5

Mini-Session: Across the Four Themes

To complement the full panels and keynote remarks, a “mini-session” was held at the end of Day 1 with three short presentations highlighting technological tools and frameworks in Spain, perspectives on the impact of artificial intelligence (AI) on teaching in Hungary, and a more global perspective on responsible AI.

SPAIN: TECHNOLOGICAL TOOLS AND FRAMEWORKS IN K-12 MATHEMATICS EDUCATION

Pilar Vélez (Nebrija University) explained Spain’s national education framework and the position of mathematics within it. Education is decentralized across Spain’s autonomous communities (the country’s subnational governing divisions), with a compulsory minimum curriculum, but the mathematics curriculum is largely uniform nationwide. Computational thinking and digital tools are explicit in the mathematics and technology curricula.

Current education-related legislation aligns national education with European Union and United Nations Educational, Scientific and Cultural Organization goals promoting competency-based learning and digital skills. The National Institute of Educational Technologies and Teacher Training provides a digital competence framework for education aligned with the European DigCompEdu framework.¹ She noted two evolving areas related to data science and AI. First, the data science curriculum has evolved from statistics to a “sense of data” with visualization and real-world modeling using digital tools. Second, regarding ethical AI usage, the Ministry of Education published a guide in 2024 that promotes AI-augmented rather than AI-led education with a focus on ethics and transparency in algorithms.

Despite the push for digitalization, Spain is challenged by a rural–urban gap, with small rural schools often lacking high-speed infrastructure. The digital competence plan aims to certify all teachers by 2025–2026, but many feel overwhelmed by the speed of AI development. The Digital Divide 2.0 is not just about access to devices (most schools have tablets and/or laptops), but the quality of their use. Mitigation strategies include program support for socioeconomically vulnerable schools to adopt AI and computational thinking equitably, as well as a unit responsible for integrated technology and for training nonuniversity staff through courses, a library of resources created by teachers for teachers, and other support.²

Programming is introduced progressively, following a no-code, low-code, high-code pathway with different tools and at different levels. GeoGebra serves as the standard mathematics platform for math teaching with detailed teaching guides. A number of other projects and resources are available through EU or regional funding.

Vélez concluded by pointing out the discrepancies between practice and policy. Ninety-six percent of Spanish students use computers but only 33 percent do so at school. Systematic classroom use of computational thinking, data science, and AI remains limited due to classroom management

___________________

¹ For more information, see https://joint-research-centre.ec.europa.eu/digcompedu_en

² For more information, see https://intef.es

challenges and high-stakes exams. Other challenges include teacher capacity, inequity of access, and contradictions between policy promotion and technology bans. Despite these challenges, she said, innovative projects show promising paths for integrating computational thinking, data science, and AI into mathematics education through problem solving and project-based assessment. It will not be easy or fast, but it can be done.

HUNGARY: HOW TEACHING IS CHANGING WITH AI

Balázs Koren (KJBK Infinity) posed a question: AI changes how we teach, but does it change what we teach? To consider the question, he first provided a brief overview of the Hungarian education system. It operates under an 8 + 4 or a 4 + 8 system. A standardized final exam is a main driver and has a big influence on what is taught and what students are interested in learning. The national curriculum, NAT2020, is being revised. An AI in education pilot is running in 60 schools for year 9 students and a policy for AI in education, applicable to every school in the country, is scheduled for completion in December 2026.

Algorithmic thinking is part of the mathematics curriculum, and data science is mostly taught within computer science. More than 50 mathematics competitions are offered to those students who are interested. AI has changed the composition of some of these competitions, including one that is more than 150 years old.

Before the pandemic, teachers were slow to embrace digital learning, but these tools became a part of everyday teaching during and after the COVID-19 pandemic. Now, he said, AI makes them feel less capable and confident. They are concerned that students will cheat in out-of-school work. Assessments include the final exam mentioned above, as well as competency tests, classroom assessments, and homework. Again, the big issue is how AI will change any of these assessments.

Koren commented that, since he has been teaching, he has heard that various innovations will replace him as a teacher, whether GeoGebra/Desmos, Wolfram|Alpha, or PhotoMath. AI will not replace teachers but it will change how they teach and assess. It is inevitable. It is not good or bad; it just will happen. When he has trained teachers, they ask what to teach if AI already knows the whole subject, what happens to homework, and how to prevent cheating. He urges teachers not to fight or compete with AI, but to collaborate with AI. He offered several examples: a CustomGPT that can serve as a math coach or tutor, Vibe coding, and Redmenta’s paper-based assessment tool.³

___________________

³ For more information, see https://redmenta.com

RESPONSIBLE AI AND AI GOVERNANCE

Ricardo Baeza-Yates (KTH/Sweden, UPF/Catalonia, University of Chile) began by sharing his favorite examination question for students: presenting them with a ChatGPT answer and then challenging them to do it better.

