Hybrid intelligence: Human–AI coevolution and learning

British Journal of Educational Technology, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 9, 2025

Artificial intelligence (AI) is becoming increasingly ubiquitous in all areas of life. A concrete example the urgent need to develop new educational practices post-pandemic world, as called for OECD 2021 Report on Digital Education (OECD, 2021). Furthermore, it has been estimated that more than half tasks associated with almost jobs have exposed AI, and at least 19% 80% exposed. As such, potential AI enhancing expert work by supporting decision making, automating routine fostering innovation solve complex problems become evident. However, also introduces significant risks, such loss human expertise, ethical concerns overreliance automated systems (Filippucci et al., 2024). The emerging paradigm hybrid (HI) can offer promising solutions strike a balance between mitigating negative impacts disruption learners' workers' re-skilling upskilling world growing population societal challenges (Akata 2020). current data-driven are still too narrow help humans, lacking social emotional restricted their ability produce realistic applicable results (Cui & Yasseri, Humans unique we capable creative flexible thinking—connecting thinking action long-term aims, values purposes—and judge activities purposes from an point view. While HI education research its early stages, this special section some works, especially highlighting human–AI coevolution learning, which be expected impact education, well-being quality This highlights ways advance multidisciplinary cross-section learning sciences computer generate future AI-based various fields education. We hope pushes forward discussions role digital data multimodal analytics (Cukurova 2020) methods technologies (Raković 2023). begin introducing concept HI, exploring how effectively integrated identifying key directions necessary field. introduce state art will contribute practice. Finally, discuss research. Hybrid field seeks bridge gap AI. By combining strengths both humans machines, aims create outperform either or machines working independently. In other words, combine through coevolutionary processes collaborate, learn reinforce each (Järvelä, Zhao, difference designed independently perform normally require intelligence, perception (Russell Norvig, 2010). Despite advancements many remain opaque (ie, 'black box') models, complicates collaboration due lack transparency interpretability (Rosé 2019). creates trust effective interaction, particularly human-centric environments. argue successful development requires fundamentally address core challenges. outperforms like pattern recognition machine lags essential attributes collaboration, adaptability, responsibility explainability. These uniquely qualities crucial teamwork, making navigation interactions, where struggles match performance. To overcome these limitations, paradigms cannot ultimate solution. Instead, integrate abilities, partnership one another. Achieving comprehensive, approach, incorporating insights cognitive science psychology adaptive, responsible transparent. transforming mere tool into trusted partner real-world applications. achieve this, not only technological required, but so critical reexamination align values, ethics objectives. there consensus about importance paradigm, robust theoretical conceptual framework understanding augmented HI. facilitate information exchange mutual data-processing computational models. Since currently under development, empirical evidence scarce. thus identify several themes greater attention global community highlight topics addressed papers featured section. Decades shown processor, superior humans. Currently, interest artificial general (AGI), across tasks, even superintelligence (ASI), exceed it. recently highlighted within academia broader public (Ororbia Friston, achieving AGI major goal research, raises philosophical, technical questions regarding safety, control society. idea ASI existential could uncontrollable surpass authority, leading scenarios humanity may struggle manage direct actions. Nevertheless, remains phenomenon encompasses abilities (Bereiter Scardamalia, 1993). It includes learn, understand, reason, make decisions adapt situations. strength lies our plan, monitor own (Zimmerman, 1989), coupled nuanced comprehensive context. metacognitive capacity, psychological regulate one's thoughts behaviours, fundamental adaptation. Empirical skilful advanced learners use skills guide studying, makes them agentic (Flavell, 1979). recent if students Open applications, ChatGPT, think automatically agency detrimentally affected (Darvishi significantly potentially decreasing impacting cognition (Ahmad 2023) do exercise monitoring strategies. believe often limited context cause–effect relationships, operates primarily correlations rather deductive reasoning. result unintended consequences dilemmas. Hence, paradigms. Ethically, oversight ensure promote fairness held accountable actions, preventing harmful biases unethical outcomes (Nikolinakos, Cognitively, falls short, requiring common sense reasoning, contextual adaptability novel ambiguous situations handle (Arslan, Most importantly, bring empathy, capacity understand respond contextualized cues, enabling meaningful