Source-linked AI summary
Hybrid Intelligence
Dominik Dellermann, Philipp Ebel, Matthias Soellner, Jan Marco Leimeister
TL;DR
The paper addresses the gap between AI’s performance on defined tasks and its limited use for complex organizational problems. It develops Hybrid Intelligence as a framework combining human and artificial intelligence through deliberate task allocation and mutual learning. The paper argues that this combination can achieve superior results, while trust, interpretability, and transparency remain important challenges.
Problem
AI performs well on defined tasks, but its use for complex organizational problems remains scarce and largely confined to laboratory settings.
Method
The paper defines Hybrid Intelligence as combining human and artificial intelligence through deliberate task allocation, mutual learning, and complementary capabilities.
Results
Hybrid Intelligence is defined as achieving complex goals and superior results through combined human and artificial intelligence that continuously learn from each other.
Takeaways & Limitations
Hybrid Intelligence can extend human-machine collaboration to complex tasks beyond games, including strategic decision making, science, and AI development.
Takeaways & Limitations
Effective Hybrid Intelligence requires balancing trust and distrust while making system states and needs understandable to humans and machines.
Abstract
from arXiv · showhide
Research has a long history of discussing what is superior in predicting certain outcomes: statistical methods or the human brain. This debate has repeatedly been sparked off by the remarkable technological advances in the field of artificial intelligence (AI), such as solving tasks like object and speech recognition, achieving significant improvements in accuracy through deep-learning algorithms (Goodfellow et al. 2016), or combining various methods of computational intelligence, such as fuzzy logic, genetic algorithms, and case-based reasoning (Medsker 2012). One of the implicit promises that underlie these advancements is that machines will 1 day be capable of performing complex tasks or may even supersede humans in performing these tasks. This triggers new heated debates of when machines will ultimately replace humans (McAfee and Brynjolfsson 2017). While previous research has proved that AI performs well in some clearly defined tasks such as playing chess, playing Go or identifying objects on images, it is doubted that the development of an artificial general intelligence (AGI) which is able to solve multiple tasks at the same time can be achieved in the near future (e.g., Russell and Norvig 2016). Moreover, the use of AI to solve complex business problems in organizational contexts occurs scarcely, and applications for AI that solve complex problems remain mainly in laboratory settings instead of being implemented in practice. Since the road to AGI is still a long one, we argue that the most likely paradigm for the division of labor between humans and machines in the next decades is Hybrid Intelligence. This concept aims at using the complementary strengths of human intelligence and AI, so that they can perform better than each of the two could separately (e.g., Kamar 2016).
Hybrid Intelligence
The paper addresses Hybrid Intelligence, artificial intelligence, and human-computer collaboration as central themes for understanding future work.
- The authors are Dominik Dellermann, Philipp Ebel, Matthias Söllner, and Jan Marco Leimeister.
- The article was published online on 28 March 2019 by Springer Fachmedien Wiesbaden GmbH.
- The paper examines Hybrid Intelligence and related forms of intelligence.
1 Introduction
The introduction contrasts AI’s success on clearly defined tasks with its limited deployment for complex organizational problems. It proposes Hybrid Intelligence as a likely future division of labor between humans and machines.
- AI performs well on clearly defined tasks such as chess, Go, and image-based object identification.
- Artificial general intelligence capable of solving multiple tasks simultaneously is not expected in the near future.
- AI applications for complex business problems remain scarce in organizational contexts and mainly confined to laboratory settings.
- The paper argues that Hybrid Intelligence is the likely near-term paradigm for dividing labor between humans and machines.
2 Conceptual Foundations and What Hybrid Intelligence is Not
The paper distinguishes Hybrid Intelligence from general, human, artificial, and collective intelligence. Its defining contrast is the combination of heterogeneous human and machine agents for complex goals.
- 2.1 Intelligence: General intelligence is the ability to accomplish complex goals, learn, reason, and adaptively act within an environment.
- 2.2 Human Intelligence: Human intelligence concerns people’s mental capacity to learn, reason, and adaptively act using existing knowledge.
- 2.2 Human Intelligence: Human intelligence has been described through general, experience-based, specialized, and componential, contextual, and experiential perspectives.
- 2.3 Collective Intelligence: Collective intelligence describes groups acting collectively in apparently intelligent ways, commonly involving combined intelligence among human agents.
