How does rtificial ntelligence (AI) influence learning processes?

Inge de Waard: In education, AI can speed up learning processes, but there is a risk. AI in its current, multi-varied forms (e.g. natural language processing, machine learning, deep learning, neural networks...) is, at best, supporting the learning processes at this stage in time. There is no magic that simply enhances earlier knowledge. I would even dare to say that it is simply adding additional speed to the learning processes. This gain in speed is a result of the more personalized suggestions that can be embedded in the overall learning processes.

Spaced interval learning is a simple example: someone is repeatedly answering the same questions but with an interval, until they get all of them right. Each individual has a slightly different memory, and with AI we can observe when memory works then program the intervals between the reissuing of the questions to fit that particular person's optimal memory capacity. As a result, that learner absorbs that particular information more rapidly. In a sense, we optimize and automate the learning process r at think we optimize it. There are some potential risks involved. If we only train children to think , and only provide them with one model that we promote as the ber-model, thn we clearly overestimate our current AI options, and we severely underestimate the complexity of learning.

There has been a lot of on whether you can use AI to follow what children are doing in the classroom (Asian test cases), and where the teacher can manage the classroom in terms of what each child is doing, whether is still all the teacher's guidelineso me, is selling our children short. It is like taking the Pavlov test and treating our children as dogs that need to be trained for one task only: sitting in a classroom and taking tests. Learning and living require time to reflect. 

We start to understand the additional options provided by AI. Thanks to our growing understanding of big data, of learning analytics, and of combining algorithms both for machine and deep learning, we are starting to see more options in which AI in can be . Complex relationships can now be analyzed and used to improve the learning process. For instance, we are now capable of analyzing semantic relationships between words and concepts, without those concepts having to be literally written.

For example, if we feed a course textbook to an AI engine that is trained to skill concepts related to a particular field (e.g. renewable energy) engine can the learning concepts that are at the core of that course. These skills can even be concepts that are not literally mentioned in the textbook, but which have been proven to be related and thus as a concept. This has a profound effect allows specific tools to compare different manuals on the same subject and construct the best manual from all of them. manuals into concept-based paragraphs, the quality between those paragraphs, , and then all the best paragraphs and them back together into a new

This saves for teachers and potentially saves costs for parallel courses.  

The AI project I am working on provides easier access to start the learning process. It identifies the industry needs (AI-driven), pinpoints emerging skill gaps in that sector (AI driven), analyses the existing workforce to know where urgent skills gaps are situated (AI driven) and then refers employees to a personalized learning trajectory addressing their skills gap (part AI, part human support).

The goal of this project is to ensure that employees of the sustainableenergy sector stay futureproof in a quickly changing working environment. In order to reach this goal, we try to bring meaningful learning options closer to their own needs, hence speeding up their own learning process.

How does it change the learning experience?

Inge de Waard: Similar to all processes that are now increasingly supported by AI, the change in the learning experience is a result of the just-in-time and more specific handling of big data. Where the learning processes are enhanced, the learning experience as such is still a very complex cluster of actions undertaken by individuals according to their own character, their motivation toward the subject matter, their technological preferences, their personal learning goals, etc.

This also means that learners can be quite diverse. To me, it is the complexity of the learning experience that propels our society forward. Therefore, I am a deeply committed supporter of keeping complexity in the learning experience. I like to compare it to Darwin's evolutionary theory. Yes, some of nature's prototypes are really say the least (e.g. the platypus), but nature is because of this constant mutation, constant testing, and organic diversity. It is the same with learning. 

Although we live in a society that is increasingly automated, we - as teachers and learning developers - must ensure diversity in approaches. If we only , linear type of learning experience, it is going to slow down progress, as nothing of interest only consists of one approach or one action (think of life itself). is why we must not only have fficiency as our main goal when we develop AI in ducation.ensure creativity, organic chaos, and even stupidity as a means to try out AI in . 

