AIXPERIMENTATIONLAB — Augmented Intelligence Systems: Opportunities and Challenges
Towards an institutionalized format for the design, development, use and diffusion of human-centred artificial intelligence applications
Shedding the light on augmented intelligence systems
Automation and data-based services are becoming increasingly prevalent in our everyday lives. The growing amount of data generated by connected devices and services has led to new ways of interacting with artificial intelligence systems. These interactions take place across a variety of industries and in various forms such as autopilot systems, automated medical diagnostics, financial services, digital productivity applications and interactive platforms for manufacturing and education. The steady development of artificial intelligence technologies motivated the emergence of a more inclusive paradigm, commonly known as augmented intelligence. It is based on the assumption that human and machine intelligence complement each other positively. This paradigm involves information technology tools built and designed around human collaboration to support and improve human cognition through evidence-informed decision-making.
The relevance of designing and developing augmenting systems for complex tasks has gained momentum as a requirement for their widespread adoption. This article intends to give a brief context-independent overview on current opportunities and challenges of implementing augmented intelligence approaches to achieve an efficient combination for decision-making.
If you missed the article on the motivation behind augmented intelligence, the Augmented Intelligence Framework developed at WZL of RWTH Aachen and its use within the research project “AIXPERIMENTATIONLAB”, you can check it here.
Opportunities and challenges
Augmented intelligence aims at synergistically complement human and machine intelligence capabilities. While artificial intelligence algorithms are able to analyze and discover hidden patterns in large amounts of data, humans can add context-sensitive information to enhance decision-making. Algorithms find solutions faster and more efficiently, but often only with limited certainty on the feasibility of the solution. To minimize this remaining uncertainty, the human counterpart provides its expertise or uses its intuition, spontaneity, and heuristics. Accordingly, the final decision-making and the resulting actions remain entirely under human control. The human in turn “feeds” the artificial intelligence system with his expertise and knowledge gained from experience in order to improve the algorithms for the next case. The principle of “human-in-the-loop” helps to combine the relative advantages of human and machine intelligence. As a result, not only does immediate decision-making becomes more efficient and effective, but also mutual learning takes place from the interaction between humans and machines (Figure 1).
While many industries have benefited from the use of advanced automation and networking technologies, human cognitive capability remains a critical success factor at the operational level. This holds especially true for complex, high-tech, and highly customized industries, where humans play an important role in understanding and solving problems, as well as driving continuous improvement beyond the automation of repetitive tasks. In this setting, augmented intelligence approaches are especially beneficial when humans can make better initial judgements than automated systems through flexible and creative problem solving. This is the case in settings where data availability for automated system training is restricted, or the tasks be performed are not well defined. In general, augmented intelligence approaches could perform well where hard-wired intelligent systems do not understand the input, the input is not correct, or the models behind an application are not accurate enough to outperform a human.
Beyond the direct performance comparison of augmented and fully automated systems, augmented intelligence approaches enable more advanced forms of collaboration and concepts of labour division. Such is the case for collaborative robots in production contexts, which share working spaces with humans in a safer and more efficient way than classical cage robots. Moreover, humans can achieve higher levels of accuracy and safety for critical operations when augmented by intelligent systems. While intelligent systems can support with inspection activities, humans can directly monitor and check for plausibility during quality-critical operations. Finally, intelligent technologies such as wearable and monitoring devices can support in mitigating effects of overburden by providing context-sensitive support when required.
Despite the potential advantages of these augmenting technologies, several challenges may arise when implemented in real-world environments. Although intelligent solutions are intended to assist humans in carrying out operations, there is a latent risk of over- or mistrusting these systems. Hence, solutions that are not adjusted to task requirements and context-specific performance objectives are prone to underperform in the long term. Moreover, some advanced technologies may fall short on expectations when implemented in real-life conditions. For instance, wearables providing task-relevant information can convey an advantage as soon as they are comfortable, ergonomic, and achieve an efficiency improvement over the result of a non-assisted operation. In this regard, not every person may be accustomed to wear some kind of device or receive automatically generated instructions when performing their tasks. Furthermore, some solutions may be hindered either by technical problems (e.g. due to accuracy and synchronization of delivered information, transmission latency or interruptions) or organizational issues, such as certification procedures and roll-out permits. Additionally, the perception of these systems changes from person to person is an important factor to consider during implementation. Thus, the adaptation to these technologies may require an extended period of time, as well as a potential paradigm shift for the existing and future workforce.