Baeza-Yates clarified that he prefers the concept of “responsible” AI over ethical or trustworthy AI. Ethics and trust are human traits, and he wants to stress that machines are not human. Trustworthiness puts the burden on the user instead of the builder. The systems do not need to be perfect, he commented, but it seems that people want them to be more perfect than humans are. He offered a taxonomy for irresponsible AI, with six elements particularly important for education. The most serious, he said, is automated discrimination. Other issues of concern to education include pseudoscience (in which the ethics of the problem to be studied is not scrutinized by Institutional Review Boards), copyright issues, disinformation, cognitive loss, and mental health. Since the Organisation for Economic Co-operation and Development has started to monitor cases of AI incidents in the news, the number has greatly increased, and that number is surely an undercount.⁴

Most authors use bioethics as a guide to define operational AI principles. In 1979, the Belmont Report for Biomedical and Behavioral Research articulated basic values: respect for persons (autonomy), beneficence, and justice.⁵ These values are articulated through such means as informed consent, risk and benefits assessment, and subject selection. These principles can conflict: for example, transparency versus privacy. The Association for Computing Machinery issued a statement on algorithm transparency in 2017 and responsible algorithmic systems in 2022. This latter statement, of which he is one of the two main co-authors, added the new elements of legitimacy and competency, minimizing harm, and limiting environmental impacts.⁶ A June 2023 statement extended the principles to generative AI.⁷

___________________

⁴ For more information, see https://oecd.ai/en/incidents?search_terms=%5B%5D&and_condition=false&from_date=1900-06-11&to_date=2026-06-11&properties_config=%7B%22principles%22:%5B%5D,%22industries%22:%5B%5D,%22harm_types%22:%5B%5D,%22harm_levels%22:%5B%5D,%22harmed_entities%22:%5B%5D,%22business_functions%22:%5B%5D,%22ai_tasks%22:%5B%5D,%22autonomy_levels%22:%5B%5D,%22languages%22:%5B%5D%7D&order_by=date&num_results=20

⁵ For more information and the full text of The Belmont Report: Ethical Principles and Guidelines for the Protection of Human Subjects of Research, see https://www.hhs.gov/ohrp/regulations-and-policy/belmont-report/index.html

⁶ For the full 2022 statement, see https://www.acm.org/binaries/content/assets/public-policy/final-joint-ai-statement-update.pdf

⁷ See https://www.acm.org/binaries/content/assets/public-policy/ustpc-approved-generative-ai-principles

New principles were needed related to limits and guidance on deployment and use, ownership, personal data control, and correctability. Responsible AI governance is based on these principles through a process that starts as an initial idea, its design and development, deployment, and maintenance. When AI fails, we must audit the system, whereas when it harms, we may need to go to court. Baeza-Yates closed by commenting that he “doesn’t worry about AI; he worries about us.” He directed participants to review these and other ideas in his 2023 responsible AI paper.⁸

DISCUSSION

Padhu Seshaiyer opened the discussion by underscoring the threat of coded bias and asked Baeza-Yates how to avoid algorithmic bias. Baeza-Yates replied that it is impossible to eliminate but can be mitigated. Algorithms must be contextualized to the problem being solved. Look at the principles of legitimacy and competency, he urged. Teachers can help learners understand the mechanisms that cause bias in systems and how to mitigate impacts, Markku Hannula suggested. Baeza-Yates opined about other types of bias in education. Why are we tied to a quarter, semester, or year? he queried. There is no reason to have subjects the same length. Moreover, he asked, why should all students take the same subjects and in the same order. Forget about time and sequence as constraints, he urged.

An attendee reflected that she has heard differing opinions about the role of AI in education and asked the presenters their thoughts as teachers. Vélez said she sees that students in Spain use AI more frequently than teachers. Koren said AI offers great tools to students, but students still need someone to lead and show them the way. AI can hallucinate. Someone left alone and completely new to a field can be misguided. Baeza-Yates related that when the first chatbot was developed at MIT in 1966, the researcher’s warning was never to confuse humans and computers. Machines have no understanding or knowledge on their own. They should not be given names or ascribed human qualities, Baeza-Yates said. There can be positive bias, too, for example, to which way to look for oncoming traffic in the United States versus Britain. It depends on the context, although it is usually done in bad way.

Another attendee commented that personalized learning is often assumed to be a benefit of AI without a clear understanding of what it is. Baeza-Yates said personalization makes sense in medicine and education, but in most cases it can be misused. In most cases, it is really

___________________

⁸ Ricardo Baeza-Yates. 2023. An introduction to responsible AI. European Review, 31(4):406–421. https://doi.org/10.1017/S1062798723000145

contextualization, not personalization. It has advantages, such as getting more data from users while protecting privacy.

Ben Galluzzo asked about interactions with social scientists and social psychologists. Research has occurred in these areas that might be useful in mathematics education. Baeza-Yates said he is involved in a movement called Digital Humanism that started in Vienna and that works broadly across disciplines. He commented that it took 20 years to understand the impact of social media on teenagers. AI is happening so fast that the impact will be known within 5 years.

Chapter 9 provides a summary of the reflections on Day 1.