health care, customer service. integrating systems, trustworthy, reliable emotionally intelligent efficiently resonate needs values. Cukurova (2024) introduced vision builds interplay He multi-dimensional view AI's emphasizing intricate processes, stressing relationship automatization. dynamic interactions Cukurova's interaction rooted in, upon, differences processing. extremely useful recognizes stages contributing systems: externalization cognition, internalization models influence mental extension via tightly systems. There numerous conceptualizing developing For example, synergy environments, share explain goals strategies 2020)? integration brings up issues, including transparency, accountability assurance perpetuate biases. section, two important themes: Data algorithms assisting mechanisms interact widely employed assess (Azevedo Gašević, 2019; Blikstein Worsley, 2016; Nguyen made possible channels modalities, providing unobservable central learning. draws diverse range data, log files online environments (Montgomery Premlatha 2016), physiological (Lämsä 2024), (Kwon 2014) self-reported measures. streams arrive learner's state, capturing trajectories. Such enables deeper exploration learner affects engagement, offering richer picture provided traditional assessments application real-time support, augmenting (Blikstein 2016). instance, tutoring provide tailored feedback, scaffold adjust difficulty levels optimize performance (Behera support skill acquisition, self-regulated (SRL), encouraging reflect progress, set accordingly. enable researchers merge efficiency. techniques now analysing vast facial expressions (Zhao 2023), body movements (Lyu gaze direction, speech patterns responses, interpret emotions, preferences decision-making 2022). techniques, deep natural language processing (NLP), allow detect behaviour previously hidden, unprecedented states individuals. New methods, emotion emotions intentions visual auditory cues. combination expressions, postures patterns, infer states, interpersonal dynamics engagement. further enhanced heart rate, skin conductance brain activity (Lian 2024; Sun result, interpretation experiences, paving way intelligent, responsive collaborate deeper, intuitive manner. Recent utilizing capabilities derive (Cukurova, Järvelä, Nguyen, Hadwin, 2023; Molenaar NLP vision, being used uncover were undetectable, facilitating holistic processes. progress enhance effectiveness interventions augment thereby promoting context, creativity judgement, supports amplifies capabilities. Identifying predicting pose researchers. timely question is: What systems? Human inherently complex, encompassing interconnected cognitive, metacognitive, motivational affective conceptualized relation SRL theory 1989). goal-directed knowledge, competence, because adaptable nature, valuable settings functioning. Central who actively realization goals. continuously achievement determine revision cycle able self-regulate and, optimal cases, successfully (Winne 1998). Technological past decade expanded opportunities sources spawned examining informing design sensitive individual they behavioural, demands individually collaboratively (Molenaar, An advantage approaches very precise individuals collected speeds sophistication beyond analysis. risk heavily analytical divorced strong theoretically empirically informed understandings complexity (Winne, Adaptive (ALTs) empowered dimensions study SRL, affordances (Dever methodological challenges, concerning design, collection solved progress. Analytically, must foundations. recently, Järvelä Hadwin theory-driven trigger regulation advances examine during task teamwork. Another issue strategies, presents Current predictive accuracy needed when engaging struggling specific strategy (Nguyen relying types, clickstream surface-level misses complicated factors compose personally situational belief prediction highly variable. When temporality crucial. Late predictions miss intervention, while premature ones unnecessary early. Individual internal conditions, prior knowledge affect personalized adaptive (Järvelä Therefore, use, sophisticated methodologies account differences, varying granularity guided theories. stage marked collaborative algorithms, interactive inputs real time Wambsganss al. proposed approach students' argumentation employing learning-based modelling. Their demonstrated behavioural modelling improved objective domains, outperforming scripted modelling, static Additionally, indicated showed large effect, simple exhibited medium equal regardless expertise levels. Contemporary focusing collaborations line investigate aspects people engage technologies. studies recognized establishing appropriate component (Mehrotra Appropriate ensures users neither over-rely nor disregard capabilities, striking enhances collaboration. Researchers explainability user experience building Yadollahi users' technology style, willingness addressing factors, intuitive, user-friendly seamless Looking ahead, poised focus NLP, allowing communication well Multidisciplinary essential, science, empathetic socially aware. problem solving, leveraging accomplish alone. broad academic fields, technologies, sciences, discussion human-driven three different other, theme underscores latest aimed bridging communicative divide machines. embedded everyday