm, to customize the course?
Julie: The algorithm assimilates the learners’ performances from their first questions, it learns from the individual and also from the group. It records patterns that emerge, and projects predictions. This is where pedagogy and artificial intelligence work together: when AI predicts failure, a remedy can be applied to prevent it.
Jean-Philippe: It’s important to keep in mind that in a typical project, we work to bridge a performance gap, to find the best process to acquire a skill. From this point of view, the algorithm gives us valuable information on KPIs (key performance indicators). They are used at a higher level of business, which reflects a very important strategic aspect as well.
Such an application is an enormous amount of programming work. The jargons of instructional designers and integrators seem so different, how can we develop translation skills to communicate?
Julie: It’s a fairly common stereotype that it’s hard to understand each other. Actually, programmers are very proactive in understanding our needs and finding a solution. We outline the objective, the data we have, and then we work to understand each other. Of course if our idea is vague, we won’t understand each other. It is a basis for communication in collaborations of all kinds. The “command” must be clear for the programmer to translate it into code. Personally I like to draw diagrams as it forces me to clarify, specify and develop my idea.
Jean-Philippe: We also all have different personalities. Designers can be very creative at a high level, but less strong on details. Julie and I work in a complementary way: if I imagine a new B12 feature, Julie will make me think of concrete aspects to clarify the ideation. We combine our forces in tandem to develop the idea. This facilitates communication with programmers and makes projects very effective from the earliest developments.
How much do you need to know or understand AI to be able to speak in the terms of programmers?
Jean-Philippe: It’s a good question, actually not so much: it also depends on the person on the other side, on his or her ability to ask the right questions to clarify the needs and link them to the solutions they have in their toolbox.
Julie: Also, we need to demystify what AI does versus basic statistics: AI takes numbers and predicts, so AI doesn’t invent anything, it feeds on facts. From there, all we need is to understand what each trade needs to be able to produce something. Programmers think in terms of data and functionality, as graphic designers think with a mood board, for example. Understanding what the other needs to produce is a valid reflex across all trades.
What are some limits of B12?
Jean-Philippe: All tools have their limits, of course. For example, B12 does not touch social learning, for that aspect we’ve integrated outside tools we haven’t developed ourselves. Another limitation relates to evaluation, and that is also a business question: many companies in Quebec are still reluctant to confront employees to an evaluation, for various reasons. On this, the human takes over to interpret the situation and talk to the client.
How do you manage the aspect of privacy, data confidentiality and ethics?
Jean-Philippe: This is obviously a major concern, it is very important. First, all the data is secure, the algorithm divides them by client and even by project; anyway it doesn’t make any pedagogical sense to share this data. Then, each learner has a user id on the platform, so that no personal or demographic data is collected by the algorithm. This wouldn’t have any pedagogical meaning either: we want to see how the person learns and answers the questions. Variables like age or gender aren’t useful unless their analysis is relevant in a particular project.
How to determine the variables relevant to the algorithm?
Jean-Philippe: There are mathematical and statistical models to determine which variables are relevant to predictions. For example, AI can determine whether the learner will perform well on a given question based on previous results on the same topic or on a different one, on confidence level, but not gender or age!
How did this idea of trust capital come about, and how is it calculated?
Julie: In the first features we imagined for B12, we integrated confidence-based learning to better understand the learner’s level of knowledge: for example, was this good answer given by chance? Or on the contrary, did a preconceived idea lead too quickly to this false answer? The AI will correct the situation. Trust capital is calculated simply by asking the learners about their confidence level before each question.
Jean-Philippe: For those interested in reading further, here are our inspirations for confidence-based learning: Darwin Hunt, Dieudonne LeClerq, Emir Shuford, and James E. Bruno.
What variables impacting performance (other than educational) can be integrated by AI?
Julie: There’s a limit here because we don’t have all kinds of data. For example, an event happening in one department versus another could impact performance, but the AI won’t take it into account unless the variable “department” is integrated. However, the AI will always be able to predict from the activity recorded in the application. Another example is what time of day people use the app, which could have an influence. We can only integrate in B12 the variables we have access to.
In education, a teacher will work hard with weaker learners, and sometimes less to sharpen and elevate the knowledge of better-performing learners. Could AI increase the motivation of gifted learners by feeding them more?
Jean-Philippe: It is indeed a hypothesis that we have. We’re currently working on a big project with Collège Ste-Anne, from elementary to college levels, in which we think AI can help teachers identify all levels of students among the class and adapt the content accordingly.
Does integrating B12 into a training require a large budget for the client?
Jean-Philippe: Designing a course requires a good budget, but integrating B12 itself doesn’t: the tool is distributed in the form of a license, then it varies according to the content, but the costs are reduced because the implementation itself is simple.
Would it be possible to expand the use of AI across the school system to assist teachers?
Jean-Philippe: The education network is a more difficult environment to reach, more than the corporate environment. There is a more political and rigid structure that makes decision-making complex: for example, we contacted the office of the Minister of Education, but we were told that the school boards are in charge of that aspect. There is also a lack of resources, although B12 would be a most welcome help to teachers, who have a lot of work to do at the moment.
What makes a platform competitive: is it the technology used, the innovative idea of data processing, or the performance of programmers?
Jean-Philippe: It’s going to sound cliché, but for us, it’s the team that makes the difference. We compete with hundreds of platforms around the world, literally! We often hear about new platforms we didn’t know about. However, our pedagogy is very strong, our experience is proven. Approximately 16,000 learners have used B12, and the overall feedback is that the application appears simple and effective, although the “engine” is complex. So what makes the difference, is the humans behind the beast!
One last question: Who do you think is the best candidate to join your team? What expectations and skills do you have for your recruits?
Jean-Philippe: The first question we ask is obviously: what music do you listen to?! (laughs). At Apprentx, we have different profiles of designers, the orientations can be different depending on the interests. What’s essential is to be able to learn very quickly, to know how to speak to the client, to have good interpersonal skills. For us, it’s better than someone with an impressive but stagnant background. We like curious people, who want to have fun. Obviously, you have to have a solid and up-to-date educational base. The affinity with technology must be there, but no need to know everything: it is curiosity that is paramount.