Notice: the WebSM website has not been updated since the beginning of 2018.

Web Survey Bibliography

Title Human vs. artificial intelligence: Are software solutions already able to replace human beings?
Author Koch, M.
Year 2016
Access date 29.04.2016
Full text PDF (537MB)
Relevance & Research Question: Human-based coding of open-ended answers is a time-consuming and tedious procedure – especially in terms of large sample sizes. Therefore, in recent years, several software solutions have been developed in order to enable an automatic process of coding. The objective of this study is to investigate if these software solutions can already replace human-based coding or if they should only be used as supportive tools.

Methods & Data: In the first step, verbatims were coded by two different people in order to calculate a basis value for inter-rater reliability: therefore, the coders independently assigned the answers to 14 predefined categories. In the next step, three software solutions (one freeware and two commercial software) were used for coding the open-ended answers: All three software programs were applied to cluster the open-ended answers based on the semantics (no predefined categories), while one of the two commercial software was also able to automatically allocate the answers to the 14 predefined categories. For analysis, the different types of coding were compared with each other (human – human, human-machine and machine-machine).

Results: Human-human: The human-based coding achieved the highest matching rate (M=86%; average Cohen’s kappa к=0,8). Human-machine: The clustering of the open-ended answers by their semantics delivered different thematic codes. However, one code was evident with all three software solutions and the human-based coding. Yet, the level of average agreement was rather low (M= 25%; Kappa к=0,3). In contrast to that, the machine-based allocation to the predefined categories performed better: the average inter-rater reliability was к=0,4 with 38% percent of agreement. Machine-machine: In terms of the code that was evident with all three software, an average agreement of 71% could be achieved between the three different programs (average Kappa к=0,6).

Added Value: In consideration of the results, it seems that human-based coding is much more precise than machine-based coding. Yet, software solutions can facilitate the tedious procedure of coding: They can be applied before coding in order to get an overview of important keywords and thematic aspects within seconds. Besides, the categorization via predefined categories is also promising.
Year of publication2016
Bibliographic typeConferences, workshops, tutorials, presentations

Web survey bibliography - Germany (639)