life, Several contributions area innovative understanding, improving awareness accurately adapts dynamically evolving contexts. Fan al.'s paper 'Beware laziness: Effects generative motivation, performance' assessed 117 university students, multichannel motivation analysed. revealed following: (1) received forms no post-task intrinsic (2) AI-assisted group outperformed essay score improvement, gain transfer different. noteworthy ChatGPT dependence 'laziness'. pioneering sheds light assistance way. Naik 'Providing reflection instructions using models' contributes previous showing modern (LLMs) student prompting reflection. provides generating contrasting cases comparing differ applicability adult learners. particular, demonstrates LLMs solutions. Evidence classroom intervention indicates LLM-enabled positive suggests training instructors tools beneficial, lower knowledge. opened pathways depth detail (Chen 2022; fusion, proving powerful uncovering difficult grasp. AI-enhanced impactful (Dang Ouyang Machine process datasets subtle shifts regulation, learners, relate outcomes. explore negotiate roles, process, AI-mediated might productivity shared (Edwards, Lämsä, utilized reported dynamics. 'Enhancing iterative game' Hare examined AI-driven system first-year undergraduate electrical engineering course. integrates multi-agent reinforcement gamified, human–computer (HCI) environment tutoring. benefits perceptions engagement system, experiences. emphasize human-centred solely automation. Moving forward, authors plan refine metacognition self-reflection broader, population. Yu 'Using peer feedback teacher video-based learning' explored improve (PF) teachers. PF suffers inconsistencies teaching expertise. developed (HIF) uses categorize summarize PF, supplemented involving 58 preservice mathematics teachers compared HIF PF. led better revision, reflective reflections. analysis explores co-creative together ideas. recommendations, predictions, outputs, apply final decisions. teacher–AI available Cohn investigated AI-generated timeline teachers' difficulties build (STEM+C). combined extending contexts, nuances domain-specific information. They argued powers STEM+C case formative students. leverage timeline. discussed events teachers: cross threshold intervention. should intervene feedback. concluded teacher–researcher–AI overcoming reciprocal ways. First, second, iteratively Giulia 'Hybrid intelligence: Lessons learned embodied experience' teaching. With deeply classrooms, designers investigating receive simultaneously coordinate sought emerge intertwined instruction perceive approach. analysed AI-powered MOVES-NL, was integer arithmetic. MOVES-NL whole-body movement immediate room-scale environment. findings practical recommendations educators aiming enriching ongoing discourse Weber legal writing skills: environment' emphasis writing. benefits, actual contexts frequently short expectations, scarcity validating effectiveness. issue, whether assist law mistakes crafting organized persuasive texts. experiment conducted 43 evaluate effects examination performance, themselves, written process. errors skills, helping recognize SRL. Lee 'Unveiling reading behavioral cues: approach' applied merged surge precise, consistent scalable limitations faced criticism efficacy interpreting signals. model helps evaluators features higher-order (HOTS) versus lower-order (LOTS). method feature interventions. frameworks insights, pinpointed cues related involves tasks. excel creativity, empathy uncertainty, proficient amounts recognizing performing repetitive high accuracy. case, leverages Edwards, members' (SSRL). Meta-skills, SSRL go acquisition represent competences strengthened trained schools age Detecting groups challenge, time, agents possible. (MAI) agent prompt raise group-level aim SSRL. methodology designing MAI presented evaluation initial prototype. implemented quantitative measures lexical alignment speakers MAI's prompts SSRL, questionnaire measure MAI, interviews qualitative perceptions. first prototype did hoped, guidelines revising next 'AI-powered vocabulary primary school Yun culmination exploratory mixed evaluated ARCHe, four functions: automatic pronunciation, handwriting (3) scoring student-generated sentences (4) recommendations. Conducted 140 six Singaporean schools, functions pre post-tests, surveys, student-created artefacts. test scores, reflecting Chinese characters. Positive ARCHe's pronunciation recommendation correlated increased favourable views higher engaged sentence creation gains. suggest holds great promise exciting advancements. evolve, seamless, resulting aligned needs. harnessing Wiedbusch, Unlike mainly automates emphasizes preserving agency. necessitates coevolution. reality (MR) revolutionize blending virtual, physical MR allows content, immersive settings, actions personalizing (Yannier thinking, solving scenarios. Research role, ensuring meets boosts making. direction focuses gathering, sophisticated, responses becomes central. Overreliance undermine self-regulation. prioritize foster adapting Molenaar, closely promotes inclusive equitable needs, freedom choice privacy. Future responsible, transparent lifelong cater challenge compromised. replace intelligence. funded Council Finland University Oulu profiling project Profi7 Intelligence—352788.

