Artificial Intelligence & Machine Learning

Wednesday 14:00- 15:00



“Human-Machine Teaming Agents: A Future Perspective”

Michael Teichmann, Marco Ragni, Julien Vitay, Martin Gaedke and Fred H. Hamker

Abstract— The rise of artificial agents that excel in complex tasks makes it feasible for humans and machines to form efficient teams. However, the design and requirements of the required software agent are still unclear. We identify important factors for human-machine teaming and characterize teaming situations by the presence of shared (sub-)goals, communication, interdependence, and the ability to learn and adapt. Given this, we introduce a framework for software teaming agents, which utilizes state-of-the-art deep neural networks and addresses how communication and conceptual information can be incorporated into such a design. Moreover, we suggest information-seeking behavior, based on uncertainty assessment, to deal with the variability of the environment and the agent’s imperfectness. Finally, we address some interdisciplinary research directions in human-machine teaming which arise from our conception.

Keywords: human-machine teaming, deep neural networks, multi-task learning, multi-channel networks, information-seeking, abstraction



“Gyro Gearloose or Little Helper? Two Perspectives Of AI And Patent Law”

Dagmar Gesmann-Nuissl and Stefanie Meyer

Abstract— Recently, patent offices around the world were concerned with the patent application for a blinking light and a fractal food container, and extensively discussed the requirements for granting a patent. This unprecedented case sparked extensive discussions as “Dabus AI” was filed as an inventor – an autonomously acting software. This case raised a number of issues concerning patent law, in particular whether an invention in the field of artificial intelligence (AI) is patentable, and moreover, whether the invention made by an AI is patentable. In this paper, we address these two perspectives by discussing the legal requirements for filing a patent application and applying the two perspectives described to these requirements. We conclude that, as of now, only the human inventor Gyro Gearloose can be a patentable inventor. The time when his little helper will occupy a similar position is still in the future – but it is worth looking beyond the horizon into the future.

Keywords: artificial intelligence, patent law, inventor, AI inventions, AI-generated inventions



“Nadaraya-Watson Time Series Classification For Gesture Recognition”

Florian Kretzschmar, Rim Barioul, Dana Uhlig, Olfa Kanoun, and Alois Pichler

Abstract— The prediction of gestures in shared public spaces is important not only to ensure good functionality of agents but also the safety of humans and their acceptance of the hybrid society concept. In this context, surface electromyography (sEMG) combined with inertial measurement unit (IMU) signals is investigated in this work to predict dynamic hand gestures. Our proposed algorithm is designed to make predictions while these signals are being observed, combining a Nadaraya-Watson kernel estimator and an entropy-based decision function. We not only achieve a high accuracy, but also a significant earliness of prediction.


Keywords: Gesture recognition, Time series classification, Surface electromyography



“Person Detection and Differentiation in Shopping Scenarios”

Divyasha Naik, Martin Reber, Tom Uhlmann and Guido Brunnett

Abstract— In the research field of human-robot-interaction, detection of people is of central importance. Moving robotic platforms create additional challenges due to an absence of a defined area of interaction between humans and robots. The use of a mobile robotic platform to provide personalized assistance requires that tracking and identification of the interaction partner in a group of people can be ensured at all times. This work contributes to an application that can detect and distinguish multiple people in public spaces. We use Mask R-CNN followed by DeepSORT for the differentiation and tracking of each individual in a video by assigning a unique ID. We apply our approach to the Multi-Object Tracking and Segmentation (MOTS20) data set and show that our method successfully performs person detection, differentiation, and tracking at 5 frames per second. Using the three evaluation metrics Multiple Object Tracking (MOTA), the Higher Object Tracking (HOTA) and the Identification metric (IDF1), we achieved an accuracy of 49.69%, 48.35% and 56.14%, respectively. Our results show good accuracy, but also possibilities for future improvements. We show how reflections, shadows, human-like clothing, and poor annotations influence the precision negatively.

