Hybrid intelligence: Human– <scp>AI</scp> coevolution and learning

Abstract

Artificial intelligence (AI) is becoming increasingly ubiquitous in all areas of life. A concrete example is the urgent need to develop new educational practices in the post-pandemic world, as called for in the OECD 2021 Report on Digital Education (OECD, 2021). Furthermore, it has been estimated that more than half of the tasks associated with almost half of all jobs have been exposed to AI, and at least 19% of tasks associated with more than 80% of jobs have been exposed. As such, the potential of AI in enhancing expert work by supporting decision making, automating routine tasks and fostering innovation to solve complex problems has become increasingly evident. However, AI also introduces significant risks, such as the loss of human expertise, ethical concerns and the potential for overreliance on automated systems (Filippucci et al., 2024). The emerging paradigm of hybrid intelligence (HI) can offer promising solutions that strike a balance between mitigating the negative impacts of AI disruption and supporting learners' and workers' re-skilling and upskilling in a world of growing population and societal challenges (Akata et al., 2020). However, current data-driven AI systems are still too narrow to help humans, lacking in social and emotional intelligence and restricted in their ability to produce realistic and applicable results (Cui & Yasseri, 2024). Humans are unique in that we are capable of creative and flexible thinking—connecting thinking and action to long-term aims, values and purposes—and we can judge activities and purposes from an ethical point of view. While the HI in education research is still in its early stages, this special section introduces some early works, especially highlighting human–AI coevolution and learning, which can be expected to impact education, well-being and quality of life. This special section highlights ways to advance multidisciplinary research at the cross-section of the learning sciences and computer sciences to generate future AI-based solutions for various fields such as education. We hope that this special section pushes forward discussions on the role of digital data in multimodal analytics (Cukurova et al., 2020) and AI-based methods in future educational technologies (Raković et al., 2023). We begin by introducing the concept of HI, exploring how human intelligence can be effectively integrated with AI and identifying key research directions necessary to advance this emerging field. We introduce the state of the art in HI research and how this special section will contribute to research and practice. Finally, we discuss future directions in HI research. Hybrid intelligence is an emerging field that seeks to bridge the gap between human intelligence and AI. By combining the strengths of both humans and machines, HI aims to create systems that outperform either humans or machines working independently. In other words, HI aims to combine the strengths of both humans and machines through their coevolutionary processes to collaborate, learn from and reinforce each other (Järvelä, Zhao, et al., 2023). This is a key difference to AI, which is designed to work independently to perform tasks that normally require human intelligence, such as perception and learning (Russell & Norvig, 2010). Despite significant advancements in AI, many systems remain opaque (ie, ‘black box’) models, which complicates collaboration between humans and machines due to a lack of transparency and interpretability (Rosé et al., 2019). This gap creates challenges in trust and effective interaction, particularly in human-centric environments. We argue that the successful development of HI requires fundamentally new solutions to address core AI challenges. While AI outperforms humans in tasks like pattern recognition and machine learning, it lags in essential human attributes such as emotional intelligence, collaboration, adaptability, responsibility and explainability. These uniquely human qualities are crucial for effective teamwork, ethical decision making and the navigation of complex social interactions, areas where AI still struggles to match human performance. To overcome these limitations, current data-driven AI paradigms cannot be the ultimate solution. Instead, HI aims to integrate AI with uniquely human abilities, fostering a partnership where both systems reinforce one another. Achieving this requires a comprehensive, multidisciplinary approach, incorporating insights from fields like cognitive science and psychology to develop AI systems that are more adaptive, responsible and transparent. These advancements are essential in transforming AI from a mere tool into a trusted partner in real-world applications. To achieve this, not only is technological innovation required, but so is a critical reexamination of how we align machine intelligence with human values, ethics and objectives. While there is a growing consensus about the importance of the HI paradigm, we still lack a robust theoretical and conceptual framework for understanding human intelligence and learning processes that can be effectively augmented by HI. We also lack an understanding of how to facilitate the information exchange and mutual learning between AI and humans with new data-processing methods and computational models. Since HI systems and solutions are currently under development, empirical evidence of HI is scarce. We thus identify several research themes that require greater attention from the global research community and highlight topics addressed by the papers featured in this special section. Decades of research have shown that AI, as an information processor, is superior to humans. Currently, there is growing interest in artificial general intelligence (AGI), which aims to match human intelligence across all tasks, and even artificial superintelligence (ASI), which aims to exceed it. These topics have recently been highlighted within academia and the broader public (Ororbia & Friston, 2023). While achieving AGI is a major goal of AI research, it also raises significant philosophical, ethical and technical questions regarding its safety, control and potential impact on society. Furthermore, the