Collaborating with Machines: Challenges at the convergence of human and artificial intelligence

I have completed a research paper after ten months into my EU funded project Augmented Humans and I’d like to share some ground truth given the growing hype and controversies that are emerging in artificial intelligence.

While this is all taking place among technologists, computer scientists and ethicists, it’s becoming obvious that organisations find themselves a challenging situation of limited practical applications, conflicting information and uneven results across industry. My research paper is currently under peer review with a journal but here are a few headlines that might be useful to organisations looking to implement augmented intelligence in their organisations. I've only left some key references in the text for anyone that wishes to delve deeper.

#artificialntelligence #augmentation #convergence #collaboration #teamwork #humancentereddesign #humanmachineinterface

Context

Contemporary business decision-making is characterised by experimentation, collaboration, and creative problem solving; skills that are critical in an environment characterised by increasing complexity, change and ambiguity. However, these skills at this point in time appear to be elusive to AI due to fundamental difficulties in replicating and scaling human knowledge representation (Dellermann, Calma, et al., 2019; Duan et al., 2019; Dwivedi et al., 2021; Goertzel, 2014; R. M. Lee, 1985) which gives rise to the ‘augmentation design problem’ when attempting to combine the unique capabilities of both humans and machines. This becomes a bigger challenge when designing augmented systems to enhance collaboration and problem solving in complex and ambiguous environments.

Key Motivation for our research

Industry uptake of augmented intelligence has been limited by systemic challenges with algorithmic models, AI system understandability, explainability and usability, limitations in machine learning methods and training data, a lack of critical and transparent research on AI capability, and pressing ethical issues (Bender et al., 2021; Birhane, Prabhu, et al., 2021; Holmstrom, 2022). Industry motivation to address these challenges are very high due to increasing concerns about low value creation and declining business confidence in AI investments. Our core research question therefore is: what are they key enablers and inhibitors in the development, implementation, and user acceptance of augmented intelligence?

Prior attempts at addressing this research problem

The field has been constrained by a fragmentation of research across multiple scholarly disciplines and technology silos, and a concentration of proprietary research in large technology corporations. Historically, there is a relative underrepresentation of the unique challenges of human-machine augmentation for cognitive work in complex environments. Most importantly there has been a lack of critical research on the value creation potential of AI in business applications (Coombs et al., 2020; Cubric, 2020; Keding & Meissner 2021; Collins et al., 2021).

There is also limited research on augmentation relative to the more popular field of AI automation which is fundamentally different in its motivation, technologies and value-creation goals. The automation and computer science bias that dominates the discourse on AI technology development and application presents a major barrier to more holistic multidisciplinary approaches required in the more complex domain of human-machine augmentation.

Our research method

The methodology for this paper was based on a literature review which consisted of systematic searches in the SCOPUS database focusing on high quality peer-reviewed journals in key domains that included AI and augmentation in information systems, management research, and organisation or systems research. The process identified 961 papers, resulting in a total of 158 papers that met the inclusion criteria for our final review. An in-depth thematic analysis was undertaken over several passes to identify key findings, themes and research directions. (The next phases of our project will include a pilot survey, expert interviews and a psychometric study).

Key results

A total of 75 key themes emerged from the literature review data, which were distilled into 10 key categories then organised into four overarching thematic clusters coded as human engagement, organisational evolution, strategic positioning and technology development, as depicted in Figure 1:

Figure 1 Human-AI augmentation challenges: Four clusters and ten categories

We found limited empirical evidence to support expositions of the decision-making effectiveness and performance outcomes of AI augmentation, and at best the literature showed contradictory and mixed results (Borges et al., 2021; Collins et al., 2021; Rzepka & Berger, 2018). Furthermore, performance results were often reported in an overly optimistic way (Cubric, 2020; Enholm et al., 2021; Newlands, 2021) and research results lack generalizability across application domains (Langer & Landers, 2021; Niehaus & Wiesche, 2021; Cubric 2020; Morley et al., 2021). A total of 28 future research directions were identified.

A recurring theme of machine learning limitations

General problems in the machine learning domain was a recurring theme in our review. There are consistent reports of limitations in systematic, structured methods and processes to develop, deploy and evolve machine learning models which has resulted in AI often exhibiting poor behavior when deployed in real-world applications. Critical problems and ethical issues emerge when inductive biases become encoded into the training models which create blind spots, spurious correlations and unreliable predictions (Marcus, 2022; Birhane et al., 2021; Newlands, 2021; Bender et al., 2021). These issues are limiting trust, usability and technology acceptance in human-machine collaborations.

AI ethicists believe that AI holds significant potential to improve many aspects of human life and business value creation. However, they also caution that AI can pose major systemic threats and harms such as bias, discrimination and safety without effective ethical frameworks in place (Floridi et al., 2018; Hagendorff, 2020; Morley et al., 2021; Taddeo & Floridi, 2018).

Conclusion

At this point in time AI has limited capability in dynamic, fluid, complex or dynamic environments where humans flourish and are at their most creative in problem solving (Dellermann, Calma, et al., 2019; Jarrahi, 2018; R. M. Lee, 1985; Rafner et al., 2021; Teresa et al., 2020). Within this technological limitation however lies the greatest opportunity for augmenting human and artificial intelligence.

The potential benefits of AI technologies are undeniable however the true value-creation opportunities may be in the human, organisational and social ecosystems that supports and utilises the technology rather than in the technology itself.?

Practitioner takeaways: What can organisations start doing now?

Our research is still ongoing but practical tips can be taken from our four key interdependent clusters critical to the development of augmented human-machine systems:

Human Engagement: building understanding of the psychological, sociological, neurological and biological factors that affect how humans engage, interact and collaborate with AI. The fundamental aim of augmented technologies is build and enhance human capability.

Organisational Evolution: creating enabling organisation systems, structures, processes and networks that are conducive to effective human-AI technology acceptance, adoption and collaborative teamwork. This is about creating human-machine-nature ecosystems rather than organisational hierarchies of power and control. Augmented systems in complex and ambiguous environments need distributed autonomy and participation.

Technology Development: developing technological capability that proactively manages limitations and enhances the potential of AI technology design, data models, algorithms, interactions and interfaces that facilitate human-machine augmentation and collaboration. This includes an understanding of next generation cognitive technologies, using ethics frameworks to design human-centred systems, and using participatory practices in technology design, development and implementation involving a wide variety of stakeholders.

Strategic Positioning: Organisations must lead with strategy, not the technology. This is because technology is constantly changing and expanding, and its design, development and utilization needs to be grounded in the strategic positioning of the organisation and its value creation objectives. This requires understanding the complexities and realities of deriving value creation from AI, achieving clarity in the source of truth from critical research and continuous performance assessment. ?

Hope this has been useful to you, we are only at the beginning of this journey.

A request from us

When our paper will be formally published it will be open access so we will freely share it with the public. Our research is ongoing and we would very much appreciate if you could take the time to complete our survey for our pilot project on your experience with AI applications. This is not a technical survey it is focuses on the user experience and will help is plan the next stages of our research. The link can be found here?https://forms.office.com/r/TvNQWqaB5g