Language: Английский

---

Pedagogical Agents Communicating and Scaffolding Students’ Learning: High School Teachers’ and Students’ Perspectives

Computers & Education, Journal Year: 2024, Volume and Issue: 222, P. 105140 - 105140

Published: Aug. 23, 2024

Pedagogical agents (PAs) communicate verbally and non-verbally with students in digital virtual reality/augmented reality learning environments. PAs have been shown to be beneficial for learning, generative artificial intelligence, such as large language models, can improve PAs' communication abilities significantly. K-12 education is underrepresented technology research teachers' students' insights not considered when developing PA communication. The current study addresses this gap by conducting analyzing semi-structured, in-depth interviews eleven high school teachers sixteen about their expectations capabilities. interviewees identified relational task-related capabilities that a should perform effectively scaffold learning. simultaneously affirmative induce immediacy, foster the relationship engagement PA, support management. Additionally, described activities technological aspects designing conversational PAs. showed applied human-to-human scripts outlining desired characteristics. offers novel recommendations researchers developers on communicational, pedagogical, must communicative discusses contributions human–machine education.

Language: Английский

---

Adaptive support for self‐regulated learning in digital learning environments

British Journal of Educational Technology, Journal Year: 2024, Volume and Issue: 55(4), P. 1281 - 1289

Published: May 5, 2024

Abstract A core focus of self‐regulated learning (SRL) research lies in uncovering methods to empower learners within digital environments. As technologies continue evolve during the current hype artificial intelligence (AI) education, theoretical, empirical and methodological nuances support SRL are emerging offering new ways for adaptive guidance learners. Such affordances offer a unique opportunity personalised experiences, including interventions. Exploring application adaptivity enhance is an important area that requires further attention. This editorial introduces contributions seven papers special section on These explore various themes related enhancing strategies through technological interventions, valuable insights paving way future advancements this dynamic area.

Language: Английский

---

Scaling goal-setting interventions in higher education using a conversational agent: Examining the effectiveness of guidance and adaptive feedback

Published: Feb. 21, 2025

Language: Английский

---

Using Serious Game Techniques with Health Sciences and Biomedical Engineering Students: An Analysis Using Machine Learning Techniques

Information, Journal Year: 2024, Volume and Issue: 15(12), P. 804 - 804

Published: Dec. 12, 2024

The use of serious games on virtual learning platforms as a support resource is increasingly common. They are especially effective in helping students acquire mainly applied curricular content. However, process required to monitor the effectiveness and students’ perceived satisfaction. objectives this study were (1) identify most significant characteristics; (2) determine relevant predictors outcomes; (3) groupings with respect different game activities; (4) perceptions usefulness simple complex activities. We worked sample 130 university studying health sciences biomedical engineering. activities Moodle environment, UBUVirtual, monitored using UBUMonitor tool. degree type explained differing percentages variance results assessment tests (34.4%—multiple choice [individual assessment]; 11.2%—project performance [group 25.6%—project presentation assessment]). Different clusters found depending group algorithm applied. Adjusted Rang Index was appropriate each case. student satisfaction high all cases. they indicated being more useful than resources for practical content both engineering degrees.

Language: Английский

---

Analytical Approaches for Examining Learners’ Emerging Self-regulated Learning Complex Behaviors with an Intelligent Tutoring System

Lecture notes in computer science, Journal Year: 2024, Volume and Issue: unknown, P. 116 - 129

Published: Jan. 1, 2024

Language: Английский

---

Capítulo 17: Potenciando habilidades del Siglo XXI: un enfoque de diseño instruccional basado en taxonomías

Published: Sept. 30, 2024

La investigación se enfocó en fortalecer competencias del siglo XXI mediante un diseño instruccional basado taxonomías. Destaca la importancia de como el pensamiento crítico, resolución problemas y creatividad sociedad actual. Los Sistemas Tutores Inteligentes (STI), impulsados por Inteligencia Artificial (IA), son fundamentales educación al ofrecer instrucción adaptativa evaluar progreso estudiante. El estudio propone modelo STI para estas competencias, con enfoque cualitativo dos fases. primera, una revisión bibliográfica usando hermenéutica seleccionar pedagógico definir XXI. segunda, empleó Lenguaje Unificado Modelado (UML) ontologías diseñar visualmente pedagógico. Usa Test Kolb estilos aprendizaje e implementó casos. Las actividades diseñaron según niveles complejidad Taxonomía Bloom revisada. metacognición relación entre metacognición. discusión abordó necesidad estrategias pedagógicas basadas IA marco ético uso educación.

---

The Expert and Planning Module of an Intelligent Pedagogical Agent for the Development of Critical Thinking

SN Computer Science, Journal Year: 2024, Volume and Issue: 5(8)

Published: Nov. 22, 2024

Language: Английский

---