idea of ASI raises critical ethical and existential risks, as it could become uncontrollable and surpass human authority, leading to scenarios where humanity may struggle to manage or direct its actions. Nevertheless, intelligence remains a complex phenomenon that encompasses several abilities (Bereiter & Scardamalia, 1993). It includes the ability to learn, understand, reason, make decisions and adapt to new situations. A key human strength lies in our ability to plan, monitor and control our own learning processes (Zimmerman, 1989), coupled with a nuanced and comprehensive understanding of context. This metacognitive capacity, the psychological ability to monitor and regulate one's thoughts and behaviours, is fundamental to effective learning and adaptation. Empirical research has shown that skilful and advanced learners use metacognitive skills to guide their thinking and studying, which makes them agentic learners (Flavell, 1979). However, recent research has shown that if students use Open AI applications, such as ChatGPT, to think automatically in their tasks, their agency may be detrimentally affected (Darvishi et al., 2024). AI significantly impacts human decision making, potentially decreasing it and impacting cognition (Ahmad et al., 2023) if students do not exercise their own metacognitive monitoring and strategies. We believe that data-driven AI is often limited in its understanding of context and cause–effect relationships, as AI operates primarily on correlations rather than deductive reasoning. This can result in unintended consequences in decision making and ethical dilemmas. Hence, we need human intelligence to guide and regulate current AI paradigms. Ethically, human oversight is crucial to ensure that AI systems align with societal values, promote fairness and are held accountable for their actions, preventing harmful biases or unethical outcomes (Nikolinakos, 2023). Cognitively, human intelligence is necessary to bridge the gap where AI falls short, particularly in areas requiring common sense reasoning, contextual understanding and adaptability to novel or ambiguous situations that AI systems struggle to handle effectively (Arslan, 2024). Most importantly, humans bring empathy, emotional intelligence, and the capacity to understand and respond to contextualized and complex emotional cues, enabling more meaningful interactions, especially in areas such as health care, education and customer service. By integrating these human qualities into AI systems, we can create more trustworthy, reliable and emotionally intelligent hybrid solutions that not only perform tasks efficiently but also resonate with human needs and values. Cukurova (2024) introduced a vision for HI that builds on the interplay of learning, analytics and artificial intelligence in education. He introduced a multi-dimensional view of AI's role in learning and education, emphasizing the intricate interplay between AI, analytics and learning processes, especially stressing a relationship between human control and automatization. Since HI requires effective dynamic interactions between humans and AI, Cukurova's (2024) conceptual framework for human–AI interaction in education is rooted in, and builds upon, the significant differences between human intelligence and artificial information processing. This framework is extremely useful for HI research, as it recognizes the stages contributing to the development of hybrid human–AI systems: the externalization of human cognition, the internalization of AI models to influence human mental models and the extension of human cognition via tightly coupled human–AI HI systems. There are numerous challenges in conceptualizing and developing HI systems. For example, how do we develop AI systems that work in synergy with humans, how do these systems learn from and adapt to their environments, and how can humans and AI share and explain their goals and strategies to each other (Akata et al., 2020)? The integration of humans and AI also brings up ethical issues, including the need for transparency, accountability and assurance that AI systems align with human values and do not perpetuate biases. In this special section, we highlight two important research themes: Data and algorithms assisting HI research and understanding core human learning mechanisms to adapt and interact within HI systems. Data and algorithms have been widely employed to understand and assess human learning and intelligence (Azevedo & Gašević, 2019; Blikstein & Worsley, 2016; Nguyen et al., 2020). The development of learning analytics has made it possible to integrate data from various channels and insights into the cognitive and emotional processes to human For example, learning analytics on a of including from learning et al., 2019; et al., data et al., social interactions et al., and These data have been integrated to at a comprehensive understanding of a complex and dynamic learning integration a of behaviours, and a of learning processes than that by (Raković et al., 2023). Furthermore, learning analytics the of advanced algorithms to offer both human learning and decision making & Worsley, For intelligent systems use these insights to learning tasks and to et al., 2020). is crucial in fostering such as learning learners to on their goals and adapt their strategies To more effective information exchange and mutual learning between humans and AI, are exploring computational models that human contextual understanding with machine For example, machine learning are capable of of multimodal such as et al., et al., and to human interactions, and processes et al., These including learning and AI systems to in human that may have been insights into the cognitive and emotional of machine learning such as computer vision and recognition systems, are enabling AI to human and more effectively by and For example, by a of and AI systems can emotional and cognitive This can be by integrating data such as or to a understanding of human et al., et al., 2024). As a these methods for a more nuanced of human the for more AI systems that can with humans in a more advancements in learning analytics and the learning sciences are increasingly AI to more nuanced insights from educational data & et al., 