- 2.4 Artificial Intelligence: Artificial intelligence refers to systems performing activities associated with human thinking, including decision-making, problem solving, and learning.
3 The Complementary Benefits of Human and Artificial Intelligence
Humans and machines offer complementary capabilities that can augment one another. The paper describes decision-support and human-in-the-loop arrangements as ways to combine these strengths.
- 3 The Complementary Benefits of Human and Artificial Intelligence: Hybrid Intelligence combines complementary human and computer capabilities to augment one another.
- 3 The Complementary Benefits of Human and Artificial Intelligence: Humans are described as stronger in settings requiring system 1 thinking, including flexible, creative, and empathic capabilities.
- 3.1 Artificial Intelligence in the Loop of Human Intelligence: AI can improve human decisions by providing predictions, while humans can train machine-learning models through human-in-the-loop arrangements.
- 3.1 Artificial Intelligence in the Loop of Human Intelligence: In business, AI is used both to automate machine-solvable tasks and to provide humans with decision support through predictions.
- 3.1 Artificial Intelligence in the Loop of Human Intelligence: Physicians can use AI-processed patient data for disease predictions while applying their own intuition and empathy in decisions.
- 3.2 Human Intelligence in the Loop of Artificial Intelligence: Humans support AI by generating algorithms, tuning hyperparameters, training or debugging models, and interpreting unsupervised outputs.
4 Defining Hybrid Intelligence
Hybrid Intelligence combines human and artificial intelligence to solve complex goals through deliberate task allocation, superior system-level outcomes, and continuous mutual learning. Its potential is illustrated by AlphaGo and extends to complex decision-making and knowledge transfer.
- Definition: Hybrid Intelligence combines human and artificial agents to achieve complex goals and superior results through deliberate task allocation.The system can outperform either humans or machines acting separately.
- Core concepts: Tasks are performed collectively, with conditionally dependent activities among agents whose goals may not always be aligned.The shared system goal can coexist with locally different objectives, such as teaching adversarial tactics.
- Core concepts: Hybrid Intelligence systems continuously improve as human and artificial agents learn from each other through experience.Performance assessment includes both the system’s superior outcome and performance increases in its individual agents.
- Illustrative example: AlphaGo learned from expert human moves, achieved superhuman performance, and introduced creative moves that expanded human players’ knowledge.The example illustrates mutual augmentation: human input improved the AI, while the AI influenced expert play.
- Implications: Hybrid Intelligence can allocate complex tasks between humans and machines while supporting knowledge transfer, interpretable learning, customization, trust, and acceptance.Proposed applications include strategic, managerial, political, military, scientific, and AI-development decisions.
6 Future Research Directions for the BISE Community
Future BISE research should focus on socio-technical system design for complex, uncertain business problems and on overcoming barriers to trustworthy AI adoption. Such settings require both analytic and intuitive human abilities alongside machine learning.
- Research motivation: Complex managerial problems are dynamic, time variant, domain-dependent, and lacking a specific ground truth, requiring both analytic and intuitive abilities.Human creativity and empathy are also relevant in these uncertain contexts.
- Research agenda: The paper proposes three interrelated directions for developing Hybrid Intelligence in BISE, focused on socio-technical system design.The supplied passage introduces the directions without enumerating all three.
- Trust and interaction: Trustworthy AI adoption requires balancing trust and distrust while translating system states and needs between humans and artificial agents.Semi-autonomous driving is given as an example requiring task distribution based on the human’s state.
- Trust and interaction: Domain-specific interface guidelines that help humans understand and process artificial-system needs are still missing.The paper therefore calls for further research on human-centered architectures balancing model transparency and performance.
5 The Advantages of Hybrid Intelligence
Hybrid Intelligence can combine human and AI capabilities to generate knowledge, support mutual learning, and enable new forms of work. Its development also raises challenges around trust, governance, incentives, and education.
- Hybrid Intelligence can generate new knowledge in complex domains, allowing humans to learn from AI.
- Human helpers can teach AI systems through task and interface designs that support human instruction.
- Maintaining machine-learning accuracy while ensuring interpretability and transparency is crucial for building appropriate trust in AI.
- Hybrid Intelligence research must examine governance mechanisms for matching experts to tasks, aggregating input, and assuring quality standards.
- AI-driven changes in task distributions create new qualification demands and motivate research on education for democratizing AI use in future workspaces.
- Internal crowd work can leverage collective expert knowledge across functional silos within a company.