The Transformative impact of AI on education, from an ethical perspective.pdf
0 likes942 views
Presented by Roland Klemke (Open Universiteit) as part of the EADTU Webinar Week 'AI in Education: Assessment, Ethics, and the Future of Learning'
More Related Content
Similar to The Transformative impact of AI on education, from an ethical perspective.pdf (20)
More from EADTU (20)
Recently uploaded (20)
The Transformative impact of AI on education, from an ethical perspective.pdf
- 1. The transformative impact of AI on Education, from an ethical perspective Roland Klemke Empower Webinar Week AI in Education: Assessment, Ethics, and the Future of Learning 03 December 2024
- 2. https://www.worldometers.info/world-population/ These People need to be educated!
- 3. They need to be educated for longer! 46,5 yrs in 1950 73,4 yrs in 2023
- 4. https://www.millennium-project.org/15-global-challenges/ They need to be educated better!
- 5. (Roser, 2024a; UNESCO, 2022; Gopalan, 2016) They need to be educated for an unknown future!
- 6. Brief History of Learning Innovation
- 7. Pedagogic change no longer comes from just educational research; it comes from insights in cognitive science and, increasingly, through technological innovation Donald Clark, 2023
- 8. The learning brain Social Learning Life emerges on earth Institutionalised Learning Learning with Technology Technology that Learns ~4 bn yrs ~500 mn yrs ~200.000 yrs ~4.000 yrs ~100 yrs now
- 9. (Roser, 2024b; Laitinen & Sahlgren, 2021; Khutsishvili, 2023; Shibuya, 2020; Smith & Lundstrom, 2024)
- 10. AI and Education
- 11. Final objective: Intelligent Tutors The vision is to create intelligent tutoring systems which fully understand the user and can maximise objectives such as: - learning gains - collaboration - skill mastery - knowledge acquisition - behavioral change . Hey, Im Clippy! In the past I was not that smart and helpful. But with AI I can become much more intelligent!
- 12. but Learning is a complex process! Bloom, B.S. et al (1956). Taxonomy of Educational Objectives, Handbook I: The Cognitive Domain. Bandura, A. (1971). Social learning theory. New York: General Learning Press. Three domains of learning (Bloom) Social learning theory (Bandura)
- 13. The 2-sigma problem (B. Bloom, 1984) The Pedagogical Bottleneck
- 14. AI-based Multimodal Learning Analytics Model Di Mitri D, Schneider J, Specht M, Drachsler H. From signals to knowledge: A conceptual model for multimodal learning analytics. J Comput Assist Learn. 2018;34:338349. https://doi.org/10.1111/jcal.12288 Body + context Mind Physical Digital
- 15. Modelling psychomotor Adam G. Kirk, James F. O'Brien, and David A. Forsyth. "Skeletal Parameter Estimation from Optical Motion Capture Data". In IEEE Conf. on Computer Vision and Pattern Recognition (CVPR) 2005, pages 782788, June 2005. Biomechanical movements are finite and take place in 3D space Body and mind are highly inter- connected! Psychological and physical factors cannot be separated ASSUMPTION 1 ASSUMPTION 2
- 16. Modelling affect (emotions) Arroyo, I., Cooper, D. G., Burleson, W., Woolf, B. P., Muldner, K., & Christopherson, R. (2009, July). Emotion Sensors Go To School. In AIED (Vol. 200, pp. 17-24) AFFECTIVE COMPUTING = the study and development of systems and devices that can recognize, interpret, process, and simulate human emotions.
- 17. Modelling social interactions Martinez Maldonado, Roberto, et al. "Orchestrating a multi-tabletop classroom: from activity design to enactment and reflection." Proceedings of the 2012 ACM international conference on Interactive tabletops and surfaces. ACM, 2012.
- 18. Learning Theories & Artificial Intelligence Examples
- 19. Behaviorist Learning with AI Learning as a response to environmental stimuli posits that behaviors can be shaped through reinforcement and feedback MILKI-PSY dance tutor (Mat Sanusi et al., 2024) Immediate feedback Reinforcement learning Continuous practice Psychomotor tasks
- 20. (Limbu, Schneider, et al., 2018; Limbu et al., 2019) AI-based Feedback provision in the Calligraphy Tutor
- 21. Constructivist learning with AI Learners actively construct their own understanding and knowledge of the world MILKI-PSY human robot interaction (Keller, Majonica, et al., 2024) Problem-solving Intelligent tutoring systems Prior Knowledge activation Student models Scaffolding Reflection Learning analytics
- 22. Social constructivist learning with AI Learning happens within social contexts and is profoundly influenced by cultural beliefs and attitudes AR Tutor (Iren, 2024) Collaborative learning CSCL Project-based learning Conversation theory Learning occurs through conversation
- 23. Online class Non-verbal cues Action Units Gestures (head movements) Learning Centered Affective States Feedback to the teacher (a) (b) (c) (d) (Shingjergji et al., 2022; Iren, Shingjergji, et al., 2023; Bottger et al., 2022; Zhang et al., 2021; Iren, Yildirim, et al., 2023) Teacher Support with Affective Computing In online learning, teachers are more disconnected from their learners. AI-based affective computing can inform the teacher about struggling learners
- 24. Risks
- 25. Whom do you recognise? https://miro.medium.com/max/1400/1*xOJ3mjjsIfud7GPS7XNJIQ.png
- 26. Image: Jarodzka, 2023 Real references Fake references
- 27. Reported ML biases, failures, and issues Gender bias in automated translation (here turkish english) Turning the neutral pronoun into a "he" when in the same sentence as "doctor" or "hard working," and a "she" for "lazy" and "nurse. Google Allo suggested man in turban emoji as response to a gun emoji Amazon Echo starts playing random music without owner at home IBM Watson for oncology cancelled after unsafe treatment recommendations Microsoft Chatbot trained with hate speech Apples face ID beaten by mask Uber self-driving vehicle killing a pedestrian Amazons recruitment AI is gender-biased https://www.techrepublic.com/article/the-10-biggest-ai-failures-of-2017/ https://medium.com/syncedreview/2018-in-review-10-ai-failures-c18faadf5983 https://www.lexalytics.com/lexablog/stories-ai-failure-avoid-ai-fails-2020
- 28. Reasons for these problems Training data already biased ML algorithms are bad at generalising Algorithms can contain bias (models, observation space, feedback functions, rewards, learning rates ) Structure and knowledge is hard to pre-encode in ML algorithms ML badly represents uncertainty and does not handle it well ML does not explain ist decision (black-box) Small changes in input data can destroy ML results Applying ML depends largely on expert knowledge
- 29. Image: Milad Fakurian on Unsplash Facts Training data Source code Algorithms Quality control Ownership Topicality Control
- 31. (Agarwal et al., 2024, in review; Liesenfeld et al., 2023) Ethics of AI in Education
- 32. How do we take control? Openness is a key Open Source Open Model Open Data Independent control is another Training Data Algorithms Quality control processes Image: Duy Pham auf Unsplash This requires collaboration of individuals, institutions, and politics!
- 33. Conclusion Education faces challenges: population, life-time, global challenges Traditional education fails to scale enough Technology is needed to drive the change Open challenges are in precision, consistency, ethics Effort is required to ensure ethical and useful use of AI in education
- 34. Thank you! Your Prompts? Roland Klemke Empower Webinar Week AI in Education: Assessment, Ethics, and the Future of Learning 03 December 2024 Roland.Klemke@ou.nl