ML in K–12
Our interactions with technology are less and less defined by rule-based systems—the kind of computing currently taught in almost all levels of education. Our interactions today are with much smarter technology: Our systems recognize our faces, voices, gestures, and speech. How those systems work is not, however, taught in schools.
Our work on machine learning in K–12 education is focused on banishing magic from technology: Children need to learn how their everyday technology works, learn that it can be mastered, and see the work opportunities they enable. Understanding how technology works is necessary for critical awareness of their use, too: Technology can create jobs but also erase them. Technology can solve social problems but also create new ones. Technology can reduce, but also worsen inequalities. Our research on machine learning education provides technological and pedagogical tools for learners of all ages to understand how the data-driven world around them works. We have taught machine learning pilot classes for different aged learners from pre-K to high school.
We are currently running machine learning in K–12 workshops in schools in Joensuu and nearby areas. Children co-design ML systems with researchers and educators: They describe problems that could be solved with machine learning, they plan what data need to be collected to train the ML models, they collect the data and train the models, deploy the models in apps, test them, and plan for further development. Our 2019–2021 publications show a range of results from the workshops.
Data-Driven Studies in Educational Research
We work on a series of articles that combine network science, social network analysis, and scientometric research in the domain of education. The outcomes show conceptual timelines, development trends, milestones, networks at many levels, and geographical shifts in educational research. We analyze learners’ interactions in terms of network formation, network robustness, and multimodal data. We use the Finnish supercomputing services at CSC for processing the massive amounts of data some of these models require.
Analytics of knowledge building and innovation processes in K-12
Learning analytics has often been rooted in the acquisition-oriented and teacher-led practices of the formal education systems, in terms of, for instance, using drill-and-practice types of tasks for assessment of predefined learning objectives of acquiring specific curricular knowledge and skills. Next generation learning analytics research track conducts research in the context of constructivist pedagogies, informed by cross-disciplinary learning sciences. In such modern learning scenarios, students proactively engage in their own learning processes, setting goals, choosing problems to work on, making choices and monitoring progress.
We work with pedagogical settings such as project-based education, knowledge construction pedagogies, and invention projects. Our ongoing pilots are conducted in the contexts of flipped classrooms, and maker-projects. We are building analytics of students’ learning trajectories and decision-making during learning activities. Our aim is to provide learners and teachers with data that they can use to understand and improve their own learning processes. Our previous work involves pilots with maker-pedagogies, the use of sensors, such as moodmetrics -sensors as part of new measures in learning, and pilots with modern metrics such as growth mindset and maker mindset as part of innovative learning.
Ethics of AI in education
The megatrends of computerization and digitalization shape the world in fundamental and unpredictable ways, raising new classes of ethical and legal dilemmas. The ethical implications of modern artificial intelligence are vast. Related new initiatives have been launched to support the general public to reflect on potential future scenarios taking into account e.g. aspects of privacy, surveillance, job losses, misinformation, diversity, algorithmic bias, and transferability. While ethics of AI has become an active research area, ethical and legal implications of AI in learning have been less researched.
The increasing use of AI in analytics of learning data creates a new set of ethical concerns, which our ethics of AI track researches. While the amount of research in learning analytics is expanding, research on ethical and legal implications of AI in education is significantly low. In our research, we develop and research methods of ethical AI for learning, which are not, e.g. based on statistical classification or profiling of students, or methods, which result in stigmatizing or creating self-fulfilling prophecies.
Our research takes a multi-dimensional view on the ethics of learning analytics. The research combines insights on machine learning with a holistic reflection on ethical perspectives in education and the society as well as child as a legal subject and the responsibility to protect children’s privacy and rights. Our research activities are focused on the design of ethical algorithms, ethical use of data, and in developing and publishing ethical guidelines in the form of the ethics governance model in learning analytics.