Canadian Academy Discover, Publish, Thrive
This review explores collaborative intelligence, emphasizing the synergy between humans and AI for enhanced performance. It introduces the concept and outlines criteria for evaluating AI systems’ ability to enable collaboration. A systematic review of 1,250 AI applications from 2012 to 2021 identified 16 systems meeting these criteria, showcasing the potential for collaborative human-AI teams to improve outcomes across various domains such as efficiency, quality, creativity, safety, and user satisfaction.
Industry 4.0 is marked by the adoption of automated systems, IoT, big data, and cyber-physical systems for enhanced productivity. Industry 5.0 emphasizes human-centric technology development, focusing on collaboration between humans and smart machines rather than mere automation. Collaborative intelligence, involving human-AI systems working together, is shown to significantly improve work outcomes and is projected to increase revenues for organizations. However, despite strong theoretical support, empirical research on collaborative intelligence applications is lacking. This paper aims to fill this gap by systematically reviewing existing applications to assess their technological and economic feasibility.
Collaborative human-AI interactions are increasingly evident, encompassing various terms such as human-robot interaction, human-robot teams, and collective intelligence. However, there’s a lack of consensus on defining these interactions. This study delineates three defining characteristics: sequential shared actions toward a common goal, AI’s ability to share and adapt to information, and enhanced performance through collaboration. Collaborative intelligence entails complementarity, a shared objective, and sustained interaction, going beyond mere division of labor.
The motivation for collaborative intelligence stems from its potential to enhance task performance and work satisfaction. While AI excels in processing data and recognizing patterns, it struggles with common-sense situations and intuitive decision-making, areas where humans excel. Collaborative intelligence optimizes performance by combining human intuition with AI’s computational power.
The potential of collaborative intelligence is exemplified in endeavors like human-computer chess teams, where combining human creativity with computational power yields superior results. Additionally, collaborative intelligence enhances the quality of work by automating mundane tasks and allowing humans to focus on more rewarding endeavors, thus improving work satisfaction.
Exploring the sociotechnical dynamics of collaborative AI presents opportunities to enhance worker satisfaction and bridge skills gaps within the workforce. While optimal design remains under investigation, current research underscores the impact of AI on worker experience, including predictability, controllability, meaningfulness, and fairness. Collaborative intelligence offers a promising avenue for leveraging human potential alongside AI capabilities, thereby enhancing both task performance and work satisfaction.
Our research aims to assess how combining human and artificial intelligence can significantly enhance performance. We established criteria for evaluating “collaborative intelligence” in AI systems and conducted a systematic review of AI applications to determine their embodiment of these criteria. Our findings demonstrate that collaborative efforts between humans and AI yield improved outcomes across various measures, including creativity, safety, and productivity.
Our review uncovered 16 instances of collaborative intelligence, where humans and AI work together to achieve more than either could alone. These examples are largely in early development stages. They span various fields such as healthcare, manufacturing, and creative writing, using diverse communication methods including sensors and natural language processing. Such collaborations enhance efficiency, creativity, and safety. While some argue AI should be seen as a tool rather than a collaborator, we emphasize the shared output aspect of collaboration. It’s crucial to distinguish between AI tools and collaborators, even if the latter lack awareness or independent initiative. This nuanced terminology can shape the socio-technical implications of AI integration.
Since our systematic review, Large Language Models (LLMs) like ChatGPT and DALL-E have rapidly gained widespread adoption. These AI tools align with collaborative intelligence criteria by enhancing human work through their computational power and ability to integrate vast data. Research shows improved productivity, quality, and customer satisfaction. LLMs support humans across various workflow stages, offering naturalistic conversation and adapting responses based on user interactions. However, due to the diversity of LLM models, they can’t be classified as a single application, warranting further research. Their popularity highlights the value of AI systems that complement human abilities and facilitate communication for shared objectives and improved outputs.
The review found that current collaborative intelligence applications have limitations in operating across various environments and performing diverse tasks compared to human collaborators who seamlessly switch between real-world and virtual collaboration. While these applications can work in virtual or cyber-physical forms, they lack the flexibility to collaborate effectively across both environments. They are also constrained to discrete tasks rather than engaging in multi-stage projects like humans. The development of future collaborative intelligence applications faces the challenge of enabling them to transfer capabilities across different domains. Additionally, numerous applications partially meet collaborative intelligence criteria, hinting at future growth in finance, defense, and scientific research driven by advancements in large language models (LLMs). The rising adoption of LLMs is increasing awareness of AI benefits and fostering collaborative work with AI, emphasizing the need to address ethical considerations in implementing and designing collaborative AI applications.
The current prevalence of AI applications automating tasks is due to the complexity required for collaborative AI, including understanding human perspectives and engaging in dialogue. Recent advancements are enhancing these capabilities, indicating potential for a new era of collaborative intelligence that boosts efficiency, safety, sustainability, and overall quality of work and life.
Source:
Emma Schleiger, Claire Mason, Claire Naughtin, Andrew Reeson & Cecile Paris (2024) Collaborative Intelligence: A Scoping Review Of Current Applications, Applied Artificial Intelligence, 38:1, DOI: 10.1080/08839514.2024.2327890