Keywords: Person detection, Person tracking, Mask R-CNN, DeepSORT



Friday 8:30-9:30



“Observational Learning in Humans and Machines”

Claudia Voelcker-Rehage, Fred H. Hamker, Javier Baladron, Julian Rudisch, Torsten Fietzek and Julien Vitay

Abstract— Observational learning is referred to as a change in performance following the observation of others. With respect to motor learning, an observed action is known to facilitate motor learning mediated by brain processes that are involved during both, the observation and the execution of a certain task. Observational learning in humans has inspired robotic researchers, as it may alleviate the necessity to explicitly program robots or require robots to extensively search for a suitable solution. Further, observational learning may become a central aspect in future hybrid societies in which robots closely interact with humans. Here we summarize the current state of the art on observational motor learning in humans and robots, with a focus on upper extremity tasks. Further, we briefly outline a roadmap for better understanding observational motor learning by means of building brain-inspired neuronal models. Bringing together these lines of research not only advances human movement science but, in the long run, may contribute to new programming approaches in robots that facilitate human-robot interactions.

Keywords: motor learning, mirror neurons, robots, machine learning, computational neuroscience



“>AI Takeover… Doesn’t Sound that Bad!< – Authoritarian Ambivalence Towards Artificial Intelligence”

Frank Asbrock, Jochen Mayerl, Manuel Holz, Henrik Andersen and Britta Maskow

Abstract— Public perceptions of Artificial intelligence (AI) are mostly positive, but recent research indicates growing skepticism and differentiated attitudes towards different AI applications. Moreover, research showed that more conservative people are more skeptical. Extending previous research, we tested the role of authoritarianism, a broad ideological attitude, in relation to attitudes towards AI in different domains in a German online sample (N = 1,027). Structural equation models showed the expected ambivalent relationship between authoritarianism and attitudes towards technology (expressed in terms of positive and negative attitudes) as well as differentiated relationships with specific AI attitudes. Authoritarianism showed a small positive total effect on attitudes towards AI in autonomous vehicles and a non-significant total effect on AI in robots (mediated by attitudes towards technology and towards AI in general), but a strong relationship with attitudes towards AI in surveillance, which is in line with the general authoritarian preference for security and social order.

Keywords:  authoritarianism, artificial intelligence, attitudes, structural equation modelling



“Feature Selection for Hand Gesture Recognition using Six FSR Sensors Bracelet”

Sajidah Al-Hammouri, Rim Barioul, Khaldon Lweesy, Mohammed Ibbini and Olfa Kanoun

Abstract—  Gesture recognition is crucial for numerous applications, from gaming controllers to remote manipulation to assistive technology for the disabled. Force myography (FMG) has recently been used in gesture recognition applications and has effective recognition capabilities. In this study, eleven gestures from the American sign language (ASL) were classified, and in order to improve the classification accuracy, three outstanding swarm optimization algorithms, which are the binary grey wolf optimizer (BGWO), binary grasshopper optimizer (BGOA), and binary hybrid grey wolf particle swarm optimizer (BGWOPSO) were investigated to test which one achieves better performance when exploited as a feature selection wrapper based on ELM. The three wrappers are evaluated in terms of accuracy. The data was collected from 10 volunteers using a bracelet with only six force-sensitive resistors (FSR) sensors. The results demonstrate that using the BGWOPSO as a wrapper based on the extreme learning machine (ELM) outperforms the other methods and can effectively enhance its performance, by increasing the classification accuracy from 55.82% to 91.36%.

Keywords: Extreme learning machine, Force myography, FSRs, Hybrid Grey Wolf Particle Swarm



“Proposals for Communicative and Cooperative AI to Promote Synergies in Hybrid AI-Augmented Socio-Technical Arrangements – A Humanities Perspective”

Arne Sonar and Christian Herzog

Abstract— We propose to explicitly adopt the requirements of cooperation and communication as constitutive features that can govern the design of hybrid artificial intelligence (AI)-augmented socio-technical arrangements. We believe that truly harnessing the synergies within concrete human-machine interaction scenarios calls for a fundamental reevaluation of the normative ethical groundwork based on which the hybridization of actions, processes and structures is advanced. For this purpose, we conduct a qualitative-explorative literature analysis as well as review and evaluate corresponding design implications of the constituents of cooperation and communication. We use the AI-augmented diagnosis of deep-vein-thromboses at the point-of-care as an illustrative example.

Keywords: cooperative AI, communicative AI, hybridity, socio-technical arrangements