2023). AI such as machine learning, and computer are to that a more understanding of learning This aims not only to the of learning but also to human intelligence, the development of HI. While humans and ethical AI and human cognitive and human learning processes significant challenges for A are the core human learning mechanisms that can and need to be to adapt and interact in dynamic hybrid learning is a of and These processes have been in to (Zimmerman, is a action to develop new skills or and of its it is not only in educational but also across various areas of human is the agency of learners goals and control their activities and strategies in of the of these In other words, learners monitor goal to the need for and to control their learning actions. This of monitoring and control makes learners to their learning in to adapt their learning to achieve their goals & advancements in the have to use multimodal data and advanced analytics to understand complex human learning processes (Azevedo & Gašević, 2019). These advancements have also emerging research directions the potential of these data in the of AI systems that are to differences in and can adapt to and learners as manage the and emotional of learning both and of these is that data on in action can be and integrated at and of that are the capacity of human However, a is that these data-driven processes have become from and of the of human learning 2020). learning technologies by AI have introduced new to the of that impact how learners regulate their own learning et al., 2024). However, these also introduce particularly research paradigms and data that need to be for HI research to ensure that the to align with its theoretical For and (2024) introduced a framework that can guide multimodal and in research to learning processes complex work and is that learning processes, such as significant challenges. models are still in development and often lack the to identify a is in or with a et al., For example, on limited data such as results in the of only interactions and cognitive and emotional is by the interaction between and emotional that unique and systems and make understanding and learning processes, is for may result in and that are too differences in learners' such as and highlighting the need for and models & 2024). the of learning processes, coupled with the challenges in identifying and their more models and in HI research. These for of and the dynamic interplay between cognitive and emotional processes by learning The current of HI research is by significant advancements in systems and the development of AI models that can adapt to and learn from human in (Akata et al., 2020). For example, et (2024) an to skills by dynamic machine research that dynamic significantly learners' abilities across various methods like and the results that this dynamic is more effective than for both complex tasks, where it a and tasks, where it a thus to learners of their HI research is increasingly on the human that influence interactions and between humans and AI systems for HI. This of research aims to how psychological and social the with AI For example, have that trust is a crucial of effective human–AI interaction et al., 2024). trust that on AI systems their a balance that are exploring how like transparency, and contribute to this trust et al., et al., 2024). are how such as a with or their cognitive impact their to with AI. By understanding and these human the goal is to create HI systems that are more and capable of fostering collaboration between humans and AI. the future of HI research is to on the integration of AI systems with human cognitive This includes advancements in for more between humans and AI, as as the development of AI models that can understand and adapt to human and social collaboration will be combining insights from fields such as cognitive psychology and ethics to create AI systems that are not only intelligent but also and The papers in this special section discuss key of HI in decision making, and other and cognitive tasks by the strengths of both humans and machines to achieve outcomes that could The papers in this special section have integrated from a of such as advanced learning the learning education, learning analytics and the computer thus a multidisciplinary on AI. We identify themes: AI for assisting humans and machines in understanding each AI to insights into human learning processes and human–AI collaboration research. This the advancements in AI at the and cognitive between humans and As AI systems become increasingly in enabling and interactions has become within this highlight designed to mutual on contextual and learning These will AI systems to human needs more and complex cognitive processes, such as decision making and fostering a human–AI partnership that to and et of metacognitive of artificial intelligence on learning processes, and and their learning, and data and The results the learners of learning difference in and the the other by of but their and not significantly is particularly is that AI technologies such as may promote learners' on and potentially metacognitive work is in that it on how AI can human decision making and cognitive in a negative et (2024) in learning to research by how models can learning work by The new insights into how and the in their and for In the an where generate to from their that this has a impact on Furthermore, the that for to effectively use as to create could be particularly especially in supporting students with of The integration of AI into educational research has up new for understanding human learning processes in greater and et al., & 2023). AI such as machine learning, and multimodal are to be in insights that to For example, learning analytics can be particularly in understanding learning et al., et al., 2023). learning algorithms can complex from learning to identify in from and how these to learning These insights to how learners how social interactions contribute to the learning and where and et al., 2024). Furthermore, AI can be to assess various emotional and cognitive of human learning processes at both the and The in this special section have the potential of AI to generate insights into these learning and learning with in