What is learning as a phenomenon
The way people learn is an outcome of millions of years of evolution and it cannot be changed with technology. In short, we learn through interaction by observing the world around us i.e. we make observations, conceptualise our observations and connect them into our existing understanding. When we connect observations we have made before, we just strengthen our current understanding. When adding new observations, we learn. When we make observations that don’t fit into our existing understanding, we might end up learning and at the same time changing our understanding radically.
In science, the learning process is studied at the for example chemistry level, biological level, neural cell level, social level and psychological level. All science supports this generalised definition of learning, no matter if we call it conceptual learning, constructivism or connectivism. Claims that AI will change the way we learn is just like a claim that television will change the way we learn. Every ten years, there have been technologies that have claimed to change the way learning happens. Games (2010), mobile devices (2000), Internet (1990), CD-rom (1980), television (1970), programmed instruction (1960), radio (1950) and spirit duplicators (1930), just to mention a few.
We dream that technology would make learning fast and easy, but that’s an illusion.
However, all technologies, including artificial intelligence, have changed our abilities to make observations of the world. Furthermore, technologies have enabled a remarkable change in education and teaching. Radio brought access to good quality educational content for millions. Internet provided access to a lot of information, readily available for learners. A decade later, there was no need to just deliver information in schools – new pedagogical innovations were developed. The Learning process, however, remained the same: we yet connected our observations into our mental understanding. Learning is hard work, but we can make it more enjoyable, accessible and effective with good pedagogy. Technology alone is never the answer.
Coming back to Artificial Intelligence (AI)
We should not spend too much time on defining what AI is. All machine performed activities that would require conscious thinking also from people, can be called Artificial Intelligence. AI is not dependent on which algorithms or technological frameworks are used.
Furthermore, AI is old. As a science, it comes from the 50’s and it has been with us for more than 60 years. The first generation AIs were decision trees and state machines, based on programmed rules. In many solutions, the rules were based on scientific work and millions of lines of data, but the rules were programmed after people found the rules. In second generation AI, there are no pre-programmed rules and machines would learn the rules from the given data. In other words, the rules with which AI works are not programmed, they are (machine) learned. There are excellent use cases for both types of AI, for some cases programmed rules are perfect, because they are fast to perform, e.g. spell checking. Some other cases require machine learning because of the complexity of the case, e.g. speech recognition. The third generation is going to be about extending the concept of machine learning. i.e. teaching machines case by case with natural language.
AI can be used as an assistant for people, as a closely managed worker or as an autonomous system. In many cases, we use AI in the form of a co-worker, letting it to do easy, boring, dirty or dangerous tasks. Assisting AI performs numerous supporting tasks for professionals and so enabling them to focus on higher order thinking. Currently, there are very limited number of autonomous AIs in use, however, when we refer to AI, we tend to think autonomous AI. It is important to note that the ethical discussion around AI is very different if we talk about autonomous systems or systems that provide assistance by performing routine tasks. Alongside AI ethics, we should discuss data ethics: who is allowed to collect data, who owns the data, is the data real and valid, what are biases in the data, what is the role of manipulation, and so on. We have to accept we can not talk AI as an isolated system, it is always connected to other technologies and processes.
And how AI will change education
There is nothing new in applying AI to learning. In 1912, Edward Thorndike introduced his idea about The Learning Machine. The machine could ask questions of the learner and suggest further readings, if required. Because there were no modern computers invented at that point, The Learning Machine applied punch cards to run Adaptive Learning features 100 years before the concept of Adaptive Learning became popular in practice.
Adaptive learning applies computational methods that fit into the AI-family, and so we can call it AI. There are also two generations of adaptive learning. Prior to the 1990s, adaptive learning was based on rules found in scientific research and then programmed into machines. Post the 90’s, real-time data was used in order to provide highly individual learning experiences for the learner and detailed learning analytics. This required other technologies like fast internet, big memories, own devices and computation power to support AI.
There are parallel opportunities to apply AI in several educational tasks to either improve or speed up the process. However, the power of AI is not in improving old process, we have to rethink the processes.
If we want to understand the major opportunities related to AI, we have to focus on one of the biggest global challenges we face in education: We don’t have enough teachers to provide education for everyone. This is critical in elementary education, but remarkable also in vocational training, higher education and lifelong learning.
In early years of education, such as elementary education and primary education, the human teacher cannot be replaced. No way. Soft skills and transferrable skills such as critical thinking, problem solving, team work and communication needs interaction with people. However, we can provide AI-based teaching assistants and helpers for teachers and so enable teachers to work with bigger groups. This, however, requires AI-pedagogy, which we don’t have it yet. We need to invest a lot in pedagogical research in order to understand the full potential of AI to elementary education.
In lifelong learning, work place training and professional development, the key factor is to understand what skills are needed in the labour markets now and in the future. Although there is research being undertaken, big data is not being applied and micro-skills are not being focused upon because there is too much data and complexity for people.
For a couple of years, we have had AI assistance that reads through terabytes of job openings, curriculums, future forecast reports and labor market analysis, to build a real time research report on skills seeking. We could do this also with people, if we just could have millions of people to do this work.
AI can build parallel models on an individual’s skills at micro level, it can easily show an individualised career path: direct jobs with person’s current skills as well as low-hanging jobs that are accessible after person gets couple of identified skills. Furthermore, when AI knows the curriculums, it can also reveal educational opportunities relevant for individual.
When the skill gap is recognised, AI can construct tests and curate content packages for the individual. In other words, AI can produce online courses from identified skills and so bring vocational training and lifelong learning for billions of learners.
Finally, AI won’t change the way we learn, but it can change the access to education, globally.
click here to watch making of B-AIM: