Refine
Document Type
- Article (2)
- Doctoral Thesis (1)
Language
- English (3)
Has Fulltext
- yes (3)
Is part of the Bibliography
- no (3)
Keywords
- Advice Accuracy (1)
- Algorithm appreciation (1)
- Forecasting (1)
- Perfect Automation Scheme (1)
- algorithmic advice (1)
- artifical intelligence (1)
- cognitive trust (1)
- confidence intervals (1)
- forecast (1)
- initial trust (1)
Institute
With recent progresses in the field of artificial intelligence (AI), algorithms can support more complex tasks (Frey and Osborne, 2017) that seemed safe from automation a few years ago (Brynjolfsson and McAfee, 2012). They can even outperform human decision-making in many task domains (Meehl, 1954; Dawes et al., 1989). The prominent defeat of Garry Kasparov, the world champion in chess, in 1997 is a milestone for the potentials of algorithms, increasing the value of IBM’s stock increased by $18 billion (Norvig and Russell, 2010). More recently, universal chatbots such as ChatGPT and the Bing Chatbot Sydney can influence millions of users and provide them answers to a broad range of tasks.
To gain a competitive edge, companies invest heavily in algorithmic systems (Kappelman et al., 2021), which are considered as one of the most important drivers of today’s economy (Brynjolfsson and McAfee, 2011; Hogharth and Makridakis, 1981; Önkal et al., 2019). Particularly in management accounting, an algorithmic data analysis offers the potential to support a high decision quality, making their usage a key factor for success.
However, algorithms can not only support, but also manipulate human decision-making (Roose et al., 2023). Algorithmic advice can lead to overtrust, i.e. to the heuristic replacement of vigilant information seeking and processing in favor of the advice provided (Mosier et al., 2001). The implementation of algorithmic decision support systems is thus no panacea to improve human decision-making. In contrast, they can entail new risks, such as the uncritical use of these algorithms (Boden, 2016). Yet, this ‘dark side’ of algorithms is not sufficiently investigated in the field of management accounting. With advances in machine learning, algorithmic support can become even more of a black box, blurring the reference points for evaluating its advice quality and thus increasing the risk of overtrust. Stephen Hawking even warns that AI will be “either the best, or the worst thing, ever to happen to humanity” (Alex Hern in: The Guardian: Stephen Hawking: AI will be 'either best or worst thing' for humanity, 19th October 2016). Interacting with such algorithmic systems therefore requires data literacy skills for an appropriate use. They seem to be a crucial prerequisite in today’s data-driven world to reap of the benefits of algorithmic systems. If no reflective and critical interaction between human and algorithm can be ensured, the trend towards a digitalized world will be a race against the algorithm (Brynjolfsson and McAfee, 2011).
This dissertation takes this understudied critical perspective on the increasing implementation of algorithmic systems and sheds light on different aspects of human-algorithm interaction, ranging from the ability to use data visualization tools to trust calibrations and critical evaluations of algorithmic advices. Study I investigates an adequate information representation of subjective multiple criteria decision problems, an important task in management accounting, to support human decision-making. Study II distinguishes different designs of the human-algorithm interaction investigated in the trust in automation literature and compiles factors influencing the calibration of trust and behavior toward the true capabilities of the algorithm. Study III examines the role of performance feedback on trust and advice usage in a forecasting task for a better understanding of the recently observed phenomenon ‘algorithm aversion’. Finally, Study IV focusses on the risk of overtrust in advice in repeated interactions with a forecasting advisor and investigates the calibration of advice usage. Each study thereby represents an independent contribution and contains all information relevant to the respective research questions asked in the study.
Algorithms are capable of advising human decision‐makers in an increasing number of management accounting tasks such as business forecasts. Due to expected potential of these (intelligent) algorithms, there are growing research efforts to explore ways how to boost algorithmic advice usage in forecasting tasks. However, algorithmic advice can also be erroneous. Yet, the risk of using relatively bad advice is largely ignored in this research stream. Therefore, we conduct two online experiments to examine this risk of using relatively bad advice in a forecasting task. In Experiment 1, we examine the influence of performance feedback (revealing previous relative advice quality) and source of advice on advice usage in business forecasts. The results indicate that the provision of performance feedback increases subsequent advice usage but also the usage of subsequent relatively bad advice. In Experiment 2, we investigate whether advice representation, that is, displaying forecast intervals instead of a point estimate, helps to calibrate advice usage towards relative advice quality. The results suggest that advice representation might be a potential countermeasure to the usage of relatively bad advice. However, the effect of this antidote weakens when forecast intervals become less informative.
There is empirical evidence that decision makers show negative behaviours towards algorithmic advice compared to human advice, termed as algorithm aversion. Taking a trust theoretical perspective, this study broadens the quite monolithic view on behaviour to its cognitive antecedent: cognitive trust, i.e. trusting beliefs and trusting intentions. We examine initial trust (cognitive trust and behaviour) as well as its development after performance feedback by conducting an online experiment that asked participants to forecast the expected demand for a product. Advice accuracy was manipulated by ± 5 % relative to the participant’s initial forecasting accuracy determined in a pre-test. Results show that initial behaviour towards algorithmic advice is not influenced by cognitive trust. Furthermore, the decision maker’s initial forecasting accuracy indicates a threshold between near-perfect and bad advice. When advice accuracy is at this threshold, we observe behavioural algorithm appreciation, particularly due to higher trusting integrity beliefs in algorithmic advice.