25 March 2021 We have talked in previous articles about Artificial Intelligence and its proven or possible impact on education. In this article, we’ll explain why we believe it makes sense to use Deep Learning to help professionals find their next career step and succeed in it. What is Deep Learning? In essence, Deep Learning is a way for a network of computers to analyze data and learn from it (through multiple processes, hence “deep”) in such a way that will enable it to “predict” a behaviour based on a very large set of previous behaviours. Why is it relevant to the SILKC Path project? With SILKC Path, we want to offer any user the best recommendation of training necessary for them to achieve the necessary skills to land their dream job as efficiently as possible. We consider that the steps to move (through training) from their current job to their dream job form a *path*, and that this path will share common steps with the paths of other people. However, the diversities in people, their jobs, their dreams and the training they can follow is so wide and includes so many details that it would be impossible for a single person, or even a group of HR professionals, to offer a reasonable solution to each user in a reasonable amount of time. Finding out the skills that given user already has and the skills he/she is missing is relatively simple: we analyze his/her current job, use a common skills framework (ESCO in this case – as it provides a common set of skills at the European level) to establish the set of skills this user probably has (the “acquired skills”) then analyze the dream job and the skills it requires (“required skills”) and subtract the acquired skills from the required skills. With this remaining set of required skills, we look for the best available training which will teach all of these missing required skills. Easy. Where things get more complex is that users: Might not share the same background.Might not all live in the same town. Might not have the same ages.Maybe they don’t speak the language of the training.Maybe they have special needs.Maybe there is no training that provides all these missing skills.Maybe the cost of the training is too high.Maybe a given training is not recommended by other people with similar aspirations.Maybe it is recommended but the people who recommended it didn’t stay for long in their “dream job”, which would indicate some kind of failure in their plan.Maybe the training is not available at the right time. These 10 criteria, and potentially many more, are the elements, the data points that we want to analyze to offer better recommendations. Imagine that each of these simple 10 criteria would have about 10 variations each (easy to imagine for age, language and acquired skills). Simple combinatory analysis would suggest more than 10 billion combinations. Add one “10 variations” criteria, and we’re at 100 billions. Doing such an analysis manually for each user would be incredibly expensive, time consuming and slow. Programming it in a classical way would require huge development efforts initially and regular adaptations to the code in order to sequentially improve our model. Deep learning, however, excels at this kind of analysis. Using Deep Learning, we can design the basic rules of the system, give it examples of the results we would expect, then let the system process this information, finetune it and continue feeding it new information as we go, to train a system that continuously improves by itself. Conclusion Given the variety and number of values (criteria) our system will have to analyse to provide custom recommendations for users to get to their dream job, harnessing the power of Deep Learning seems to be a very attractive option. Processing this information manually (or programmatically) would require excessive resources and would have to be abandoned because of a lack of funding. By using Deep Learning, we are training machines to do this work, making it accessible to all.