On a more technical perspective, these systems require a considerable amount of resources to achieve desired outcomes. Besides the direct monetary costs, further resources involve the data required for training the solutions, as well as the timely constraints on the operative schedule of focus activities. Besides, some systems may require sensitive or protected information in order to be implemented successfully in the expected conditions. The latter requires special attention, as data privacy concerns might represent a major obstacle in the implementation of intelligent data-based solutions. This challenge addresses an important trade-off between the amount of data required to make a solution more effective, and the degree to which data can be disclosed in favour of enhanced work efficiency. Therefore, a broad acceptance of these systems is required to mitigate possible risks that could be detrimental for their roll-out in the first place.
Conclusion and outlook
With ever-increasing digitalization, new opportunities arise to improve human activities within sociotechnical systems. The present article introduced some of the opportunities and challenges of implementing augmented Intelligence solutions in context-independent operations. While these approaches promise several potential benefits, many practical challenges exist for their wider implementation. For this reason, two important aspects need to be considered when developing augmented intelligence systems: the position of humans within the intelligent system and the potential gains accomplished compared to non-augmented operations. In this sense, a holistic view of the problem is required, starting from a cost-benefit-analysis, through the technical and organizational requirements, to the design of user interfaces and data acquisition methods. As pointed out, trust and privacy may represent an important hurdle for wider acceptance. In order to overcome mistrust, a high share of judgement should be given to humans. On the other hand, the purpose of building augmenting systems should be clear to the user, with involvement of preferences at an early stage of development. Finally, human engagement during interactions is a crucial aspect to be considered when integrating intelligent solutions in our everyday lives. On this basis, the sustainable design of intelligent solutions should focus on both the experience meaningfulness of the interactions and the system outcome alike.
This article was generated as an informative contribution to the central technical research subject within the project “AIXPERIMENTATIONLAB” funded by German Federal Ministry of Labour and Social Affairs (BMAS) — EXP.01.00016.20.
The authors gratefully acknowledge the support of the German Federal Ministry of Labour and Social Affairs (BMAS).
For more information please visit the project’s official website (not optimized for mobile, in German only).
References
Akash K.; McMahon, G.; Reid, T.; Jain.: Human Trust-Based Feedback Control: Dynamically varying automation transparency to optimize human-machine interactions. In: IEEE Control Systems, 2020, pp. 98–116.
Bisen, V. S.: Medium: What is Human in the Loop Machine Learning: Why and How used in AI? URL: https://medium.com/vsinghbisen/what-is-human-in-the-loop-machine-learning-why-how-used-in-ai-60c7b44eb2c0 [Accessed: 10.01.2022]
Engelbart, D.: Augmenting Human Intellect: A Conceptual Framework. Summary Report. Stanford Research Institute: Palo Alto, CA, USA, 1962.
Emmanouilidis, Ch.; Pistofidis, P.; Bertoncelj, L.; Katsouros, V.; Fournaris, A.; Koulamas, Ch.; Ruiz-Carcel. C.: Enabling the human in the loop: Linked data and knowledge in industrial cyber-physical systems. In: Annual Reviews in Control. Vol. 47, 2019, pp. 249–265.
Human-Centered Artificial Intelligence (HAI), Stanford University; Wang, G.: Humans in the Loop: The Design of Interactive AI Systems. URL: https://hai.stanford.edu/blog/humans-loop-design-interactive-ai-systems [Accessed: 10.01.2022].
Hassani, H.; Silva, E. S.; Unger, S.; TajMazinani, M.; Mac Feely, S.: Artificial Intelligence (AI) or Intelligence Augmentation (IA): What Is the Future? In: Artificial Intelligence. Vol. 1, 2020, pp. 143–155.
Levity; Wolfewicz, A.: Human-in-the-loop in machine learning: What is it and how does it work? URL: https://www.levity.ai/blog/human-in-the-loop [Accessed: 31.01.2021].
Oliveira, M.; Arica, E.; Pinzone, M.; Fantini, P.; Taisch, M.: Human-Centered Manufacturing Challenges Affecting European Industry 4.0 Enabling Technologies. In: Constantine Stephanidis (Eds.), HCI International 2019 — Late Breaking Papers, 21st HCI International Conference. HCII 2019 Orlando, Florida, USA, 26.-31. July 2019, pp. 507–517.
Rueckert, P.; Muenkewarf, S.; Tracht, K.: Human-in-the-loop simulation for virtual commissioning of human-robot-collaboration. In: Procedia CIRP. Vol. 88, 2020, pp. 229–233.
Zanzotto, F. M.: Viewpoint: Human-in-the-loop Artificial Intelligence. In: Journal of Artificial Intelligence Research. Vol. 64, 2019, pp. 243–252.