an intelligent educational by et the development of an educational designed to particularly in an The learning in a interaction to automated The highlights the of which and with the leading to more learning the results the importance of combining AI with to educational outcomes rather than on the to the by and in as as exploring the long-term impacts and of the on a more et in their hybrid intelligence to for in how HI can the quality of in learning for is crucial for but often from due to of To address this, this work an HI that to and by AI-based multimodal data A with The results that to thinking and more attention in This that integrating with human can the quality and impact of in learning environments. HI research various models of collaboration, including decision systems, machine learning and systems where humans and AI work to generate new AI systems can insights or and humans can these contextual understanding and make For example, collaboration can the work of human by for new insights into learning and that not to 2019). et (2024) how collaboration between and an multimodal can guide and as work to computational models and solve problems in sciences tasks This humans and AI, understanding of in learning the of social interaction and that combining the of humans and AI can address challenges. et (2024) in their that can use the to guide for Furthermore, can to help the multimodal two of interest to students a and the point of These are where a or to effective et (2024) that collaboration can help students in in two can use the multimodal to guide for and can the multimodal et (2024) from an learning the collaboration between humans and AI in AI becoming more integrated in educational are how students learn from both AI systems and human However, there is a need for a more comprehensive understanding of how to effectively interactions between and AI in educational environments. This to identify the challenges and that AI is with and to assess how students this hybrid The a an called which to the learning of is an tool that and in a The offer insights and for and to the integration of in the on the role of HI in education. et (2024) The impact of in a hybrid intelligence learning how humans and AI in learning, with a on While learning systems offer promising their in human–AI falls of and there is a of empirical their To address this the in an HI learning can students in learning from their and more and A field with students to the of on the students their work and the The results that incorporating to significantly highlighting the role of AI in students their and fostering et (2024) cognitive processes in digital through A hybrid intelligence an HI research that human with machine intelligence to cognitive processes digital by The recent in the use of AI within learning analytics aims to achieve a more and understanding of learning, the of human intelligence However, machine intelligence has for its lack of contextual and in complex human and social To overcome these issues, the a machine learning that human identify which are essential in learners' use of thinking skills thinking skills This insights for future and the of educational By and expert the to learners' attention which are crucial for understanding their cognitive HI research the integration of human intelligence with AI to decision making and other cognitive Cukurova (2024) that humans in and complex decision making under AI systems are in of and tasks with In the HI the synergy of these strengths and the unique of both humans and AI. and et (2024) a on human–AI collaboration to in learning such as & and unique human that need to be and in in the of AI. These are the skills require to become agentic which are also essential in and successful and supporting in is a especially in but and et (2024) that HI such as AI can make this designed a metacognitive AI that can and students to their metacognitive with the of their for the and empirical results on their of between and the of on the partner as a of of and to the context of these While their not facilitate as their for for the learning for by et (2024) is the of an methods that introduced and an learning with core on of and with students and across the how these to and learning outcomes through and and The results an understanding of and of and with emotional on to cognitive students in greater learning The offer insights for and learning regarding AI's role in learning and directions for future research. The future of HI research for As AI technologies their integration with human intelligence will become increasingly in systems that are more and with human abilities, values and societal A is the synergy between human intelligence and AI to create learning environments. This developing AI systems that with human fostering metacognitive skills and understanding (Azevedo & 2023). AI, which tasks, HI mutual interaction and learning, enhancing human cognitive processes This new data-processing and computational models that facilitate effective information exchange and In HI can education by augmented and environments. learners to interact with both digital and real-world in In these HI can adapt to learners' and cognitive educational through data et al., 2024). This students to their critical and collaboration skills in realistic in this aims to AI's it not only learners' needs but also emotional and decision crucial research on the ethical integration of AI in education. As the tasks of and multimodal data AI solutions become more from to emotional cues, the of ethics even more on automated could essential human skills like decision making and research AI systems that human agency and to needs (Järvelä, & 2023). By incorporating emotional intelligence and empathy, AI can align more with human values. The ethical integration of AI in education and other fields and learning and solutions that adapt to their of and agency and data need to how HI can promote and accountable AI solutions that learning and to As AI systems become more integrated into learning environments, a major is that human agency is not HI research promote the of AI that rather than human decision making, and emotional This research has been by the of and the of Hybrid

Topics & Concepts