Lifelong learning is a must in this rapidly transforming society in which citizens need to be informed of what AI can do for them, of what effects data has on their own life. As the world evolves ever more rapidly, all of us learners need to stay on top of our field, or ideally, find our way through life by learning what matters to us in such a way that we can use it to our societal and personal advantage.

In the context of AI, how can it be guaranteed that the learner actually acquires the requisite knowledge?

Inge de Waard: Validation of learning is a long-standing challenge for teachers and trainers. Even the learners themselves aren't always aware of their learning or learning progress. At present, AI is very conservative in its implementation: making better classroom books, keeping track (or monitoring) learners and employees, distilling synthesis from a long report to offer as an exercise, and so on. These are all classic implementations of learning. It has no bearing on creativity, philosophy, the process of thinking, or the actual benefits of learning (getting a feeling of fulfillment). If we look at optimal learning, we can see how this differs from AI-in-education solutions.

What do we learn if we can choose whatever topic we like? In this ideal situation, the learner will be intrinsically motivated to learn something they will choose something they like they will actually overcome challenges related to the subject area they are learning and finally, they will get a feeling of accomplishment having mastered this new information and having turned it into new, useful knowledge. To the , their voluntary learning has then resulted in requisite knowledge, validated by themselves. 

When looking at the AI in , the automation and speeding up of AI solutions do nothing to our understanding of what it means to acquire requisite knowledge. In classical settings, we as trainers or teachers use to measure understanding: multiple choice, open essays to test understanding and memorization. As AI solutions we use spaced interval learning, virtual as well as augmented realities to create simulations or to provide a real work floor process that the employee needs to follow with feedback (eg. construction on an assembly line)e monitor employees to see whether they work efficientlymeans work as hard as they can within a process.

Acquiring requisite knowledge is something that can happen in formal and informal moments, as insights in learning can manifest both at the work floor or outside (admittedly, the latter is the case for knowledge workers). Drilling learners is not the only way to provide and test requisite knowledge. With AI we should go past production move towards qualitative efficiency, providing optimal living conditions to our employees and learners.

Emotions are at the core of learning, which means that optimal learning and the acquisition of knowledge are also embedded in our holistic human brain, including emotions, preferences, and paralleling each of our individual capacities.  
 

What experience has your company gathered so far in the use of AI?

Inge de Waard: For InnoEnergy the AI experience also starts from something quite conservative and built upon a long-standing challenge: How can we reskill or upskill employees so that they meet the requirements of a rapidly changing field? The first steps in that process are feasible and comprehensible: By screening industry reports, government reports, white papers, etc., we can do a predictive analysis of all these documents and get a pretty good view of the direction in which a specific field of expertise - in our case renewable energy - is moving. We do this by using natural language processing (NLP). We can also screen courses and signpost learners to these courses in order to enhance their skills and competencies. An observation here is that it is easier to guide them for hard skills than soft skills.

Where it really becomes difficult - or to me exciting - though, is when it comes down to the actual learning. Can we program or start an AI engine so that it can actually surpass the simple suggestion of courses based on NLP? Can we take into account people’s characteristics and cluster courses that they should take, but also provide the learners with a meaningful, personalized learning path that will be manageable and fit their own learning interests and preferences?

As I researched how adult learners self-direct their learning, I can use that framework to pinpoint potential indicators that can influence the learning process (eg. learners who prefer learning on their own versus more social learning-oriented learners). But one can feel how much more complex and difficult AI solutions once you surpass the correlations, the algorithms based on measurable facts, and especially when we move into unexplored the validation of formal as well as informal learning. 

Looking at what we can do with AI, it is quite exciting for coders for administrators. in a way, there is nothing really new except for speeding up the processes. We measure with our old learning and teaching standards, we should be using AI to go beyond what we had up until now, and move towards a more holistic human plan all of our human capacities are into account and used to their full potential, for instance by turning learning into the satisfying experience it can be. We cannot even inclusivity in our AI systems, so there is a long way to go.